Wikipedia:AI noticeboard

 Cleanup has been requested.

Yet another enormous swath of AI edits going back to 2023. What makes this especially telling is that not only do their edits seem like AI, the way in which they seem like AI tracks the differences in AI model output exactly. Special:Diff/1203341299 reads like 2024 AI, Special:Diff/1288603486 reads like 2025 AI, Special:Diff/1347250351 reads like 2026 AI, etc. This also applies to their edit summaries and talk page comments, e.g. Talk:Picayune Rancheria of Chukchansi Indians, Talk:Culture of California, etc.

I warned them in February; they did not answer, despite being actively editing. Gnomingstuff (talk) 20:14, 20 June 2026 (UTC)

they responded Kowal2701 (talk, contribs) 22:36, 16 July 2026 (UTC)

from Nov 2024 had utm_source=chatgpt. Today, the site where the Sentinel Hotel once stood serves as a memory of the rich history of Yosemite Valley's development into a major tourist destination. Its enduring legacy continues to captivate those interested in the evolution of Yosemite's hospitality industry. in Sentinel Hotel (Yosemite Valley) (July 2023). I think I'm gonna draftify from Discovery Tree (Jan 2023) as it has lots of WP:AILEGACY and promo which differs from earlier creations and writing. Kowal2701 (talk, contribs) 17:34, 9 August 2026 (UTC)

I see you are moving many articles I contributed to back into Draft and out of the mainspace. I spent many hours editing and researching, especially around areas that were not well covered, including the history of the Sierra Nevada and Yosemite National Park, and these articles are well cited, and believed to have met quality guidelines for wikipedia. If tone is the issue, then could these pages be cited for tone? Guywelch2000 (talk) 17:49, 9 August 2026 (UTC)

Hi @Guywelch2000, I was about to leave a message on your talk. Normally these articles would get deleted (see WP:LLMPROD), but I can see you've been acting in good faith so I draftified them instead for you to work on if you wish (see here for a list of your drafts). The biggest issue with LLM use is hallucinations (i.e. WP:V failures), though there are other important issues like WP:NPOV/WP:WTW and WP:COPYVIO. The signs that appear as promotional writing are only how we detect LLM-generated content, copyediting is not how you fix such content (see WP:AISIGNS). Usually we recommend people to stubbify or rewrite LLM-generated articles from scratch since it's often very time-consuming to attempt to preserve such content, but it's up to you. You may want to go through WP:AFC for the first few, idk. Please let us know if you have any questions.

Unfortunately your edits to existing articles during the timeframe of LLM use will also need cleanup, can I ask if my estimate of late Jan 2023 as a start date was about right? I'll create a tracking page for this in a bit Kowal2701 (talk, contribs) 17:59, 9 August 2026 (UTC)

Thanks for the guidance and suggestions here. I have been a part of Wikipedia for about 20 years and began contributing much more often during the COVID timeline. I enjoy doing research and found that for some topics, the Chapt GPT Deep Research tool was very helpful to surface citations that were otherwise hard to locate. For example my article on Mono Winds was something that computer aided tools helped me find citations from the US Forest Service and others that I would have struggled to locate through traditional search engines. I understand what you're now saying about tone, especially neutral point of view. I did take that into account, however I have done a lot of promotional copy-writing in the past, so what sounds neutral to my ears needs to be stepped back a few more notches. Guywelch2000 (talk) 18:13, 9 August 2026 (UTC)

Can you give us an approximate start date for using LLMs to generate text? That would be very helpful in directing the cleanup. Kowal2701 suggested January 2023, is that correct? M kuhner (talk) 01:20, 10 August 2026 (UTC)

@Guywelch2000 can you give us an update? Dr vulpes (Talk) 04:29, 16 August 2026 (UTC)

 Cleanup has been requested and is being tracked at /2026-08-09 Yahya0071, 19 pages need review.

 Yahya0071 (talk · contribs · logs · block log) (9 August 2026)
Editor has been blocked from editing the mainspace (articles only) for AI/LLM violations. 

Yahya0071 (talk · contribs · deleted contribs · logs · filter log · block user · block log)

See the recent articles created in rapid succession with AI signs and User talk:Yahya0071#Question. Jfire (talk) 02:24, 23 June 2026 (UTC)

Hi Jfire, thank you for reaching out and pointing me toward the policy guidelines. You are completely correct. I have been translating and porting established articles from the Arabic Wikipedia over to the English side to help expand coverage on these topics. To speed up the translation and Wikitext formatting process, I have been using an AI assistant.

I realize now that creating these pages in rapid succession without proper translation attribution or thorough manual sifting is a problem. I sincerely apologize for the oversight. I am going to pause creating new articles immediately, and I will go back through the articles. I appreciate your vigilance in keeping the project clean! Yahya0071 (talk) 02:29, 23 June 2026 (UTC)

Thank you for your honesty, your understanding of the guidelines regarding LLM use, and pledging to pause. I appreciate that. LLM-aided translation is allowed, but only as outlined in WP:LLMTRANSLATE. In particular only when you have checked the sources in the origin language article and you are sure the translated text reflects them fairly. Given the rate that you were creating new articles and the fact that many of the sources refer to offline texts, it seems unlikely that the articles you created comply with that guideline. Can you confirm? Jfire (talk) 02:35, 23 June 2026 (UTC)

Bumping thread. Kowal2701 (talk, contribs) 05:57, 22 July 2026 (UTC)

They created United States–Saudi strikes on Iraqi militias (29 July), which appears to be a translation of ar:الضربات الأمريكية السعودية على الميليشيات العراقية 2026 (which Yahya has written most of ). Ref #4 doesn't say "30 drones", only saying "several", and it doesn't give the background (which is pretty WP:AITREND) nor even mention Israel. Ref #3 doesn't even mention the Houthis let alone the other info about them (there's also WP:AIDASH). @Admins willing to patrol AINB: I think we're out of rope, especially given they said they'd go back through the articles, but instead created more pages and went inactive for 3 weeks before pumping out this. Kowal2701 (talk, contribs) 15:35, 31 July 2026 (UTC)

Mecca Joint Defence Agreement (8 August) was created with WP:AIREDCAT among other signs, @Admins willing to patrol AINB: I think a mainspace block is needed Kowal2701 (talk, contribs) 18:05, 9 August 2026 (UTC)

They also made an edit with blatant markdown (around Director of Finance) mere days after making the comment above ! Kowal2701 (talk, contribs) 18:07, 9 August 2026 (UTC)

pblocked from mainspace per above. Also scoring 100% on Pangram: https://www.pangram.com/history/51c12e52-c427-4218-880e-3d3a89f70fd1?ucc=WdBFgECBm3a ARandomName123 (talk)^{Ping me!} 18:29, 9 August 2026 (UTC)

Saw that @Kowal2701 has alerted WP:ITN/C § (Posted) Mecca Joint Defence Agreement about it. ‑‑gurkubondinn 17:04, 11 August 2026 (UTC)

While third-party AI detectors like Pangram are notoriously unreliable and prone to false positives, I acknowledge the formatting glitches and citation cleanup needed on my recent translations. I accept the mainspace restriction and will be more precise on future writing. Yahya0071 (talk) 19:18, 9 August 2026 (UTC)

Created tracker so we can see what the issues are, but probably want to wait and see if there is a block appeal. I will write to the editor. M kuhner (talk) 23:35, 9 August 2026 (UTC)

(btw it might be easier to create tracking pages after LLMPRODs go through, because their edits to those articles get deleted as well and won't show up) Kowal2701 (talk, contribs) 13:18, 11 August 2026 (UTC)

The only argument for doing it earlier is so I can keep an eye on the LLMPRODs--it's very easy to see the link turn red, or not, and to inquire why not if it doesn't. M kuhner (talk) 14:38, 11 August 2026 (UTC)

true, but for ones w a large number of creations (like 100+) that's rough, ofc up to you though Kowal2701 (talk, contribs) 14:42, 11 August 2026 (UTC)

Cleanup efforts are ongoing and being tracked at /2026-07-23 Vayro.7, 0 pages need review.

Vayro.7 (talk · contribs · deleted contribs · logs · filter log · block user · block log)

yet another enormous, ongoing swath of most likely AI edits; the ones that tipped me off were these (note markdown etc), but there are a lot more and I do not have time to dig through all of this Gnomingstuff (talk) 20:13, 24 June 2026 (UTC)

they were previously warned November 2025 (after the markdown diff) , to which they replied

Hi Jlwooda, Thank you for your contributions, you are not mistaking at all, actually I have used AI chatbot to assist me in some parts of the article, I did not knew it is unsuitable to use such a thing. I will make sure to check each citation and remove fictitious references if there are. Thank you for your help! Vayro.7 (talk) 07:09, 6 November 2025 (UTC)

and

Hi Jlwoodwa,
Thank you for your helpful feedback and for mentioning the sandbox — I actually wasn’t familiar with it before, and I appreciate you pointing it out. I’ve carefully reviewed and updated the Biological intelligence article, verifying all references and improving the structure to align with Wikipedia’s standards of neutrality and verifiability.
I’ll make sure to use my sandbox for future drafts and experimental edits before publishing. Thanks again for your guidance and patience! Vayro.7 (talk) 08:11, 6 November 2025 (UTC)

What a bizarre editing style, the recent edit summaries appear AI-generated at least Kowal2701 (talk, contribs) 23:37, 24 June 2026 (UTC)

yup, reverting the obvious ones and i'm seeing ~-10K bytes, which is absurd ~2026-35622-93 (talk) 00:51, 25 June 2026 (UTC)

@~2026-35622-93, btw your edits got reverted, you may want to link to the report when doing cleanup Kowal2701 (talk, contribs) 22:42, 25 June 2026 (UTC)

@Kowal2701 it was solved! see https://en.wikipedia.org/w/index.php?title=Procrastination&diff=prev&oldid=1361005691 and https://en.wikipedia.org/w/index.php?title=Self-control&diff=prev&oldid=1361005645 ~2026-35622-93 (talk) 22:54, 25 June 2026 (UTC)

What a wholesome interaction! Gladdens my heart. M kuhner (talk) 03:26, 26 June 2026 (UTC)

can someone please spotcheck their latest contribs for LLM use? ~2026-38678-64 (talk) 01:06, 23 July 2026 (UTC)

Checked their articlespace edits from July 16th till now, seems all good to me - modest additions supported by the added sources. I'm unclear on how much cleanup has been done; since they admitted to previously using AI shall we make a list? ^purring_maggot (talk) 02:10, 23 July 2026 (UTC)

i think it'd be interesting for backlog reasons (also i don't believe i can create the pages myself because TAs can't create subpages, but i can get refunded or have a userpage) ~2026-38678-64 (talk) 03:01, 23 July 2026 (UTC)

Tracker created. (-64, you are a strange creature, neither fish nor fowl.) M kuhner (talk) 05:03, 23 July 2026 (UTC)

i think it's a silly joke to spoof the admin hopeful template into autoconfirmed hopeful, arguably the best joke i have made on wikipedia ~2026-38678-64 (talk) 12:42, 23 July 2026 (UTC)

haha, that's brilliant! (TM:User Wikipedia/Administrator someday is the original template for anyone unaware) Kowal2701 (talk, contribs) 13:16, 23 July 2026 (UTC)

I looked at the user's edits to Biological intelligence, which they represented to Jlwoodwa that they had fixed. In one edit summary they claim to have corrected bad references but though they say they validated Friston 2019, it's a dead link. In another edit summary they say they added a DOI, but it goes to a totally different article. I conclude from this that, like the sources themselves, their claims to have improved the article are hallucinations. (IMO they are too stupid to be lies: why go to the trouble of pasting in a new link if it's going to 404?) Given that, we can't assume good faith on subsequent edits: I think some action is needed promptly. M kuhner (talk) 02:17, 25 June 2026 (UTC)

Thank you for taking the time to review my edits and for raising these concerns. I understand why some of my recent contributions have prompted scrutiny. My intention has always been to improve articles, not to introduce inaccurate information. Looking back, it's clear that some edits did not meet Wikipedia's standards, particularly with regard to source verification. I'll make an effort to revisit previous edits where feasible and fix any issues that are brought to my attention. Going forward, I'll make sure that substantial additions are based directly on reliable sources, that every citation is checked before publication, and I'll use my sandbox for larger drafts where appropriate. I appreciate everyone who took the time to identify problematic edits. Vayro.7 (talk) 10:15, 9 July 2026 (UTC)

@Vayro.7, see WP:AITALK Kowal2701 (talk, contribs) 14:19, 6 August 2026 (UTC)

• They're still editing, can someone check their recent edits for WP:V? We might have to add more to the tracking page Kowal2701 (talk, contribs) 22:25, 14 August 2026 (UTC)

  I think the user's most recent edits are just technically passing WP:V, but they are using ChatGPT to find sources and unsurprisingly almost everything it comes up with is a primary source (just look at any of their edits to Travel visa). The content added is just a single sentence here and there (also, does anyone else find it annoying when someone adds content and then adds citations in another edit? makes it hard to follow) mostly adding sources in place of existing citation needed tags. This is the biggest stretch with a single government website standing in to prove that "Many countries frequently...(have this requirement)". I'm not sure we need to be on AI-watch for this user anymore but they could use a primer on primary vs. secondary sourcing. ^purring_maggot (talk) 18:04, 16 August 2026 (UTC)

 Cleanup has been requested.

LumenArchivorum (talk · contribs · deleted contribs · logs · filter log · block user · block log)

More high-volume AI edits. Specifically, they are using Perplexity, and I know this because Special:Diff/1345574273 left in a citation to a Perplexity S3 bucket ("ppl-ai-file-upload.s3.amazonaws.com"). And also the usual AI signs everywhere (from that one diff, the edit summary, the it functions not only as a place of worship but also as a major religious, social, and cultural center, enhancing ventilation, etc) Gnomingstuff (talk) 18:15, 25 June 2026 (UTC)

Hello @Gnomingstuff and fellow Wikipedians

It has been a bit over a week, and I have now reviewed the correctly reported diff and the article, I then removed the AI content, and I then rewrote what needed fixing in my own words from sources I have read. I have also looked at my edit history wherever AI tools were involved outside of Wikipedia, to make sure no similar problems remain.

From now on, I will not paste AI‑generated things into maonspace or rely on AI tools for facts or summaries. When I use AI at all, it will be restricted strictly to what I do outside Wikipedia. I’m very grateful to @Gnomingstuff that lifted this problem and to all editors that guided me through the cleanup.

I deeply apologise for I entered AI‑generated text and the Perplexity URL got into mainspace. That was a bad mistake I have to own, and I’m so sorry for the extra work and concern it has rightfully caused.

https://en.wikipedia.org/wiki/User:LumenArchivorum LumenArchivorum (talk) 09:21, 5 July 2026 (UTC)

Block from editing. In my experience, they were being disruptive to consensus-making at Talk:Yemen. Their huge paragraphs of text, even when it is not necessary, takes up editor time. They can be unblocked if they show that they will not use AI HerrHeulsuse (talk) 06:25, 8 August 2026 (UTC)

I'm sorry to hear I came of as disruptive, can you please give me a concrete example for me to consider so I can learn from it? Adding the excerpt from the health of Yemen into the Yemen article's health section was meant to be helpful. LumenArchivorum (talk) 07:57, 8 August 2026 (UTC)

Hi Gnomingstuff,

Thanks for taking the time to leave me a note and for flagging the diff at the noticeboard. I’ve gone back to look at the edit you pointed out (1345574273) and at the current guidance about large language models.

You’re right that there’s a problem with that edit. My intention was to copy and paste source information that I had been working with, but in the proccess I mistakenly copied over a Perplexity‑generated summary and its file URL instead of the underlying source details. That meant AI‑derived prose and an AI artifact ended up in mainspace, which isn’t okay under the pressent rules.

I’m starting to fixing this by removing the Perplexity citation and either rewriting or cutting the sentences that came from AI output, making sure that what remains is based on solid sources I’ve actually read and written in my own words. I’m also going through my edits in general to review how I ells my have worked in anyway but in accordance with Wikipedia rules and guidelines.

From here on I’ll be much stricter with myself, if I use tools like that, it’ll only be outside Wikipedia for brainstorming and searching for academic sources.

I appreciate you raising this rather than ignoring it. If you come across other specific diffs where you think AI‑derived text or citations are still lurking, please do point me to them and I’ll sort them out.

LumenArchivorum (talk) 18:54, 25 June 2026 (UTC)

Hi all,

I’d like to report to my fellow Wikipedians about my edits, and thank Gnomingstuff for taking the time to look at the diff and raise the concerns here.

After reading through the current guidance on large language models, I can see that what I did in the reported edit (1345574273) isn’t in line with the rules. My intention at the time was to copy over source information I’d been working with, but I mistakenly copied a Perplexity‑generated summary and its file URL instead of the actual underlying sources. The result was that AI‑generated prose and an AI artifact were added to the article, which clearly shouldn’t have happened.

I’m now cleaning that up, the AI‑related citation is being removed, and I’m rewriting or removing the sentences that may have came from AI output so that the page only contains text I’ve written myself from reliable sources that I’ve checked. I’m also revewing my recent edits to see where similar problems may exist and correcting them.

Going forward, I’ll treat AI tools strictly as off‑wiki helpers and for ideas searching the web for academic sources, and I definitely won’t paste their output directly into mainspace. My goal is to keep my contributions clearly within the current policies on AI use and sourcing.

If anyone spots other specific diffs that need attention, I’m happy to look at them and fix them. I appreciate the work that’s gone into clarifying these standards; I’ll make sure they’re reflected in how I edit and in how I talk about editing with others. LumenArchivorum (talk) 19:07, 25 June 2026 (UTC)

Talk page comments should not be written with AI either. Gnomingstuff (talk) 01:14, 26 June 2026 (UTC)

I will drop the wording suggester feature and the obsessiveness in correctness in talkpagespaces because it does come of as rather robotic. I'm so sorry I allowed AI output to enter mainpace with these edits you correctly noted, I am dedicated to ensure I perform a complete cleanup. LumenArchivorum (talk) 00:29, 27 June 2026 (UTC)

Hi @LumenArchivorum, thank you for doing this, but can I ask that you do the same for any other articles you've add LLM-generated text to? Kowal2701 (talk, contribs) 17:42, 26 June 2026 (UTC)

Hi

@Kowal2701, I am dedicated to carefully review all my edits where any AI tool was part of the research to ensure ethical and responsible use, (maximum one more week, I won't keep editing new articles until I feel confident it is done) and thank you for guiding me through the clean up. Using AI to help me find sources has been a game changer in my ability to research, gives a bigger selection of content to investigate, which has to take its good time, slow and steady wins the race. Don't stress and edit, triple check all sources and text you create yourselves carefully to be sure no AI generated content is copied over from any pre research. Basic editorial integrity is at stake.

In today's world I got to better and more humbly learn when to and how to use AI outside of Wikipedia, definitely not allow AI output to enter Wiki mainspace, and I need to work slower and more carefully not to do this again. I'm so sorry about this, I am dedicated to go through all my edits where I have used AI in anyway outside of Wikipedia in research to make sure no more violations exist. LumenArchivorum (talk) 00:20, 27 June 2026 (UTC)

What do you mean by "no AI generated content is copied over from any pre research"? Were you using AI tools to gather a list of potential sources on a topic? SenshiSun (talk) 01:09, 27 June 2026 (UTC)

I should be clearer. I mean that in any case where AI tools are used outside of Wikipedia, as in helping to gather sources, nothing AI create can be allowed to enter Wiki mainspace, I am deeply regretful for this miss of editorial due diligence. LumenArchivorum (talk) 13:19, 27 June 2026 (UTC)

Please be very careful. LLMs can NOT be trusted to summarize a source. In particular, if it "thinks" you are looking for details about something, but no details are present, it will invent some. They are useful to suggest sources, but no more than that. M kuhner (talk) 03:47, 28 June 2026 (UTC)

Yes. Indeed, it is so important to keep reminding oneself about that, don't trust AI conclusions on anything. LumenArchivorum (talk) 19:52, 29 June 2026 (UTC)

They appear to be still using AI... Gnomingstuff (talk) 18:33, 16 July 2026 (UTC)

Yeah- their comments on Talk:Health in Yemen look human-written, their edit summaries on that article don't. Sarsenet•he/they•(talk) 22:44, 16 July 2026 (UTC)

Was going to open a NEW section for a review of Health in Yemen becaus of what I clearly believe are mass AI additions. This thread has simply reinforced my view.

AI-generated lead

Modern health care in Yemen began to take shape in the early 20th century, initially through facilities established by colonial administrations and missionary groups. Despite these developments, most people continued to rely on traditional medicin. After Yemen joined the World Health Organization (WHO) in 1953, public health efforts became more coordinated. Health services were gradually improved with increased involvement of international health personnel and the expansion of basic medical education. During the second half of the 20th century, health services expanded further, but unevenly across regions.

What is the normal process? selective reverts? Mass rollback to before editor in question involvement ?

Not sure if I am looking for others to edit the page for clear problems ?....or deal with it on my own

...... I guess I am looking for analysis of article from fresh eyes! Moxy🍁 04:27, 8 August 2026 (UTC)

Hello @Moxy

I understand that AI is a big worry these days but this is not an AI generated text. LumenArchivorum (talk) 06:24, 8 August 2026 (UTC)

FWIW Pangram says it's 100% human-written, and the pace seems more choppy than what AI does Kowal2701 (talk, contribs) 12:37, 8 August 2026 (UTC)

odd i get 100% of text is likely AI. Moxy🍁 13:31, 8 August 2026 (UTC)

yeah I'm getting AI hits all over.... I think I'm going to remove all from country articles and let other people deal with the health ones.... I simply don't have the time. Moxy🍁 23:53, 10 August 2026 (UTC)

Pangram does give high false negatives so if you're getting positives (with the wikitext markup removed) that's pretty strong Kowal2701 (talk, contribs) 00:03, 11 August 2026 (UTC)

With Pangram i am getting 73% AI generated text on Health in Yemen Moxy🍁 00:09, 11 August 2026 (UTC)

When I test it I get - it says impossible to determine if written by a Human or AI.

These tools are not foolproof. Compounded by the fact that MEDRS does not leave much room for a personalised way of writing, it is mostly very standard terms written in a standard order. LumenArchivorum (talk) 07:21, 11 August 2026 (UTC)

I am afraid you are mistaken. Further, it requires more than that the style seem like AI to conclude that it is. Wiki rules are clear here. LumenArchivorum (talk) 07:16, 11 August 2026 (UTC)

Please wait for a group consensus before removing my edits. You have already started. LumenArchivorum (talk) 08:08, 11 August 2026 (UTC)

Wikipedia:Do not lie about AI Moxy🍁 11:21, 11 August 2026 (UTC)

That is an incorrect characterisation. Show us all here the evidence. LumenArchivorum (talk) 12:30, 11 August 2026 (UTC)

WP:NOPROOFOFAI

https://quillbot.com/ai-content-detector

Modern health care in Yemen began to take shape in the early 20th century, initially through facilities established by colonial administrations and missionary groups. Despite these developments, most people continued to rely on traditional medicin. After Yemen joined the World Health Organization (WHO) in 1953, public health efforts became more coordinated. Health services were gradually improved with increased involvement of international health personnel and the expansion of basic medical education. During the second half of the 20th century, health services expanded further, but unevenly across regions.

100% of text is likely AI

Thinking best just block. Moxy🍁 23:11, 11 August 2026 (UTC)

@Kowal2701 & @M kuhner have expressed different takes than you on this particular text and so far it is not a big group discussion in this conversation.

Furher, Quilbot is reasonably useful as a quick screening tool for indicators, but not reliable to establish authorship or support high-stakes decisions. Wiki rules do not support using this type of style analysis to base decisions on which can be easily reviewed here: Wikipedia:Writing articles with large language models - Wikipedia

I have not had a problem admitting to and resolving previous problems raised regardless of type, I rather embrace the opportunities to grow, but the following text is absolutely written by myself and not AI:

Modern health care in Yemen began to take shape in the early 20th century, initially through facilities established by colonial administrations and missionary groups. Despite these developments, most people continued to rely on traditional medicin. After Yemen joined the World Health Organization (WHO) in 1953, public health efforts became more coordinated. Health services were gradually improved with increased involvement of international health personnel and the expansion of basic medical education. During the second half of the 20th century, health services expanded further, but unevenly across regions. LumenArchivorum (talk) 00:06, 12 August 2026 (UTC)

@Moxy, generally-speaking, for cases where you're confident someone's using AI but can't quite prove it, it's frustrating but easier to keep an eye on them and wait for more evidence. I go through AINB's archives as part of #Backlog bonanza, so everything gets done eventually Kowal2701 (talk, contribs) 12:48, 12 August 2026 (UTC)

That's fair enough. It is not fair to single handedly take upon yourself what the truth is and act on it in these cases without sufficient evidence. Process and evidence matter.

@Moxy Now please stop removing everything I do without bothering to look into things further, because fair is fair. LumenArchivorum (talk) 13:56, 12 August 2026 (UTC)

I've given them some advice on Talk:Health in Yemen § August 2026, but I've otherwise not been involved here (apart from restoring an {{ai-generated}} tag). ‑‑gurkubondinn 14:21, 13 August 2026 (UTC)

I appreciate it a lot. In fact I appreciate all editors in this conversation who are just trying to get this encyclopedia right, and since this conversation started with an accurate observation made by Gnomingstuff, which naturally raises everyone's antennas, I may cause unneeded concern when I keep editing while this conversation is still open.

I should make it easy for diligent editors like Moxy by demonstrating a text analysis to save the work which is due while an ongoing conversation in the AI noticeboard indicates. I have more than 1700 edits so I should give fair time for editors to have a chance to review if my other edits contain any AI.

I will take a pause for a month (from adding any content, I can focus on reviewing possible vandalism via recent changes and other ways to contribute where LLM is not a consideration.) lets you guys look through things in your own time, but I will remain responsive for any questions during this period. If you need more time later on let me know, and here is an analysis of the health of Yemen lead text which I wrote and is inaccurately pointed to as generated by AI.

LumenArchivorum (talk) 13:48, 15 August 2026 (UTC)

Hello

I appreciate that a AI noticeboard conversations informs you to analyse the concerned editor's edits for AI more than with other editors currently being discussed in the AI noticeboard, so it is not right for me to be anything but be helpful when concerned is raised. Here is an analysis by the program most respected by Wikipedia, keeping in mind no program is acknowledged to be a deciding factor on writing style alone.

LumenArchivorum (talk) 14:04, 15 August 2026 (UTC)

WP:AITREND Moxy🍁 21:50, 15 August 2026 (UTC)

I ran the same paragraph as in the screenshot through pangram, and I got back "100% AI-generated". ‑‑gurkubondinn 22:04, 15 August 2026 (UTC)

Trying to deceive the community is another big concern here. Moxy🍁 22:13, 15 August 2026 (UTC)

I have admitted to, apologised for, and cleaned up what Gnominstuff correctly pointed to as AI. Your example from the health of Yemen is not made by AI, I wrote it. I have no problem taking accountability and cleaning up when needed, it does not mean that everything I ever done on WIkipedia is AI. LumenArchivorum (talk) 05:49, 16 August 2026 (UTC)

I'ts very strange that you keep getting positive results when I and 2 other editors here in this conversation have found the opposite results. LumenArchivorum (talk) 05:46, 16 August 2026 (UTC)

 Cleanup has been requested.

Rejoy2003 (talk · contribs · deleted contribs · logs · filter log · block user · block log)

(Blocked for unrelated reasons, but I asked about AI prior to that; the reason I bring this up now is stumbling across yet another article by them independently.)

More AI edits going back to 2023. I asked about this and their response was "don't remember other than copy editing in the past," which I suppose makes sense, as these are older edits.

However, I believe it goes beyond just copyediting. This is somewhat annoying to check since their editing pattern is to make a huge amount of tiny edits to an initial draft, but looking at initial versions of articles, they not only show AI signs but show ones that are consistent with the years they were created: Special:PermanentLink/1172775823 is the 2023 AI output pattern, Special:PermanentLink/1226227624 has the 2024 AI output pattern, etc. (Threw one into Pangram fwiw and it also came back as AI generated) Gnomingstuff (talk) 14:30, 29 June 2026 (UTC)

Unblocked now, it was just a 24 hour one. @Rejoy2003, can you please respond to the above? I’m unsure if you’re aware, but see WP:NOLLM Kowal2701 (talk, contribs) 18:06, 29 June 2026 (UTC)

Still no reply nor edits. ~200 creations since Nov 2022 Kowal2701 (talk, contribs) 06:18, 22 July 2026 (UTC)

Hi, sorry for the late reply. I am quite busy these days. Regarding your question, I was unaware of the implications of AI in the past. I have already given my clarification to the OP that my use of AI in "some" articles have strictly been regarding copy editing which was basically grammar fixes. There was no WP:NOLLM in the past, something which has been introduced quite recently. I plan on fixing the AI issues (if there are honestly as we know how reliable those tools are) myself and dragging this to the noticeboard seems a bit too much on my end. Rejoy²⁰⁰³(talk) 13:17, 31 July 2026 (UTC)

@Rejoy2003, WP:NOLLM applies to all content regardless of when it was written, it just means behaviour pre-March 2026 isn't sanctionable under it. LLM-generated content typically has a whole host of problems, primarily hallucinations (i.e. WP:V failures), but also others like WP:NPOV/WP:WTW and WP:COPYVIO. I'm going to start draftifying your creations from August 2023 onwards later today for you to work on if you wish (normally we'd delete such articles under WP:LLMPROD), you can see a list of your drafts here. Usually we recommend that people stubbify or rewrite such articles from scratch as attempting to preserve such content is often very time-consuming, but it's up to you. Best, Kowal2701 (talk, contribs) 18:23, 9 August 2026 (UTC)

Thank you for your reply. However as per WP:NOLLM, Editors are permitted to use LLMs to suggest corrections to their own writing, and to incorporate them after human review. This is limited to spelling, punctuation, capitalisation, grammar, and other simple mistakes.

I have already stated that articles that of from around 2023 have been copyedited only using AI. I have not generated articles from scratch nor rewritten using it. From what I see, I don't think I have violated anything here. And also I would like to add that I have not used AI for all the articles created post 2023. So definitely it needs further scrutiny. I am confident that the content used in my authored articles follow WP:V and I have done my best to confirm the same after manually reviewing. Rejoy²⁰⁰³(talk) 08:33, 10 August 2026 (UTC)

Frankly there is no way that whole articles appear LLM-generated if the LLM only did copyediting. LLMs are notoriously bad at sticking to a defined scope, so it's possible that you asked it to copyedit and it largely rewrote your work instead. Please let me know which articles you didn't use AI for and I'll have a look. If you're near 100% sure that an article complies with WP:PAGs then feel free to move back to mainspace, but please don't underestimate how problematic such content tends to be (you can see other cases here where people did this) Kowal2701 (talk, contribs) 12:44, 10 August 2026 (UTC)

 Soldatipro (talk · contribs · logs · block log) (3 July 2026)
Editor has been warned about the use of AI/LLM content and WP:NOLLM. 

Edits by the user have been reverted (1, 2, 3, 4, 5, 6, 7) by Gurkubondinn but without actual accountability as to whether they used AI or not, they might just keep using it. More so, their drafts have been declined for being generated by LLMs (Draft:Pando S.A. and Draft:Dos en Uno) PeepeeDino (talk) 07:55, 3 July 2026 (UTC)

They have definitely not stopped, I just tagged Wings Store for G15: AbuseLog/44748026. ‑‑gurkubondinn 16:24, 28 July 2026 (UTC)

I've reminded them of NOLLM and asked them in User talk:Soldatipro § July 2026 to stop using LLMs, and gone through and declined their outstanding drafts. ‑‑gurkubondinn 16:42, 28 July 2026 (UTC)

It is evident that the final spell-checks have made unintended changes to my articles. I will improve my writing process. Thank you very much for the time you have dedicated to this. Soldatipro (talk) 18:08, 28 July 2026 (UTC)

What AI tools are you using? You should be aware that AI generated text is prohibited on Wikipedia. That includes "corrections" done by tools such as Grammarly. ~Anachronist (who / me) (talk) 23:27, 28 July 2026 (UTC)

(btw corrections by LLMs are permitted as one of the carveouts, but if we can tell, then they're doing more than correcting) Kowal2701 (talk, contribs) 23:30, 28 July 2026 (UTC)

There was no carve-out for grammar until 10 days ago. And Grammarly is specifically called out as a problem child in a footnote. Grammarly is known to do substantive rewriting, far more than merely correcting subject-verb relationships and such. ~Anachronist (who / me) (talk) 23:48, 28 July 2026 (UTC)

Clearly, the AI ​​systems went overboard with my spell-check request and did whatever they wanted (we're getting closer and closer to Skynet, haha). I think I'll go back to good old reliable Word!

@Gurkubondinn @Anachronist @PeepeeDino I apologize for all the inconvenience. Really. I will review each of the articles that were flagged. Soldatipro (talk) 17:44, 29 July 2026 (UTC)

This is not from a "spell-check request": Special:Diff/1355214243. Why do you not use the spell checker in your browser? ‑‑gurkubondinn 18:23, 29 July 2026 (UTC)

Yes, I clearly need to use a different tool that doesn't allow for so much leeway, so I can have more control over the content. I should have paid closer attention to the edits I made. In any case, I’ve already been working on fixing some of the proposed articles. If you have the time, I’d really appreciate it if you could take a look. Thank you again! Soldatipro (talk) 20:59, 29 July 2026 (UTC)

https://en.wikipedia.org/wiki/Draft:Pando_S.A.

https://en.wikipedia.org/wiki/Draft:Operation7

They are still displaying "Your reviews should be put on hold until your ANI investigation is done," so once the investigation concludes, I will submit them to the review process again.

Thank you all Soldatipro (talk) 21:02, 29 July 2026 (UTC)

That is supposed to say AIN, @Sulfurboy made a typo. ‑‑gurkubondinn 21:19, 29 July 2026 (UTC)

Yes, I read about the issue with your Red Bull, haha. Anyway, I’ll be keeping an eye out for the results of the investigation and will help out however I can. Thanks as always! Soldatipro (talk) 21:23, 29 July 2026 (UTC)

This is 'the investigation' (personally I usually say 'filing' instead of 'investigation'), and the 'results' more or less just depend on you at this point. We've established a likely cause/problem (you using AI chatbots for spell checking), now we're waiting to hear what your plans are for preventing this problem in the future. Our goal is always to retain you as a productive and contributing editor, never to punish you or to drive you away. ‑‑gurkubondinn 21:43, 29 July 2026 (UTC)

As I should have as well, I didn't mean to make it sound as dramatic as I did Soldatipro, we just want to make sure we have a solid gameplan going forward and then we're happy to review your submissions again. Sulfurboy (talk) 21:50, 29 July 2026 (UTC)

I understand perfectly; my clumsy use of tools and lack of proper oversight clearly caused a lot of issues. My plan is simple: I’m going back to more rudimentary, classic tools (though I certainly won't be using a typewriter), but for spell-checking, I’ll return to my trusty old Word. As for the content, I will forgo using AI technology for research or processing. In fact, I’ve been working on the two articles I mentioned earlier—primarily by manually translating from other languages ​​or consulting reliable sources.

I’m not sure if there is anything else I should do, beyond whatever advice you might offer. Thank you for all the valuable time you have dedicated to me @Gurkubondinn@Sulfurboy@Anachronist Soldatipro (talk) 21:52, 29 July 2026 (UTC)

Sorry to ask this, but did you write this reply in the discussion box here on Wikipedia or was it copied from somewhere? Was it polished by any AI tools? ‑‑gurkubondinn 22:54, 29 July 2026 (UTC)

I used Google Translate because sometimes there is a specific term I want to use and I don't want to make a mistake. Obviously, in a discussion about the "non-use" of AI, I wouldn't actually use AI... I hope it's not a problem to rely on a translator's help occasionally. Soldatipro (talk) 00:18, 30 July 2026 (UTC)

By the other hand, I also managed not to leave any stray Markdown behind this time. That its something! jaja Soldatipro (talk) 00:23, 30 July 2026 (UTC)

Good morning/afternoon:Well... I’ve been working on my drafts! I ve taken careful note of everything you mentioned, so and I’m making precise and manual improvements.If its not too much trouble, Id like to ask you to take a look at them whenever you can :) I dont want to request a new review until you give the ok. Thank you again! @Gurkubondinn@Sulfurboy

This are:

https://en.wikipedia.org/wiki/Draft:Pando_S.A.

https://en.wikipedia.org/wiki/Draft:Operation7

https://en.wikipedia.org/wiki/Draft:Fundaci%C3%B3n_El_Libro

https://en.wikipedia.org/wiki/Draft:Dos_en_Uno Soldatipro (talk) 15:45, 30 July 2026 (UTC)

Did you copy the whole comment from Google Translate, or did you use it more like an "advanced dictionary"? ‑‑gurkubondinn 16:12, 30 July 2026 (UTC)

No... it is not appropriate to copy entire texts, I know that.

In fatc, you ll be able to see in my edited and improved articles as well the effort I am making to adapt my edits. Of course, I still have to improve, but I trying to go in the right way :)

When you have chance take a look at them if it si not too much trouble. Thanks again! Soldatipro (talk) 17:37, 30 July 2026 (UTC)

You are absolutely sure that you didn't write that reply into Google Translate in your native language and then copy the machine translation into Wikipedia? There is nothing unusual about how you composed the reply, and you didn't copy it from anywhere else?

Another thing that I wanted to say is that Microsoft Word now comes with Copilot and AI-features enabled by default, so you need to be careful to disable them as much as you can. Word also isn't very good for working with wikicode or the inline references used on Wikipedia. If you aren't already using it, you should enable the spell checker in your browser. It is actually very good, and my suggestion would be to just use the wikieditor with the spell checker in your browser. If you prefer a WYSIWYG editor like MS Word, then you can try out using the WP:VISUAL editor on Wikipedia instead of the source editor. ‑‑gurkubondinn 18:04, 30 July 2026 (UTC)

My dear friend... believe me,, I do not need Copilot or any AI to respond during a work discussion about Wikipedia articles. I have my own ideas. For sure. In fact, I only used the web version of the translator—which is very basic—on that one isolated occasion here. As for me, Ill take your advice about using a spell checker! I really want to focus on contributing to the community and stop having to defend myself against past mistakes... Ill submit my revisions for validation whenever you're ready! thanks again! Soldatipro (talk) 18:20, 30 July 2026 (UTC)

Soldatipro: can you then explain why there are zero-width spaces in Diff/1366734954 (between "other languages" and "or consulting") and in Diff/1366705831 (between "the AI" and "systems went overboard")?

These are even the Unicode characters U+200B and not the HTML entities ​, so it is not possible for you to have typed them on your keyboard in the reply box. ‑‑gurkubondinn 11:35, 2 August 2026 (UTC)

Probably just some weird leftover from a site I used as a source... Its legal if may say. You’re getting a bit obsessed with me at this point lol. Soldatipro (talk) 12:22, 2 August 2026 (UTC)

How does a leftover from a site [you] used as a source end up mid-sentence in two discussion replies? This is the last time that I am going ask you about this before I will be asking admins to block you from article space, so please answer honestly. You can take a look at the WP:USED-AI? essay before answering, but I will not be asking again. ‑‑gurkubondinn 12:30, 2 August 2026 (UTC)

Just noting here that I have determined that this pattern with two zero-width space characters, placed after a normal breaking space, are most likely an artifact of text produced by Google Translate. I figured this out by finding other examples of identically placed zero-width space characters on Wikipedia and then asking the editors that added them how the text had been produced. ‑‑gurkubondinn 14:53, 13 August 2026 (UTC)

@Admins willing to patrol AINB: Kowal2701 (talk, contribs) 16:35, 28 July 2026 (UTC)

(sorry for the superfluous ping) Kowal2701 (talk, contribs) 18:28, 28 July 2026 (UTC)

 Cleanup has been requested and is being tracked at /2026-08-10_Thoppy234, 458 pages need review.

Yet another swath of AI edits. (Should not need too much explanation, the talkpage comments are especially blatant.) I've reverted, reviewed or flagged all of the prose edits, but they have also added dozens-to-hundreds of infoboxes. These are AI too (WP:AIPLACEHOLDER shows up constantly, and I mean also they're using AI for everything else so...) and thus need review. Gnomingstuff (talk) 06:00, 4 July 2026 (UTC)

Kind of non-actionable by an admin, except for cleanup, since that editor hasn't been active for nearly a year. ~Anachronist (who / me) (talk) 06:51, 4 July 2026 (UTC)

I'm not asking for admin action, if I was I'd go to ANI; this is just to document the stuff needing cleanup so it doesn't fall through the cracks Gnomingstuff (talk) 06:53, 4 July 2026 (UTC)

We now have the capability to make cleanup tracking pages from a specific start date, not just the start of the ChatGPT epoch. Can you venture a guess as to when they started? I'll set up a tracker. M kuhner (talk) 14:50, 4 July 2026 (UTC)

Everything. They have no substantive edits before 2025. I don't mean to be rude, but this is something you could easily have checked yourself to answer your own question instead of making me do it. Gnomingstuff (talk) 20:16, 4 July 2026 (UTC)

You're right, I'm sorry. I should have looked before asking. You are much better at telling when LLM started than me, but there's no reason to bug you when it's obvious. M kuhner (talk) 21:28, 4 July 2026 (UTC)

No worries -- in this case it's mostly that they basically have no edits whatsoever before a given point. Gnomingstuff (talk) 22:26, 4 July 2026 (UTC)

They have reappeared and have requested undeletion of at least one prodded article (now Draft:Adetokunbo Omishakin). —David Eppstein (talk) 22:56, 18 July 2026 (UTC)

I’m the editor behind User:Thoppy234, and I like to respond to the concerns raised here. My substantial edits begin around 2025 are being taken as evidence of AI use; in reality, that timing reflects when I started contributing more actively, and I have never used AI tools for any of my edits.

The previous article on Draft:Adetokunbo Omishakin was written completely without AI. I will be rewriting the draft again now, also without AI, and making sure that every significant claim is supported by reliable, independent sources so it complies with Wikipedia’s policies on biographies of living persons, neutrality, and verifiability.

I also understand there is concern about my infobox work. I know infobox syntax well and can create infoboxes quickly, the speed of my edits comes from practice, not automation. If specific infoboxes or past edits contain errors or look like AI placeholders, I’m happy to fix or remove them once they’re pointed out.

My goal now is to improve existing content in line with policy and to avoid any future issues with AI‑suspect edits. If editors have concrete suggestions or concerns about particular parts of the article or my earlier edits, I’m willing to address them. Thoppy234 (talk) 08:19, 19 July 2026 (UTC)

And that was a 100%-probable AI generated response according to AI detection tools. We are not stupid. See Wikipedia:Do not lie about AI. And especially once you are caught at it, don't compound the problem by adding more. We want to talk to you, not your chatbot. ~Anachronist (who / me) (talk) 17:19, 19 July 2026 (UTC)

so you mean anything written is AI generated. We then need to write single sentence. ENGLISH human writing and writter still exists. Some of us are just bless with using words. Thoppy234 (talk) 23:32, 19 July 2026 (UTC)

In this post you haven't produced one single grammatically correct sentence, but you expect us to believe that you wrote four perfectly grammatically correct paragraphs without using AI. ^purring_maggot (talk) 00:32, 20 July 2026 (UTC)

So now it comes down to a single sentence with a grammar issue. Anyway, thanks for the feedback. I’m not here to trade words with a bunch of AI editors. English is not my first language, and I’ve learned my lesson about arguing over things that don’t help me grow. Civility

Reading all the threads here shows me there’s a kind of power I can’t control. Anything outside my control, I prefer to leave to those of you who are in control or to those of you whose power is English. No personal attacks

And just to be clear: saying my grammar is bad doesn’t prove the text is AI generated; read alot of articles daily and AI slop is in most of the content does not mean Human dont write . Human writers still exist, and some of us are simply comfortable playing with words in our own way. Thoppy234 (talk) 10:47, 20 July 2026 (UTC)

Why did your grammar and spelling deteriorate towards the end of this reply? ‑‑gurkubondinn 10:54, 20 July 2026 (UTC)

Because i didnt go to Reykjavik University Thoppy234 (talk) 12:31, 20 July 2026 (UTC)

Not sure what that has to do with anything. They don't even have an English Studies program (but University of Iceland does) or even a faculty of languages/linguistics, at least they didn't when I was there more than 15 years ago.

Your reply started with perfect grammar and perfect spelling, but towards the end it completely changed. All of the grammatical errors and spelling mistakes are in the last paragraph, but there are none in the previous paragraphs. How come that happened? It's a genuine question, and I'll take a reasonable answer at face value. ‑‑gurkubondinn 12:51, 20 July 2026 (UTC)

Are you thinking the above comment was AI-generated? I think it bears a lot of difference from Thoppy234's original comment in this thread, but if they can't or won't acknowledge that we can easily tell the difference between that original comment and the other comments in this thread, this conversation is almost certainly going nowhere. ^purring_maggot (talk) 12:55, 20 July 2026 (UTC)

No idea, honestly. I just wanted to know why the last paragraph was so different from the rest of the reply. ‑‑gurkubondinn 13:53, 20 July 2026 (UTC)

Ai detectors arent reliable Froglord (Finnfrog99) (talk) 19:08, 26 July 2026 (UTC)

Yeah but it was also blatantly AI generated anyways so the point is rather moot. Fermiboson (talk) 19:21, 26 July 2026 (UTC)

Do you have a source for this? I've seen studies indicating that certain detectors can achieve very high accuracy and low false positive rates. fifteen thousand two hundred twenty four (talk) 19:45, 26 July 2026 (UTC)

MIT Froglord (Finnfrog99) (talk) 19:51, 26 July 2026 (UTC)

University of San Diego Froglord (Finnfrog99) (talk) 19:59, 26 July 2026 (UTC)

Every study cited in both of those is 2023. This is a fast evolving field and 2023 is a long time ago.

It's policy here not to rely entirely on LLM detectors but they can certainly help double-check a conclusion reached by a human. One interesting study in this regard is from 2025, which showed that both experienced users and recent algorithmic detectors are quite accurate, while inexperienced humans did not do particularly well. (A problem familiar here as we deal with occasional reports that something has typos or bad grammar so must be LLM, which is definitely not true anymore, in fact it's an anti-sign.) M kuhner (talk) 20:11, 26 July 2026 (UTC)

Added a tracker. I am hoping much of it is already dealt with but this allows a systematic check. M kuhner (talk) 01:36, 10 August 2026 (UTC)

I could use any tips on how to quickly validate an infobox, because there are a squillion of them, and some at least are incorrect. M kuhner (talk) 23:20, 15 August 2026 (UTC)

Uzsport (talk · contribs · deleted contribs · logs · filter log · block user · block log)

Yet another large swath of AI edits with various people pointing out symptoms of the problem (source-to-text integrity issues/hallucinations, deleted draft notices) but no one identifying the problem.

Is unambiguously using ChatGPT for at least some things, and their articles have the usual slate of other WP:AISIGNS (infobox "add file name here" placeholders, "plays a key role" crap, AI talkpage comments, the whole gamut.

This goes back at least to mid-2025. Not sure about 2023-24; I have not done a full check of all articles. Definitely by 2025. Gnomingstuff (talk) 20:15, 4 July 2026 (UTC)

the Anthropometry section at Guljahon Turdikulova from April 2023 and from June 2023, there's also « symbols in this from July 2023 but idk if that's markup (at least seems to come from being copied somewhere) Kowal2701 (talk, contribs) 08:55, 7 July 2026 (UTC)

I think they're quotation marks on a Cyrillic keyboard or in other languages. I have them because « I write in French ». SenshiSun (talk) 19:52, 7 July 2026 (UTC)

Thank you for raising this. I want to answer honestly. I accept that I used AI sometimes, mostly for translated/paraphrased sentences, and help with references. The article where I used it more was WBC Boxing Grand Prix, mainly because I had difficulty with the brackets and formatting. I didn't understand at the time that using AI in this way could be a problem on Wikipedia. I did not use AI every time or to write full articles from start to finish, but I understand the concern. Could you please guide me on what I should do next and which issues I should focus on first? I want to correct the problems properly. Uzsport (talk) 04:21, 8 July 2026 (UTC)

Thanks for responding and clarifying the situation. You're in a better position than anyone else to know which are the AI articles, but those are the ones to start with, unfortunately the only thing to do is go back to the beginning and write them from scratch without AI by reading the sources. Gnomingstuff (talk) 15:30, 8 July 2026 (UTC)

Thanks for the honest reply. If you could clarify which articles may contain LLM text, that would really help us out, and would also protect your non-LLM work from mistaken deletion. Even a rough start and end date would be helpful. M kuhner (talk) 18:57, 19 July 2026 (UTC)

@Uzsport, can you please answer M kuhner's question above? Thanks, Kowal2701 (talk, contribs) 12:58, 14 August 2026 (UTC)

I looked at their August 11 2026 edit to William Cheung which adds a source. This source doesn't support the claims to which it's attached, though it is easy to verify that they are correct: that has the smell of recent LLM source-shuffling. M kuhner (talk) 04:13, 15 August 2026 (UTC)

Kharbaan Ghaltaan (talk · contribs)

AI slop going back to at least 2024, which I first noticed in their 2024 edits on the article for Palestine (yes, certainly an article you want written with AI); I have not looked at all their edits on grounds of their being a lot of them, but through 2025 they have unambiguously used ChatGPT for at least some purposes, and had a page (User:Kharbaan Ghaltaan/Baghdad) speedied for ChatGPT usage. Also what appears to be an AI userpage.

They were asked about this last summer on their talk page as well as on this ANI thread, which points out the marked difference in their writing style between some of their talk page comments/edits and other ones (see Talk:Palestine for a sampling, apologies to use that as an example again but it's the one I just noticed and I have not gone through their enormous amount of edits to find others)

This isn't a conduct report as going through their recent edits it doesn't seem like they are using ChatGPT for article content anymore. It's mostly to flag the older stuff for cleanup, it is very frustrating to figure out where this stuff starts and ends. Gnomingstuff (talk) 00:30, 7 July 2026 (UTC)

• Please take a look at the user talk page. It is a nightmare of rejected drafts, edit wars and earthquakes in general. The user should either clean up the old stuff themselves or be warned and blocked, and have the material deleted. If someone throws trash on the floor, get them to pick it up themselves. Yesterday, all my dreams... (talk) 00:42, 7 July 2026 (UTC)

  All those edits were previous one. Now I no more use AI for content writing Kharbaan Ghaltaan (talk) 16:44, 7 July 2026 (UTC)

  Yes, that's what I said in the original post. Gnomingstuff (talk) 18:50, 7 July 2026 (UTC)

Those older stuff is mix of AI and original written and its not that easy to filter out AI written and if that content is deleted entirely, then my entire original written content will also go in vain. Kharbaan Ghaltaan (talk) 16:58, 7 July 2026 (UTC)

You know which parts were AI and which were not. I suggest you tackle this right away. If we have to do it, you are right, some of your original writing may be lost.

If nothing else, can you give us a date when you started using AI, and a date when you stopped? M kuhner (talk) 15:31, 8 July 2026 (UTC)

There is not a proper date I remembered using AI. Just I was new and I used it and now I stopped it. The problem is, AI generated content is totally mixed with originally written ones.

Is it necessary to cleanup AI generated content, because most of them are in sandboxes and drafts, not actual articles, I guess Kharbaan Ghaltaan (talk) 07:28, 12 July 2026 (UTC)

We don't normally clean up anything but published articles and edits to published articles. I can make a cleanup report for you to look at, if you want to see what articles we would be examining. Let me know if you want to see that.

There's no way around it: if you can't tell us which articles or parts of articles are LLM, we're going to have to try to figure it out ourselves, and some of your own writing will probably get deleted. I'm sorry for that. But experience shows that LLMs can get things seriously wrong, so we can't afford to leave LLM text in Wikipedia articles. M kuhner (talk) 08:02, 12 July 2026 (UTC)

As far I remember, I didn't made any edits to published articles using LLM and whenever I did, then it was already removed. Mostly, I have used them in my draftspace and sandboxes. and also some of those articles which I wrote and published using LLMs have been either deleted, redirected or moved to draftspace. Anyway, if there is a for you to scan and inform before deleting, then I will see to it Kharbaan Ghaltaan (talk) 12:34, 12 July 2026 (UTC)

I have created a "tracking page" which lists all your edits since LLMs were publicly introduced, grouped by article. Our next step would be to go through this list and try to determine when LLM use started, then go article by article and take appropriate steps. I am not linking this page for immediate action, since you say you will help. Please have a look. If you can use this list to remember when you started using LLMRs or when you stopped, that would be extremely helpful.

Here is the tracking page: M kuhner (talk) 15:23, 13 July 2026 (UTC)

Sorry for late reply, as I have been busy nowadays. Seeing the list, one thing I am sure that most of the LLM contents were published on articles around overall Middle East topics, like countries (Iraq, Palestine etc.) and I remember, I stopped using LLMs in my India-related articles. and regarding helping, suddenly I can get busy and won't be available. Because it was all of a sudden. And please let me know that content will be removed or entire version. Because if version is not deleted, then I can rewrite those contents originally by myself. Kharbaan Ghaltaan (talk) 21:15, 19 July 2026 (UTC)

In general we delete articles that were created as LLM, and revert large LLM edits; we try not to touch anything else. M kuhner (talk) 18:25, 22 July 2026 (UTC)

regarding the edits, if the entire version is not deleted, then it will be better, because i will be able to rewrite originally that content Kharbaan Ghaltaan (talk) 09:41, 25 July 2026 (UTC)

Cleanup efforts are ongoing and being tracked at /2026-07-09 SyzygyRebel, 86 pages need review.

SyzygyRebel (talk · contribs)

Sorry for all the topics.

Anyway, another user who was warned a while back (2025) about repeated AI usage, including hallucinated sources, and responded with This is a bit embarrassing as I actually supplied all the information to be constructed while I intended to be neutral. I will go through every other created articles to clean up for anything I might have inadvertently made worse.. However, as recently as a month ago, they're still not only using AI, but doing stuff like not even bothering to remove the chatbot response. Gnomingstuff (talk) 07:12, 7 July 2026 (UTC)

Nominated one of their drafts for speedy deletion under G15 since you pointed out they can't remove the chatbot response. Beta Beta Beta - talk 07:32, 7 July 2026 (UTC)

Support presumptive reverting, they started creating articles again in 2025, this was their first. Latest edit 19 June so need consensus here Kowal2701 (talk, contribs) 08:39, 7 July 2026 (UTC)

Count me on your side for consensus. Thanks Yesterday, all my dreams... (talk) 11:15, 7 July 2026 (UTC)

Support presumptively reverting and deleting as much of this as possible. ‑‑gurkubondinn 13:44, 7 July 2026 (UTC)

Support as well. Beta Beta Beta - talk 00:36, 8 July 2026 (UTC)

Support prompt start of cleanup. M kuhner (talk) 15:29, 8 July 2026 (UTC)

Looks like July 2025 is the start date. Setting up a tracking page. It is a lot. M kuhner (talk) 04:26, 9 July 2026 (UTC)

@Gnomingstuff,@Beta Beta Beta,@Gurkubondinn, @Kowal2701, @M kuhner and @Yesterday, all my dreams...

Thank you for flagging this, and apologies for the delayed response. I haven't been active on Wikipedia recently due to work and health challenges, and I only saw these messages, including the proposed deletion notices just now.

For context, I have a repetitive strain injury that makes extended typing difficult, so I've been using AI tools to help draft text from my own dictated points, and to help surface sources I might have missed. I understand now that this produced text that reads as unreviewed AI output, and in at least one draft, unremoved chatbot text made it through, which is not acceptable regardless of the reason.

Going forward, I will not use AI to generate or rewrite article prose. I'll switch to dictation software for my own wording instead.

Any source I cite, I will personally verify against the original text before it goes into an article.

I will go through the 104 pages on the tracking list myself, finishing by end of August.

I would ask to be allowed to do this review myself rather than have it fall entirely on other editors, but I understand if presumptive reverting is still the preferred approach for pages I haven't gotten to by end of August. While I thank you for your understanding in this matter, I also want to tender unreserved apologies for all my oversight. SyzygyRebel (talk) 20:57, 11 July 2026 (UTC)

Thank you @SyzygyRebel, that sounds like a good plan. The most important thing to review for is whether the cited sources support the content (ie. WP:V), because LLMs have a tendency to hallucinate. Hope your injury gets better soon. Best, Kowal2701 (talk, contribs) 21:06, 11 July 2026 (UTC)

Thank you very much for your honesty and cooperation. I'd certainly support you taking responsibility for the cleanup, and like the idea that we'll touch bases again at the end of August.

Please note that LLMPRODs are often done in batches, and only marked on the tracker when someone else notices that they have been done. So there may be LLMPRODs not currently marked on tracking.

I would strongly recommend, if you intend to rescue a created article currently under LLMPROD, removing the LLMPROD tag and immediately draftifying the article. Unreviewed LLM articles should not be in mainspace, and this will give you time to fix them. If not rescued they are likely to be deleted after 5-7 days. You can see instructions at WP:DRAFTIFY or ask here for help. M kuhner (talk) 21:13, 11 July 2026 (UTC)

WP:USERFY might be better because we've had problems with people draftifying in the past (like someone would draftify an article with the intention to work on it there, then someone else moves it back to mainspace later without it undergoing any changes) Kowal2701 (talk, contribs) 21:18, 11 July 2026 (UTC)

Thank you so much for your kindness and for your understanding @Kowal2701 and @M kuhner I really appreciate it.

I'll start fixing the flagged issues right away and work through as much as I can, article by article. I'll come back to this thread if I run into anything I'm unsure how to handle. I take Wikipedia's quality standards seriously, and even assuming good faith on my part, I'm genuinely unhappy that this situation arose at all.

For now, I'm going to focus solely on fixing these existing flagged articles and will hold off from creating or curating any new articles until I'm confident I can produce reliable, properly verified work going forward.

Thank you, once again. SyzygyRebel (talk) 22:57, 11 July 2026 (UTC)

You're very welcome. Your response is 1000x better than what we usually see, and frankly made my day. M kuhner (talk) 23:12, 11 July 2026 (UTC)

@M kuhner, oh, thanks for acknowledging that. I value this community and I am also learning to do better. SyzygyRebel (talk) 00:54, 12 July 2026 (UTC)

Piece of advice, not related to LLMs, on Caroline Kouassiaman: Source 1 lives on a web site which says it will close down in December 2026. Strongly recommend that you archive it now while it's still alive! M kuhner (talk) 08:07, 12 July 2026 (UTC)

@M kuhner, thank you for flagging this. I had missed it. I have now archived it. Please, if you are not too busy, kindly assist to review the pages I have fixed thus far in case I have missed other important things. Plenty gratitude. SyzygyRebel (talk) 22:00, 12 July 2026 (UTC)

I checked Caroline Kouassiaman and have put comments on its Talk page. I am sorry to say, but you are underestimating how bad LLM sourcing is, and there are still multiple sourcing issues here including one flat-out hallucination. M kuhner (talk) 15:39, 13 July 2026 (UTC)

@M kuhner Oh, no! I will go back to it. SyzygyRebel (talk) 21:08, 13 July 2026 (UTC)

@M kuhner, I have been wondering about some of the issues that @Gnomingstuff, and I discovered that a user Special:Contributions/~2026-22682 had copied and pasted an entire mathematical bold Unicode characters paragraph into Satang Nabaneh's article. And I was not notified of this editing and I was never aware of it. Isn't this a kind of garbled formatting from a copy-pasted source vandalism? SyzygyRebel (talk) 12:41, 14 July 2026 (UTC)

This user is now removing their own LLMPRODs with little to no action taken. I've seen it on other articles, too: . I would suggest an indef mainspace block based on this disruption, as the user is clearly not following through with their promise to "focus solely on fixing these existing flagged articles". Somepinkdude (talk | contribs), in solidarity 15:43, 16 July 2026 (UTC)

Despite their apparent cooperation, there have been source/text mismatch issues in both of the "cleaned up" articles I've spot-checked so far (see talk pages for Caroline Kouassiaman and Erica Woodland. I don't think the cleanup is sufficiently rigorous. M kuhner (talk) 15:50, 16 July 2026 (UTC)

@Somepinkdude and @M kuhner I am still editing. I didn't just remove them. I am compiling. I asked to finish all the editing in August. I have been taking time from work as busy as I am with my health to edit. I was editing everything as much as I can at the same time. I have no reason to take any of you for granted or not comply with community standards. Please, instead of recommending punitive actions, especially where I am complying, I will appreciate your help in any way possible. I was sleeping at 4am including working all day to get this done. Thanks for your understanding SyzygyRebel (talk) 18:03, 16 July 2026 (UTC)

Have you considered draftifying the articles (moving them to Draft) so that there will be no time limits? We certainly don't want you to damage your health.

From experience doing it, LLM cleanup is very difficult. In many cases it might be better to cut down the article to a stub, then rebuild with human text. You can salvage any good references, but will have to read them as what LLM says they contain is not reliable. It's a big project, which is a major reason we're so hardline on LLM use: a problem can be caused very quickly that will be extremely slow to cure other than by deletion.

Please do not write or edit other articles until this is dealt with. Thank you. M kuhner (talk) 18:20, 16 July 2026 (UTC)

@M kuhner, oh no. I do not intend to edit or write any article for a while even after this problem is fixed. I am taking this to heart too much to do that because integrity matters to me. If you look at distance between when I joined Wikipedia, wrote the first set of articles and came back again, it was a loooong time, because I do not have the kind of software skills that many people do here and I do my best as someone who loves to contribute to meaningful project to learn how best to participate, still. For instance, some of the errors you flagged like Early life and education on Erica Woodland's article was my error, not LLM's. And I used LLM because I thought it was accepted but minimally, especially when human contribution is involved, and can help with dictations for people who are neurodivergent, suffering some form of disabilities or have health concerns who want to participate in meaningful things. That said, the reasons I have not moved any article to drafts is; 1. because I thought some of the problems are quick fixes, 2. that even if I dratify, I may not still see all the problems and someone else can approve it as ready while I still get into trouble, after all most of the articles have been reviewed and accepted while having all these issues. 3. I have not received much help in terms of anyone helping to contribute to articles in AFC even when I declared a COI. It felt like I am largely on my own in this community, so I learnt to just do my best otherwise as the Women in Red, and LGBTQ+ Studies projects are meaningful for me. I am not making excuses but being transparent and accountable in this process and appealing that I am not punished for unintentional mistakes but supported to address them. I hope I can have the corporations of everyone who has joined the conversation to help me out like you are doing by assisting to look through all the 16 articles under scrutiny for what else I might have missed so I can go back and fix them. Your pointers on Erica Woodland, and Caroline Kouassiaman have been nothing but helpful, and I hope others kind editors will support me too. @M kuhner, would you kindly educate me on how to revert an article to a stub or build one? I have seen stub articles before and helped to develop them but I have never created one because I thought it was never intentional but an editing decision. Thank you. SyzygyRebel (talk) 07:52, 17 July 2026 (UTC)

As long as you are genuinely trying to improve the encyclopaedia in good faith, you don't have to worry about being sanctioned. They are WP:NOTPUNISHMENT, so they are not "punitive actions". Sanctions are only restrictions that are enacted to protect the project when needed.

I would think that the best thing to do here is to WP:USERIFY the articles, and move them into your userspace. Then there are no timelimits and no stress. Remember that there are WP:NODEADLINES on Wikipedia, there is no reason for any of this to become a stressful situation for you. If you want to, I can do it for you. Just let me know. ‑‑gurkubondinn 08:25, 17 July 2026 (UTC)

@Gurkubondinn, Thank you so much for your understanding and for this helpful suggestion.

I think the reason I've been so stressed and autoautopiloting is that @Gnomingstuff had recommended a block, twice, and I've been doing my best to make sure that didn't happen. I also noticed that @Kowal2701 suggested moving the articles to my userspace rather than the Draft namespace, but I didn't really understand what that meant until you reiterated it now.

Before I move the articles to my userspace, may I ask whether someone could look over all the editing done and point out what I am still missing so I can see if I can fix them without moving them?

Once I review my options based on how much correction I still have to do on each article, I will come back to you for assistance in either to move or show me how, in case I haven't figured it out myself.

Thank you, and @M kuhner for all the help rendered so far. SyzygyRebel (talk) 10:30, 17 July 2026 (UTC)

@Gurkubondinn, I actually thought there were deadlines because what I know thus far is that unfinished drafted articles would be deleted after 6 months, and that I had 7 days to resolve the issue here or all the articles will be deleted, hence rushing to fix them as I have done. SyzygyRebel (talk) 10:38, 17 July 2026 (UTC)

I don't recall bringing up being blocked and don't see anything here that is worth being blocked over. It wasn't here if so. Gnomingstuff (talk) 12:42, 17 July 2026 (UTC)

@SyzygyRebel: Drafts that haven't been touched in six months can be deleted if they are in the Draft namespace, or are in the User namespace and have an AFC submision template. If you keep them in your userspace and don't submit them to AfC or move them to mainspace before they're ready, they'll be safe in your userspace and you can take your time.

And thank you for your efforts to counter systemic bias on Wikipedia. Such important work deserves to be done thoroughly and with care. Representation is important, but there is no deadline. —ClaudineChionh (she/her · talk · email) 11:10, 17 July 2026 (UTC)

I want to echo the last paragraph in Claudine's response. We need more of this here on the English Wikipedia. We have a lot of blind spots that need filling in. ‑‑gurkubondinn 11:20, 17 July 2026 (UTC)

I am sorry @Gnomingstuff, I must have misread you. I apologise for that. Thank you @ClaudineChionh, and @Gurkubondinn. I appreciate your support and reassurance. May I take a few days to decompress and return to look through all the articles to see which ones need moving to my userpage and which have been sufficiently salvaged? I have already worked through all 16 articles over the past 5 days, but based on @M kuhner's comments, I still have work to do. I am still hoping one of you might have time to look at the articles for me. Your feedback would be of immense help. Thanks a lot. SyzygyRebel (talk) 20:50, 17 July 2026 (UTC)

No need to apologize, and also just want to echo what everyone else said as well, really really appreciate you doing this Gnomingstuff (talk) 20:57, 17 July 2026 (UTC)

I have put comments on a few more of the articles in the cleaned-up list. I'd suggest going back to Godwin Murunga right away as there are mismatches between text and source in every paragraph. As I said in my comments there, it really is better to rewrite the text based on reading the sources: salvaging the LLM text seems appealing but it turns out to be more trouble than it's worth. (Speaking from experience--for example for New Vision I wish I'd started clean rather than trying to sort out someone else's LLM text, Hypocnemis too. I badly underestimated how hard these would be; wouldn't do it that way now that I have more experience.) M kuhner (talk) 01:32, 20 July 2026 (UTC)

@Gnomingstuff, thank you so much for your support and everyone else's.

@M kuhner, massive appreciation for all your help. I have decided to dedicate next week to cleaning the articles and call upon @Gurkubondinn to help move any article I cannot salvage to my userpage or teach me how to do it by the end of the month if I am stuck. I will however go to Godwin Murunga article, right away, to address your comments so I don't leave it unattended until next week. I am also paying keen attention to your recommendations to rewrite some of these articles. It won't be fun but it is necessary. SyzygyRebel (talk) 09:13, 22 July 2026 (UTC)

Top of the day to you all. I am starting the cleaning up afresh as promised. I believe I may be able to complete it without moving the articles out of mainspace. This may be somewhat ambitious, but I say so because I feel I now have a better grasp of what needs to be done. However, if I still fall short, it will be due to human error on my part, and the possibility that I may be blindsided to things I am missing. I would therefore appreciate it if you could still nudge me in the right direction should you spot anything amiss. Thank you so much for all your support thus far.

Completed CleanUp👇🏾

1. S. M. Rodriguez: ✅ Done
2. Zeedah Meierhofer-Mangeli: ✅ Done
3. Toyin Ajao: ✅ Done — confirmed by M kuhner, no issues found

SyzygyRebel (talk) 12:38, 27 July 2026 (UTC)

I reviewed S. M. Rodriguez and left comments on the Talk page, but despite a few small flaws, I think you've nailed this one. Well done. M kuhner (talk) 04:11, 31 July 2026 (UTC)

I am happy to read this 🤸🏾. Thank you so much for all your help in resolving the flagged issues @M kuhner.I have addressed the points raised in S. M. Rodriguez's talk page. I am now done with Zeedah Meierhofer-Mangeli. I will appreciate if you can look at this as well at your convenience. SyzygyRebel (talk) 19:31, 31 July 2026 (UTC)

Hi, everyone, just some of the articles I have worked on including where @M kuhner had left comment for me. I am still on them doing my best to not miss what I had previously missed. I haven't rewritten any from the scratched because I hadn't seen any I needed to begin all over yet. Stating this here for accountability. For the ones I have salvaged, if there's still any issue with them, I take full responsibility for my oversight and seek your help to nudge me back where I must look. I have also not come back to @Gurkubondinn for the user page move if necessary but I am keeping this in mind. Thank you.

Completed CleanUp👇🏾

4. Maxi Schoeman: ✅ Done
5. Brian Kagoro: ✅ Done
6. Romi Crawford: ✅ Done
7. Priscilla Ikos Usiobaifo: ✅ Done
8. Munira Hussein: ✅ Done

SyzygyRebel (talk) 00:51, 5 August 2026 (UTC)

9. Siphokazi Magadla: ✅ Done
10. Godwin Murunga: ✅ Done
11. Caroline Kouassiaman: ✅ Done
12. Mshai Mwangola: ✅ Done
13. Ruth Murambadoro: ✅ Done
14. Eka Ikpe: ✅ Done

SyzygyRebel (talk) 01:24, 9 August 2026 (UTC)

Hi everyone 👋🏾. I hope you're all doing well.

I wanted to update you that I have now cleaned up fifteen articles, with just a few more to go. I'm confident I can complete the rest on or before the end of the month as initially requested. I appreciate your continued patience.

I also want to thank @M kuhner, once again, for all the assistance. Thank you. Your helpful comments on the talk pages of some of the articles have made a real difference throughout this process. 💜

I have a question, is Wikipedia:AI noticeboard/2026-07-09 SyzygyRebel tracking page automatically updated once I cleanup all the articles? I am just curious about the process as I am eager to see if I have made any progress worth updating. I look forward to your feedback.

Thank you in advance. SyzygyRebel (talk) 22:27, 13 August 2026 (UTC)

Completed CleanUp👇🏾

Satang Nabaneh: ✅ Done

Naminata Diabate: ✅ Done

SyzygyRebel (talk) 22:28, 13 August 2026 (UTC)

They don't auto-update but can be easily updated by hand. I'll try to have a look within the next few days. M kuhner (talk) 22:33, 13 August 2026 (UTC)

Roger that @M kuhner. Thanks. SyzygyRebel (talk) 14:10, 14 August 2026 (UTC)

 Cleanup has been requested and is being tracked at /2026-08-10_0xReflektor, 117 pages need review.

0xReflektor (talk · contribs · deleted contribs · logs · filter log · block user · block log)

see here, here. and there

I do not know if they were using LLMs on other edits, but they have a pattern of using LLMs whenever un-redirecting an article ~2026-38678-64 (talk) 03:01, 8 July 2026 (UTC)

Forgery and Counterforgery (created Sep 2025) has WP:AIVOCAB and WP:AITREND ("highlighted") ("sets the New Testament writings within that broader landscape"), as well as WP:AITABLE. A lot of ther articles are like this. Wikipedia:Articles for deletion/High-speed solar wind was Nov 2025. Some sizeable articles were created very soon after one another. This looks like a rather straightforward LLMPROD run. Will probably do it tomorrow if noone else fancies it Kowal2701 (talk, contribs) 23:04, 16 July 2026 (UTC)

Happy to do it. Your content creation area deserves attention over this. Did you use something fancier than copy-paste to tag Juan Martínez de Medrano? I activated Twinkle not too long ago. Cedarshadow (talk) 23:37, 16 July 2026 (UTC)

Thanks @Cedarshadow, I use User:Kowal2701/LLMPROD to copy-paste the tags and edit sums, unfortunately it hasn't been added to Twinkle yet Kowal2701 (talk, contribs) 09:55, 17 July 2026 (UTC)

Don't worry, I've done the LLMPRODs (honestly quite enjoy them for some reason) Kowal2701 (talk, contribs) 12:18, 27 July 2026 (UTC)

I noticed one of the recent deletions and ventured to recreate it free of AI. I also noticed a related article was un-redirected by this same user and appears to be AI-generated. Going through many of their large contributions indicates a great deal of LLM use. This will indeed require much cleaning up. Οἶδα (talk) 05:03, 8 August 2026 (UTC)

I think September 2025 was the start of LLM use (next most recent edit is September 2024, so a big gap). Making tracker accordingly. M kuhner (talk) 04:57, 10 August 2026 (UTC)

Can some of these AI articles just be promptly deleted? I am thinking of Pauline interpolations and forgeries as it is all AI. Ramos1990 (talk) 19:19, 11 August 2026 (UTC)

most of their creations have been deleted under WP:LLMPROD, but that doesn't apply to ones that have significant edits from other contributors. In this case, that contributor's you lol. I'll LLMPROD it and point to your comment here Kowal2701 (talk, contribs) 19:36, 11 August 2026 (UTC)

LOL. Appreciate you very much. Thanks for the support. Ramos1990 (talk) 01:42, 12 August 2026 (UTC)

I just want to marvel at this edit: However, [[Nina E. Livesey|some scholars]] continue to challenge the authenticity of letters within this group. That is the LLM going out of its way to do the thing they like to do, namely claiming that a lot of people support something when really there is only one! I did not expect it to be quite so blatant. (At least, I hope that that's LLM, because if it's human, it's exceptionally slimy.) M kuhner (talk) 23:25, 15 August 2026 (UTC)

That's art Kowal2701 (talk, contribs) 23:28, 15 August 2026 (UTC)

 Medelinezezie (talk · contribs · logs · block log) (13 August 2026)
Editor has been blocked site-wide for AI/LLM policy violations. 

 Cleanup has been requested and is being tracked at /2026-08-15_Medelinezezie, 12 pages need review.

This account has existed since 2023, but the vast majority of their 209 edits have come in the past few days. Their first edit after a 2.5 year hiatus was creating a clearly LLM-generated article on a music artist, which was sent to AfD. I also very kindly asked that they refrain from publishing LLM-generated content. The user proceeded to absolutely bludgeon the AfD discussion, acting very offended that their obviously LLM-generated text was identified as such.

Their second article, Yeyo (musician), also appears like it could be LLM-generated, although it is certainly not as obvious. They then started making huge additions to existing articles on music artists, adding thousands of bytes of content. Not only were these posted mere minutes apart, but they also contained the classic long-ass edit summaries, such as Article cleanup: improved neutral wording and organization, removed promotional language, expanded the career chronology through 2026, added sourced releases and chart information, updated the discography, and corrected formatting and citations. I believe that these additions (1, 2, 3, 4) were probably LLM-generated.

The user then moved on to other topic areas, making the same huge additions minutes apart with excessive edit summaries: (1, 2, 3). As of two days ago, they seem to have moved on to space-related articles. I'm not gonna post those diffs cuz they're a lot, but they've started creating articles again, adding and deleting hundreds of thousands of bytes worth of content in the past 48ish hours. Galaxy scaling relations, a 58k+ byte article, was published 6 minutes after publishing a 38k+ bytes article.

When asked again on their talk page how this was possible, they again got offended and denied the accusation, saying they've been working on these articles for months before publishing. I think it's time someone takes a closer look at all this silliness. JTtheOG (talk) 20:15, 10 July 2026 (UTC)

Re-wrote early life with verified sources to not contain no AI generated paragraphs.

at least they're honest Gnomingstuff (talk) 01:22, 11 July 2026 (UTC)

I looked at Galaxy scaling relations which has a recent-LLM tic of citing the material in the first half of each paragraph and then running on without citations in the second half. It foregrounds a lot of trivial math for no clear reason. Variable names come and go (what is gamma in the evolving zero point equation?) The Historical development section is completely unsourced and is just a jumbled list. Comparison of selected relations just below it is another unsourced list but this time in table format. Also duplicates material between citations and Further reading. The combination of extreme technical polish and intellectual emptiness is what I'd expect from more or less pure LLM.Interesting topic but the article doesn't say much despite 58K text. Super frustrating to read as someone who is actually interested in this sort of thing. M kuhner (talk) 15:46, 11 July 2026 (UTC)

The use of LLM in the deletion discussion Wikipedia:Articles for deletion § Gone Thursday is painfully clear: it hits the entry in the list I made yesterday about responding to a different argument than what was made. I think that we should lay out the case and ask for admin intervention immediately. M kuhner (talk) 15:54, 11 July 2026 (UTC)

I'd say wait until the next substantive contribution. We need more articles on topics like this editor is creating, and less pop-culture, autobiography, and COI company articles. We just need the editor to do the writing. ~Anachronist (who / me) (talk) 01:32, 12 July 2026 (UTC)

Could you further explain what you mean? JTtheOG (talk) 01:48, 12 July 2026 (UTC)

Meaning I'd like the editor to have a chance at redemption before blocking. ~Anachronist (who / me) (talk) 06:45, 12 July 2026 (UTC)

Well, I agree that it'd be great if they bothered addressing the concerns. JTtheOG (talk) 03:31, 13 July 2026 (UTC)

The replies from the editor are still there on User talk:Medelinezezie, where she claims she has lengthy research she did himherself already prepared, she uses text to speech tools, and she wasn't happy with what an AI gave himher in the past. It's up to the community to decide whether assuming good faith is warranted. ~Anachronist (who / me) (talk) 04:32, 13 July 2026 (UTC)

@Anachronist the user probably uses she/her pronouns (also, please use they/them when unsure of someone's pronouns please) ~2026-38678-64 (talk) 04:36, 13 July 2026 (UTC)

Woops, you're right. I'll correct my comment. ~Anachronist (who / me) (talk) 05:18, 13 July 2026 (UTC)

User has now cleared most of their talk page and resumed creating articles on galaxies (Kormendy relation and Abell 3667, created 30 minutes apart). It's possible they're no longer using AI, but it's still quite concerning that none of this has been addressed. JTtheOG (talk) 03:27, 13 July 2026 (UTC)

@Admins willing to patrol AINB: Could we temporarily block User:Medelinezezie from article space as the concerns raised here haven't been addressed and the user is continuing to create articles? thank you ~2026-38678-64 (talk) 03:36, 13 July 2026 (UTC)

I am looking at Abell 3667. It claims redshift of 0.0553 sourced to 1, which says 0.0550: while this could be a simple typo, source 5 says 0.0553. (struck out because turned out to be an error on my part) 23:55, 13 July 2026 (UTC)

Article says there is a northwestern and a southeastern radio relic, sourced to 5 which consistently says they are northwestern and southern; however source 6 uses northwestern and southeastern.

Sources exist and support claims, except for the two cases above where a claim is cited to source A but actually found in source B. While this is typical of recent LLM, it is also an easy mistake to make if you are familiar with the literature and writing from memory.

There is quite a lot of close paraphrase, though with material this technical it's hard to avoid.

(Disclaimer: I am a biologist and statistician, not an astronomer.) M kuhner (talk) 04:34, 13 July 2026 (UTC)

yeah, there's also a subtle error in Kormendy_relation where the article says that the kormendy relation is "a scaling relation between the effective radius of a galaxy and its effective surface brightness. It is mainly observed among elliptical galaxies " but most sources i see online agree that the kormendy relation is "a scaling relation observed between the effective radii of elliptical galaxies and their surface brightnesses at the effective radius." which is a subtle difference but seems to change a lot ~2026-38678-64 (talk) 04:45, 13 July 2026 (UTC)

Looking at Kormendy relation too. The author does say Some studies instead use the surface brightness measured directly at R_e. which I think captures what you're saying. I'm more bothered by the fact that I cannot find anything in source 2 to support what I just quoted: as you say, it's true, but it's not in that paper. I also wonder why (kpc) is in the formula in the article and not in the paper used to source it. The reminder that magnitude has reversed scale isn't in there either, though it's true. I'll keep going. M kuhner (talk) 05:05, 13 July 2026 (UTC)

i reverted my comment and now you're replying to nothing! argh let me do some magic ~2026-38678-64 (talk) 05:12, 13 July 2026 (UTC)

Don't sweat it.

I am not an astronomer but I think the first part of "Interpretation and use" is bogus: specifically The Kormendy relation is a two-dimensional projection of the fundamental plane of early-type galaxies. I can only see an unsearchable image of the cited source 3, but I'm pretty sure it does NOT say this. It says there are two independent axes describing elliptical (not "early-type") galaxies, of which one is Kormendy and the other is based on velocity. I'm not at all sure what a two-dimensional projection of a plane would even be: aren't planes inherently two-dimensional?

M kuhner (talk) 05:35, 13 July 2026 (UTC)

that probably comes from https://iopscience.iop.org/article/10.1086/340021/meta which says "Finally, the structural parameters enter the fundamental plane of early-type galaxies, of which the Kormendy relation is a two-dimensional projection." ~2026-38678-64 (talk) 05:41, 13 July 2026 (UTC)

you can search the full-text for source 3 here (yay open access! nay bad user interface) ~2026-38678-64 (talk) 05:48, 13 July 2026 (UTC)

Thanks. I think I can say with moderate confidence that there's a pattern of citing source A for fact found in source B.

I am not sure what so a linear fit derived for bright galaxies cannot automatically be extended to every spheroidal system means. It is cited to 6 which uses the word "spheroidal" only to talk specifically about dwarf spheroidal galaxies. This may be another source A/source B issue.

If it is LLM, though, it's some of the best I've seen; far less vacuous than Galaxy scaling relations. More likely, I think, a human/LLM joint project where the human knows what they are doing. M kuhner (talk) 05:53, 13 July 2026 (UTC)

I have gone as far as I can with these: we need an astronomer. Learned a lot through reading all those papers, though! M kuhner (talk) 05:57, 13 July 2026 (UTC)

i think that we have enough info and i don't think it'd be interesting or necessary to get cool people to review an article which we're moderately sure is either LLM-generated or where something has gone catastrophically wrong ~2026-38678-64 (talk) 06:08, 13 July 2026 (UTC)

Source 5 also had an invalid bibcode that lead to an 404 not found. That generation has since been deleted under G15, but now Abell 3667 has been regenerated again. ‑‑gurkubondinn 09:55, 13 July 2026 (UTC)

seems to be substantially the same article, with the citations leading to the same places ~2026-38678-64 (talk) 12:34, 13 July 2026 (UTC)

Yep, I guess AfD will be required since they decided to paste in a new generation. ‑‑gurkubondinn 12:37, 13 July 2026 (UTC)

a user (myself) has made me recently aware that i probably didn't see the original before-june-13-4-UTC article, which, oops ~2026-38678-64 (talk) 12:51, 13 July 2026 (UTC)

It's in the filter logs: AbuseLog/44612907. ‑‑gurkubondinn 13:43, 13 July 2026 (UTC)

oh, interesting (how did you find it? i don't know how to search in the filter logs) ~2026-38678-64 (talk) 15:31, 13 July 2026 (UTC)

I just remembered that I had previously seen that it got caught in the 'large creations by inexperienced user' filter, so I looked at the filter logs for Abell 3667. Even though a page is deleted ('removed' is a more accurate term), the filter logs will persist. ‑‑gurkubondinn 16:49, 13 July 2026 (UTC)

oh, awesome! ~2026-38678-64 (talk) 19:07, 13 July 2026 (UTC)

I can't see where the error is regarding the redshift. Abell 3667 says "It has a measured redshift of 0.0553", which matches the source, which says "from this sample, we derive a cluster redshift of z = 0.0553 ± 0.0002". ARandomName123 (talk)^{Ping me!} 20:57, 13 July 2026 (UTC)

I believe that the article has been deleted and recreated since I wrote that, and that in the original version the redshift was sourced to source 1 and not source 2. But it is very possible that I made a mistake. M kuhner (talk) 23:03, 13 July 2026 (UTC)

Abell 3667 was a 36k article before you nominated it for G15 deletion, and I deleted it. Now it's a shorter 6K article. In both versions the redshift was sourced to Owers et al, which is source 1 in the deleted version and source 2 in the current version. ~Anachronist (who / me) (talk) 23:11, 13 July 2026 (UTC)

Did you mean to reply to me instead? But yeah, it's not the same article (and has fewer references), hence why I said 'regenerated' somewhere above. ‑‑gurkubondinn 23:16, 13 July 2026 (UTC)

In that case I must have made a mistake, and will strike out this comment above. M kuhner (talk) 23:53, 13 July 2026 (UTC)

Update: User has responded to my question on their talk page and claimed that they have only used AI for Gone Thursday. However, edit summaries such as these are concerning:

• Expanded and reorganized company history and operations; updated leadership and 2026 acquisition information; added public-company, Sony ownership, catalogue, digital-video and former film-division coverage; added freely licensed Commons images; removed repetition and outdated update template.
• Expanded and reorganized the article chronologically; updated Farruko’s career through 2026; corrected grammar, dates, album classifications, and repetitive wording; added sourced coverage of his musical style, Carbon Fiber Music, major releases, faith-based period, recent projects, tours, selected awards, and freely licensed images.
• Created an article on IC 4965 covering its identification, morphology, luminosity, role as the brightest galaxy of Abell 3667, peculiar velocity, head-tail radio source, intracluster-light envelope, 400-kiloparsec stellar bridge, discovery, coordinates, equations, images, and astronomical catalog designations.
• Created a comprehensive article on JO171, covering its detached stellar ring, central spheroid, counter-rotation, stellar populations, ionized gas, metallicity, ram-pressure stripping, Abell 3667 environment, formation history, MUSE and ultraviolet observations, equations, tables, and related images. 2months of editing and research. Ready to go live and share knowledge.
• Comprehensive cleanup and reorganization: improved grammar, clarity, consistency, and encyclopedic tone; corrected citation and formatting issues, repaired links, removed repetition, clarified scientific claims, and expanded the article with sourced material on cooling flows, AGN feedback, X-ray cavities, the radio mini-halo, molecular gas, and JWST observations.

Strangely enough, they have abruptly stopped these overwhelmingly exhaustive edit summaries after this discussion was brought up. JTtheOG (talk) 07:05, 13 July 2026 (UTC)

Note: I had moved all three galaxy to draftspace, So the issue can be solved. Abdullah1099 (talk) 06:49, 14 July 2026 (UTC)

I had wondered about some of the galaxy and physics pages as part of NPP. Having read this, I have completed the cleanup by draftifying in a couple of cases, asking other editors to check in another and doing a CSD on Draft:Abell 3667 so Abell 3667 can also be draftified. In my opinion all their pages should at a minimum now go through AfC. I leave open the quest of a page creation block. Ldm1954 (talk) 02:26, 10 August 2026 (UTC)

@Ldm1954, I am bit confused if the work on Abell 3667 is from User:Mu301/Abell 3667 then this mean Abell 3667 is a copy of it Abdullah1099 (talk) 02:49, 10 August 2026 (UTC)

@Abdullah1099, if you check the history you will see that Mu301 created a very short draft in Oct 2025, apparently they never submitted it. They moved it to their user space in March 2026 before it hit the 6 month delete, but left a redirect. Much later the page Abell 3667 was created. The infobox looks almost identical, the image is the same which might mean nothing. I see no evidence that Mu301 did anything wrong beyond leaving a hanging chad redirect in draftspace. Ldm1954 (talk) 03:02, 10 August 2026 (UTC)

Ok, Thanks Abdullah1099 (talk) 03:31, 10 August 2026 (UTC)

My draft was merely an infobox and a list of references. There really wasn't much there to copy. I merged a couple of those refs into the current draft. --mikeu ^talk 17:11, 10 August 2026 (UTC)

Ok, I understood Abdullah1099 (talk) 03:01, 11 August 2026 (UTC)

Blocked Medelinezezie sitewide for LLM misuse and for belligerently denying the full extent of her LLM misuse when the issue was raised. I have verified that all of the issues identified in the associated deletion log entry did exist in Medelinezezie's 9 July 2026 creation of the Abell 3667 article, and I have confirmed that Medelinezezie has been misusing LLMs to generate content on multiple articles, contrary to her denials. Also, the last "Further reading" entry in Special:Permalink/1363301103 has an invalid Bibcode, and citation #2 in Special:Permalink/1363295425 has an invalid Bibcode. A number of Medelinezezie's article creations have nonexistent "See also" entries () and nonexistent categories (). — Newslinger talk 05:26, 13 August 2026 (UTC)

Tracker created, however I don't know for sure when LLM use started. Only a few non-recent edits though. M kuhner (talk) 05:02, 15 August 2026 (UTC)

Actually it's very clear: all the non-recent edits are human. July 7 2026 was the LLM start date. But there are a fair number of HUGE edits to existing articles after that date that will not be easy to deal with. M kuhner (talk) 05:11, 15 August 2026 (UTC)

• Afus199620 (talk · contribs · deleted contribs · logs · filter log · block user · block log)
• article count

This user has made 22,000 edits. I don't think all of them are AI, but I do know the most recent ones are. They've made five articles within minutes of each other. All show signs of AI, for instance, the first diff of this article is filled with broken refs. I've already nominated Rutunga for CSD during NPP.
First hoping for a response from the user, however looking for a consensus here on whether they should be drafted or deleted. Fortek67 (talk) 20:31, 10 July 2026 (UTC)

The articles where translations from short German articles. The text is not AI. I wont make these translations anymore.--Afus199620 (talk) 21:06, 10 July 2026 (UTC)

Nobody's stopping you from making these translations. It's just that they don't seem to be of the best quality. Are they machine translated? How exactly did you translate them/what did you use? Because once again this seems to have been done very quick. Fortek67 (talk) 21:36, 10 July 2026 (UTC)

Yes, I used a automatic translation program with editing. You are completely right. It was too fast. I will be more careful. --Afus199620 (talk) 07:10, 11 July 2026 (UTC)

You should take your own advice and check more thoroughly next time. Lupe (talk) 07:22, 11 July 2026 (UTC)

unreviewed/poorly reviewed LLM translations are still subject to the same issues as normal LLM-generated content, hallucinations and WP:V failures are the biggest problem Kowal2701 (talk, contribs) 10:01, 11 July 2026 (UTC)

I know that you have to review sources of machine translated articles and that that takes maybe a minute or so longer even for these short articles, but that wasn't the accusation (it wasn't even mentioned that the articles are translations). So I stand by my statement for him to work on himself. Lupe (talk) 11:39, 11 July 2026 (UTC)

I don't see that Fortek said anything wrong here? The articles were AI-generated and problematic, it's just that we allow AI-generated translations that are reviewed. Though I agree saying that they're translations in the OP would be ideal Kowal2701 (talk, contribs) 11:46, 11 July 2026 (UTC)

Not just ideal but required. The original source needs to be credited. See WP:Translation. M kuhner (talk) 15:56, 11 July 2026 (UTC)

(I was referring to Fortek's OP here, but yeah, unattributed translations are a pain) Kowal2701 (talk, contribs) 16:03, 11 July 2026 (UTC)

I'm sorry, I wasn't clear: what I was trying to say is that if these are translations, they are unattributed translations, which is also a policy violation. M kuhner (talk) 16:07, 11 July 2026 (UTC)

The Translations where actually not unattributed. The revision history showed that these articles were translations.--Afus199620 (talk) 17:44, 11 July 2026 (UTC)

You are right. I apologize and will strike out my remark. M kuhner (talk) 17:48, 11 July 2026 (UTC)

(The following has been copied from a separate case, with PARAKANYAA's permission, as they turned out to be about the same editor.) M kuhner (talk) 03:47, 26 July 2026 (UTC)

This is more subtle and I'm not sure if I'm being paranoid but I'd like a second opinion. My concerns are about these edits to Peter Levenda I removed a chunk of it for being undue weight on the author's self-description /borderline copyvio but reading further I think AI might have been involved. Not sure though. My reasoning:

1) a lot of the analysis of the books is not in the sources and reads, to me, like AI extrapolating the contents from more dry summaries in reviews. The above section is just basically copying from the author summary (which also seems like copyvio ) so this mostly shows up in the views section

2) phrasing like Levenda's books generally approach occultism as a historical, political and religious phenomenon rather than as a purely literary subject. Unholy Alliance examines the relationship between Nazi ideology, esotericism and occult currents, while later works such as Sinister Forces and The Dark Lord connect occult history with American political culture, H. P. Lovecraft and modern esoteric movements. In the Sekret Machines series, written with Tom DeLonge, Levenda applied similar themes to the modern UFO and UAP discourse, arguing that the subject should be considered in relation to religion, mythology, consciousness and political authority rather than only as a question of aerospace technology and was described in contemporary press coverage as being based on interviews with intelligence officers, scientists, engineers and military officials really smell of AI to me.

3) repeating the same information across multiple redundant sections, e.g. it covers the UFO books twice? PARAKANYAA (talk) 17:51, 25 July 2026 (UTC)

Oh, I notice this user is already being discussed above (Afus199620). My bad. Still, the problem here is a bit broader than what is discussed there, the problem goes beyond translations. PARAKANYAA (talk) 17:57, 25 July 2026 (UTC)

 Mcapdevila (talk · contribs · logs · block log) (13 July 2026)
Editor has been notified and asked to discuss AI/LLM editing concerns. 

Hello all, I am not too experienced with all of wikipedia but I wanted to bring to attention this article as it looks like it has been completely made with an LLM, first on another language wiki and then translated/generated into English? The main thing that stood out to me other than the general vibe of the text is the references.

Not a single one is actually linked to anything and in particular, the report referenced as "Scientists skeptical of new Columbus DNA claims" by The Guardian does not exist and no reports about such a thing can be found before 2024. I invite more experienced editors to look into this, please. In fact, when I look at the user who created that page it looks like they use LLMs profusely. Hooples (talk) 16:55, 12 July 2026 (UTC)

Hi Hooples, thank you, this looks like a large case.

Mcapdevila (talk · contribs), XTools, Possible Jewish origin of Columbus

They've created ~60 articles and ~3000 edits since Nov 2022, and had two creations WP:G15'd (both October 2025). Warned October 2025 after adding hallucinations to which they responded They were partialy LLM generated in the beginning, and although I have checked them.. they needed more checking of the "hallucinated references" you've cited.. They had an article at AfD yesterday in which they socked and got a temp block for that, idk what's going on with . Would appreciate if someone could check for WP:V failures in their recent edits Kowal2701 (talk, contribs) 17:14, 12 July 2026 (UTC)

I asked on their talk regarding the references for that article. ARandomName123 (talk)^{Ping me!} 00:30, 13 July 2026 (UTC)

Even it this wasn't LLM-generated (which it clearly is), it wouldn't be acceptable as an article. We already cover the topic in Origin theories of Christopher Columbus#Crypto-Judaism hypothesis in considerably more detail. AndyTheGrump (talk) 00:38, 13 July 2026 (UTC)

Oh, that is a much better and more informative article. Looking at it's references and the whole of the section, it's fairly obvious it's been trained on information along those line. Then what whatever LLM used just condensed it and hallucinated some information and weirdly translated some reports. Hooples (talk) 01:55, 13 July 2026 (UTC)

Does it look better now? Mcapdevila (talk) 22:09, 29 July 2026 (UTC)

The current version of the article still appears to have quite substantial AI-generated elements, both in terms of text and in terms of references that don't seem to exist. For example, the only google hits for "New DNA study reignites debate on Columbus' Jewish roots" (2021) is simple:Possible Jewish origin of Columbus and es:Posible origen judío de Colón. Andrew Gray (talk) 22:32, 29 July 2026 (UTC)

@Admins willing to patrol AINB: , given the background I think a mainspace block would be appropriate (or just an ordinary indef)? Kowal2701 (talk, contribs) 05:42, 13 July 2026 (UTC)

Given previous issues its probably going to be an indef. AI and socking is burning too much good faith editors time.©Geni (talk) 17:41, 13 July 2026 (UTC)

Cleanup efforts are ongoing and being tracked at /2026-08-16_Levixius, 122 pages need review.

Levixius (talk · contribs · deleted contribs · logs · filter log · block user · block log)

massive potential LLM cleanup case, see here and one where they added utm source chatgpt Their user page says "and use neural language services (primarily GPT-4) to reduce labour while editing on Wikipedia." ~2026-38678-64 (talk) 02:32, 13 July 2026 (UTC)

The editor removed ... and use neural language services (primarily GPT-4) to reduce labour while editing on Wikipedia ..."; ergo, I'm adding link here for historical reference. --Rosiestep (talk) 12:49, 16 July 2026 (UTC)

Levixius (talk · contribs) Kowal2701 (talk, contribs) 05:43, 13 July 2026 (UTC)

Hi, I manually and painstakingly review all my edits, and frankly my use to neural language is rather sparse, and has more to do with readability/copywriting (which, again, I manually review) then writing entire articles. LΞVIXIUS💬 07:48, 13 July 2026 (UTC)

could you please explain this edit where the url parameter utm_source=chatgpt.com parameters was added then? ~2026-38678-64 (talk) 12:06, 13 July 2026 (UTC)

Just want to note the use of AI for source-searching is not in violation of WP:NOLLM. No comment on actual case, however. Ca ^{talk to me!} 15:24, 13 July 2026 (UTC)

@Levixius, i'm also curious, why did you change Kane to a piped link about Kane (wrestler) here? ~2026-38678-64 (talk) 20:00, 13 July 2026 (UTC)

must be a typo lmao, I just Ctrl+K'd it, didn't realize it landed on the wrestler instead of historian P.V. Kane(?) LΞVIXIUS💬 01:56, 14 July 2026 (UTC)

yeah, that's P.V Kane in the source ~2026-38678-64 (talk) 02:05, 14 July 2026 (UTC)

fixed LΞVIXIUS💬 04:11, 14 July 2026 (UTC)

Still a manually verified source. Not quite sure why it matters if there's chatgpt on the URL. LΞVIXIUS💬 01:59, 14 July 2026 (UTC)

manually and painstakingly review my edits

As of this year ,AI-generated text is not allowed on Wikipedia regardless of how much "manual review" is done. Gnomingstuff (talk) 16:24, 13 July 2026 (UTC)

Noted. LΞVIXIUS💬 01:57, 14 July 2026 (UTC)

@Levixius, I've draftified some of your post-2022 creations, please review them for WP:V failures and other issues like WP:NPOV/WP:WTW and WP:COPYVIO. You can see a list of your drafts here. I'm seeing issues in some of the articles you've said you've manually reviewed, like at Special:Permalink/1336666545 Throughout his career, Mahapatra has published extensively in the fields of physical chemistry and chemical kinetics. is cited to a primary source which doesn't even have any prose, making this WP:OR. That's the first source-text I checked, so I'm going to draftify the others as well. Please let us know if you have any questions Kowal2701 (talk, contribs) 19:11, 9 August 2026 (UTC)

So what start date? Jan 2023? I will set up a tracker, but it's a lot of edits and I don't want to catch more than I must. M kuhner (talk) 03:27, 10 August 2026 (UTC)

Having put some of their edits through Pangram, (11 June 2024) is the first that shows AI use, though Pangram does have a problem with false negatives. are two creations from Dec 2023 that showed AI signs but came back negative from Pangram (I'd already draftified both). They also were pageblocked after edit warring over added Dec 2023 which failed WP:V. There aren't really any big edits in between. I'd go with 25 Dec 2023 (obv not including edits after 14 July 2026) Kowal2701 (talk, contribs) 08:18, 10 August 2026 (UTC)

Okay, made tracker 25 Dec 2023-14 July 2026; some later edits will be included if editor also had edits to that article within the time frame (because filtering those out was too much work, sorry). M kuhner (talk) 04:00, 16 August 2026 (UTC)

 Wikiwizardinho (talk · contribs · logs · block log) (13 July 2026)
Editor has been notified and asked to discuss AI/LLM editing concerns. 

Wikiwizardinho (talk · contribs), XTools

See Wikipedia:Administrators' noticeboard/IncidentArchive1198#Use of unreviewed LLM content by User:Wikiwizardinho, they had two articles AfD for being LLM-generated and were warned but didn't respond, and the report got closed for some reason. Since then they've created 10 more articles for a total of 65. Jhalana Amagarh leopard conservation reserve, Rajasthan Forest Department, Menar Wetland, and Folk dances of Rajasthan were all created April–June 2025 and are tagged as AI (ANI thread was 31 August). Last edited 4 July so I've asked them, but would appreciate someone looking at their edits in other months Kowal2701 (talk, contribs) 14:59, 13 July 2026 (UTC)

Overwrote a redirect on Surha Tal Bird Sanctuary in Diff/1331096373 with an LLM-generated article fork instead, contained WP:AIREDCAT, WP:AIBOLD, and references to the main pages of websites. The redirect to the existing article Surha Tal has since been restored, but they're still using LLMs as of last month. ‑‑gurkubondinn 20:35, 13 July 2026 (UTC)

@Admins willing to patrol AINB: I think a mainspace block is necessary here given the ongoing (though infrequent) issues and lack of communication? Kowal2701 (talk, contribs) 20:45, 13 July 2026 (UTC)

This is almost a stale account, with only one non-AI edit in the last six months. ~Anachronist (who / me) (talk) 21:50, 13 July 2026 (UTC)

True, but tbh I'm just worried that such editors who've ignored warnings and even ANI reports and continued to use LLMs, will just come back and create more mess for people to clean up. There's Tbound2, Marcodicaprio, UrusHyby, PeppermintSA, BookOfDisquiet, NiseEdits, Nightsturm, and CoffeeBeanAffair, all of whose more recent edits were preventable, but instead we need to play whackamole where they come back for some edits, and we only notice a few weeks later when they've gone inactive again (rinse and repeat). Imo if someone's ignored a warning/not communicated and continued to use LLMs, a friendly mainspace block where it's only needed for them to commit to not using them in future would be best (regardless of having recently edited), but I get that this contradicts convention re sanctions and I wouldn't be the one liable for making them. Kowal2701 (talk, contribs) 10:36, 14 July 2026 (UTC)

Where are we at with this? Time for a community consensus for cleanup? Do we have any idea of the start date of LLM use? M kuhner (talk) 03:28, 10 August 2026 (UTC)

The auto-archiver on the user talk page was set to a week, so any warnings left have been archived and won't be seen. I just set it to 60 days. I'd say leave another warning and see if there's a response. ~Anachronist (who / me) (talk) 04:31, 10 August 2026 (UTC)

They made no edits between April 2024 and April 2025, their 2024 edits are minor. I'd start at 5 April 2025 (also, support WP:LLMPRV) Kowal2701 (talk, contribs) 08:37, 10 August 2026 (UTC)

Done LLMPRODs Kowal2701 (talk, contribs) 18:27, 15 August 2026 (UTC)

Just to clarify the timeline a little, Jhalana Amagarh leopard conservation reserve was created before the AN/I thread and was featured in it. I think leopards are cute, so I cleaned it up at the time when I saw it at AFD. (And have now thus reverted the AI tag after seeing it on my watchlist -- hope that's okay!) GreenLipstickLesbian💌🧸 18:48, 15 August 2026 (UTC)

All good, sorry I wasn't sure if you'd looked at the sources Kowal2701 (talk, contribs) 19:50, 15 August 2026 (UTC)

Ah, that makes sense -- no worries! And, similarly, no need to worry-- I have many flaws, but unquestioningly trusting AI summaries of sources is, fortunately, not one of them. (Heck, I don't even trust human summaries of sources -- I've done too many article cleanups of my own.) GreenLipstickLesbian💌🧸 20:20, 15 August 2026 (UTC)

Wikiuser829 (talk · contribs), XTools

Yet another large pile of AI edits going back at least a year (most visible in new articles); no warnings until now. Gnomingstuff (talk) 17:26, 13 July 2026 (UTC)

Chandrapur, Chhattisgarh (Dec 2024) had lots of AI signs, I'm unsure if their articles after that were consistently AI Kowal2701 (talk, contribs) 09:29, 10 August 2026 (UTC)

 Cleanup has been requested and is being tracked at /2026-08-10_Gamji231, 373 pages need review.

Gamji231 (talk · contribs) XTools

see this edit, it links a paper named "Preferences for Cash vs. Digital Payments" that has nothing to do with a Harley-Davinson advertisement featuring a hijab, says " in place of the brand's traditional leather-clad rider imagery" when the cited source talks about an ad that shows half the face of one woman in a hijab and half of another in leather, has an overwhelmingly exhaustive editing summary, and this textbook WP:AICOMMENT featuring oaicite

had a warning for using LLMs in february ~2026-38678-64 (talk) 18:38, 13 July 2026 (UTC)

Some signs they gamed ECP with wikilink addition (a ton of those now reverted as useless/harmful). The edit summaries change over time but are never reassuring (they like to say "A small edit" as an edit summary and then add a sizeable chunk of text). I'd be inclined to say LLM suspicions go all the way back to the start (July 2025), but I haven't found conclusive evidence. (Edit summary "Added a new section with credible sources from reputable media outlets." does not make me happy but it's not a smoking gun either.) By November 2025 I'm sure: Added “Miss Universe 2025” section with verified sources and internal links. Expanded coverage to include Nadeen Ayoub’s participation as the first Miss Palestine delegate, international media reporting, and contextual significance. Maintained neutral tone per WP:NPOV and cited multiple reliable sources (The New Arab, Times of Israel, Israel Hayom, AS, Grazia ME, NY Post, The Sun, The Times, The Canary).. Making a tracker, though the wikilink gaming is going to clutter it badly. M kuhner (talk) 03:40, 10 August 2026 (UTC)

There is currently a discussion at Wikipedia:AI noticeboard regarding an AI cleanup investigation which may be related to you.

 BlueTigerBlue00 (talk · contribs · logs · block log) (25 July 2026)
Editor has been warned about the use of AI/LLM content and WP:NOLLM. 

BlueTigerBlue00 (talk · contribs)

LLM, exhaustive summary, here and there ~2026-38678-64 (talk) 21:37, 13 July 2026 (UTC)

The second edit (Diff/1351359381) adds a source with an utm_source=chatgpt.com url parameter. Not necessarily a smoking gun in and of itself, but probably worth looking into. ‑‑gurkubondinn 21:44, 13 July 2026 (UTC)

i was alerted to their edits by mistakenly thinking that they added referrer=grok to the article, it seems that i have to file another AINB for that sockpuppet (eugh) ~2026-38678-64 (talk) 21:47, 13 July 2026 (UTC)

They responded with

Thanks for the note, but thats not correct. I write all content myself, based on reliable sources. The only tool I use sometimes is an AI spellchecker to correct my grammar and typos before posting, because I'm native German but mostly do work on Engl wiki. The text itself is entirely my own words, not LLM generated BlueTigerBlue00 (talk) 00:07, 18 July 2026 (UTC)

Re this edit (23 July), the source doesn't even mention "Seen", nor does it mention commercialisation (also WP:IMDB). @Admins willing to patrol AINB: , I think a mainspace block is needed here Kowal2701 (talk, contribs) 11:02, 24 July 2026 (UTC)

Hi Kowal2701, Seen and the film details are mentioned in the cited IMDb page (cited archive link), so I can't conclude LLM use from that, despite IMDb being an unreliable user-generated source. However, their explanation does not address their addition of a ChatGPT-generated link. — Newslinger talk 17:54, 25 July 2026 (UTC)

Sorry, I need to be checking archived links. Will keep an eye on this then, thanks Kowal2701 (talk, contribs) 20:08, 25 July 2026 (UTC)

All, or nearly all, of their big content edits have an "Edit Check (paste) shown" tag, which means they're copying and pasting from somewhere into Wikipedia. They could be copying from AI, or they could be copying from their spellchecker program. ARandomName123 (talk)^{Ping me!} 01:42, 27 July 2026 (UTC)

Oh great, they removed an LLMPROD saying they'd reviewed it all , the first ref I checked (ref #8 ) doesn't mention several of those guest stars. Kowal2701 (talk, contribs) 14:59, 31 July 2026 (UTC)

Ivanhardybirt (talk · contribs), XTools

Their first LLM article appears to have been on 12 April 2025 (stark difference to their earlier creations, obvious signs), they've created 26 since with some deleted. Was brought to ANI October 2025 (at which time they were warned for LLM use ) which got archived as unresolved. Their most recent creation was 11 July Kowal2701 (talk, contribs) 22:42, 14 July 2026 (UTC)

well, that is upsetting Kowal2701 (talk, contribs) 17:43, 17 July 2026 (UTC)

I've draftifed creations that weren't obviously human-written from 12 April 2025 onwards, but I'm not convinced that's all of them as their use of AI seemed sporadic Kowal2701 (talk, contribs) 21:16, 9 August 2026 (UTC)

noting User talk:Ivanhardybirt#LLM use, nice to see Kowal2701 (talk, contribs) 18:36, 10 August 2026 (UTC)

• Lebu Ayiga (talk · contribs · deleted contribs · logs · filter log · block user · block log)

More promising than most? Several AI articles throughout early 2026, was told repeatedly about the policy. They didn't respond on their talk page directly (they did post this AI-generated comment elsewhere), but they do seem to have taken the guidelines into consideration as their most recent edits, Draft:Justina Wamae and their edits to 2017 Somali drought, do not seem to be AI-generated either in v1 or in the subsequent revisions. (There's also this comment, not sure what to make of it. I didn't come across these via an AI detector and ZeroGPT sucks in particular.)

So the gist is that this applies to the articles created before May 27, 2026. Gnomingstuff (talk) 22:44, 14 July 2026 (UTC)

Sorry for not responding on the talk page but I have taken the guidelines into consideration and will improve on my edits Lebu Ayiga (talk) 11:35, 16 July 2026 (UTC)

I understand that AI generated content is not allowed on wikipedia at all and going forward I am going to follow the WP:LLMS guideline on working to improve and creating articles Lebu Ayiga (talk) 11:54, 16 July 2026 (UTC)

Thanks for responding. Just to clarify, this section is mostly just for backlog purposes. Gnomingstuff (talk) 18:26, 16 July 2026 (UTC)

(noting User talk:Kowal2701#Drafting LLM articles) Kowal2701 (talk, contribs) 08:56, 10 August 2026 (UTC)

 The Royal Herald (talk · contribs · logs · block log) 
Editor has been blocked site-wide for AI/LLM policy violations. 

 Cleanup has been requested and is being tracked at /2026-07-15 The Royal Herald, 0 pages need review.

The Royal Herald (talk · contribs · deleted contribs · logs · filter log · block user · block log)

See case at WP:ANI § User:The Royal Herald. Editor has been indeffed. In the closing statement voorts said "I suggest cleanup discussion continue at AIN." The user has been inserting extraneous material into a LOT of articles, and there are AISIGNS especially in their edit summaries; their responses during the case were also painfully clear-cut LLM. I ask for a cleanup consensus; this kind of insertion LLM material is painful to clean up if it sits around. M kuhner (talk) 05:27, 15 July 2026 (UTC)

examples: ~2026-38678-64 (talk) 05:31, 15 July 2026 (UTC)

Tracking list template can't display: too many edits, I suppose. Xtools may be a better initial choice. M kuhner (talk) 05:53, 15 July 2026 (UTC)

600KB? dear god

good news: the page is under the category Category:Pages where post-expand include size is exceeded ~2026-38678-64 (talk) 06:18, 15 July 2026 (UTC)

If we can find a more recent start date for LLM use we can cut it down. M kuhner (talk) 06:20, 15 July 2026 (UTC)

Trying to fix this right now. sapphaline (talk) 09:42, 15 July 2026 (UTC)

Fixed. sapphaline (talk) 09:50, 15 July 2026 (UTC)

Perhaps better for elsewhere, but still think its worth flagging. Believe this may have been a Sock Puppet from a previously blocked user, User/Geronimo Virula Medrano El: Special:Contributions/Geronimo Virula Medrano El. Found while reviewing some edit history on the Medrano page . Edits match the same pattern as the user as far as I can tell, and same subject matter. ~2026-39818-62 (talk) 06:38, 15 July 2026 (UTC)

Good catch. We should keep an eye out during cleanup. I think Geronimo is way too stale for the checkusers to be interested, but there may be something fresher. Geronimo was blocked pre-LLM so at least we don't have to vet their edits. M kuhner (talk) 07:47, 15 July 2026 (UTC)

SPI confirmed by suspicions, for better and for worse Wikipedia:Sockpuppet investigations/Geronimo Virula Medrano El. Sorry for such a large case being on the table. On the one hand I'm glad to have found and flagged so it can be fixed, on the other it looks like a mess to remediate. Gwyain1 (talk) 20:13, 15 July 2026 (UTC)

yeah good luck with this one, I knew the name sounded familiar and it's because I was on the receiving end of the typical sealioning and stealth removing tags Gnomingstuff (talk) 07:53, 15 July 2026 (UTC)

Does their LLM use go back to their early edits? (Sep 2023) was created as a GA (presumably they asked the chatbot to write a "good article" lol), but I'm not seeing signs other than WP:AITITLECASE, and there are hyphens in the date ranges which is more of a human thing. (Nov 2023) has WP:AIDASH and WP:AIVOCAB like "reflects" and "recognition of their importance in shaping human existence", throw in title case and a bolded list as well. Kowal2701 (talk, contribs) 09:18, 15 July 2026 (UTC)

not sure, sorry, haven't gone through this thoroughly because there is so much

would sure be nice if we ever got a straight answer! Gnomingstuff (talk) 09:27, 15 July 2026 (UTC)

Not sure if they're AI or not, but their early edits are entirely unsourced: . Pattern seems to be, write an article, then slow add sources to it, the sources themselves may be part of what's AI generated. Gwyain1 (talk) 15:25, 15 July 2026 (UTC)

It looks like a mix of what the LLM gave them and lack of interest in learning formatting. Things like "Idem id., cajon 12, num 59" and 29-page PDFs with no page number. Best to WP:TNT as part of the WP:LLMPRV process rather than have it in newcomer tasks with a maintenance tag. Some future history buff can make a stub. Cedarshadow (talk) 20:53, 15 July 2026 (UTC)

I'm gonna LLMPROD from Sep 2023 onwards, though they'll all probably be covered by WP:G5 given the SPI Kowal2701 (talk, contribs) 21:54, 15 July 2026 (UTC)

I do agree with WP:LLMPRV. InfernoHues (talk) 16:37, 15 July 2026 (UTC)

I've done low-hanging fruit down to Pedro de Mena in the tracking page (about halfway down) Kowal2701 (talk, contribs) 21:31, 3 August 2026 (UTC)

now down to Favourite, it's getting fiddly though Kowal2701 (talk, contribs) 23:39, 3 August 2026 (UTC)

Never seen an article with 4400 edits by the same editor before; no wonder it broke the tracker! M kuhner (talk) 03:48, 19 July 2026 (UTC)

Just deleting that page took several minutes, and led to me thinking it failed to delete. That was interesting. ASUKITE 12:31, 22 July 2026 (UTC)

I have removed the 4400 edits from that article's entry, in the hopes of getting the tracking page to behave properly (as is common with too-long trackers, the update buttons aren't always working). We'll see if it helps. M kuhner (talk) 01:42, 4 August 2026 (UTC)

Cleanup efforts are ongoing and being tracked at /2026-08-07_Sila_Handsome, 2 pages need review.

Sila Handsome (talk · contribs · deleted contribs · logs · filter log · block user · block log)

Sila Handsome (talk · contribs) registered today. Large edits to military/firearm articles. I reverted two of their article changes: Iron sights and Madsen machine gun as not only clearly AI produced but also lacking references, as their edits replaced properly referenced text with unreferenced text. I haven't been actively patrolling AI issues recently so I'm not sure as to how much of the rest of their contributions are problematic, but certainly deserves scrutiny. — rsjaffe 🗣️ 18:57, 15 July 2026 (UTC)

There's huge edits (often multiple kB) being made within minutes or seconds of each other. ‑‑gurkubondinn 19:09, 15 July 2026 (UTC)

I understand the concern about the size and speed of my edits. I will slow down my editing process and make smaller, more focused changes so they can be reviewed more easily by other editors. Sila Purachato (talk) 15:11, 25 July 2026 (UTC)

@Rsjaffe This looks like block evasion and evidently they still don't understand the reason for their block. Jfire (talk) 15:14, 25 July 2026 (UTC)

No good reason to have that second account so I blocked it. — rsjaffe 🗣️ 15:19, 25 July 2026 (UTC)

also markdown in Special:Diff/1364288022 Gnomingstuff (talk) 20:45, 15 July 2026 (UTC)

And now we have pure vandalism coming from the same user. This itself is blockable, in my opinion. Somepinkdude (talk | contribs), in solidarity 23:44, 15 July 2026 (UTC)

They are continuing to make large unsourced, likely LLM-generated expansions to articles. @Rsjaffe seems like a block is warranted. Jfire (talk) 02:55, 18 July 2026 (UTC)

Blocked from articles and invited to discuss here. — rsjaffe 🗣️ 03:50, 18 July 2026 (UTC)

Thank you for bringing these concerns to my attention. I understand that some of my edits were too large and lacked sufficient references. I will review my contributions, restore referenced information where appropriate, and make sure future edits are supported by reliable sources. Sila Handsome (talk) 05:07, 18 July 2026 (UTC)

It is not just that they were too large and lacked references; it's that they were generated by a large language model, which is forbidden on Wikipedia. Many of the LLM generated claims were false. Please commit to not using LLMs to create text for Wikipedia. The relevant guideline is WP:NOLLM. M kuhner (talk) 05:12, 18 July 2026 (UTC)

Girandoni air rifle has source to text issues, making a claim about using the weapon in "demonstrations" for Native tribes when the source knows nothing of this (and only speculates that it was a Girandoni at all). M kuhner (talk) 05:41, 16 July 2026 (UTC)

Making a tracker for master account, as there doesn't seem to be an unblock request in the works. No mainspace edits on the sock. M kuhner (talk) 22:01, 7 August 2026 (UTC)

Teknuccio (talk · contribs · deleted contribs · logs · filter log · block user · block log)

Contributor with English as a second langauge. Was asked about AI earlier this year on User talk:Teknuccio#LLM_use, and said I have used AI-based tools quite often as auxiliary tools for English wording, copyediting, structure, and checking neutral phrasing, and apologied for this going "beyond the guideline."

On the one hand, this is more cooperative than most of the responses we get (see above) ...but it doesn't really seem like they've stopped, based on their very rapid editing a few days ago, each rapid edit paired with an AI edit summary. A bit hard to tell from what's here how well this conforms to WP:LLMTRANSLATE; the text doesn't seem like obvious unreviewed AI text but unclear how much verification is being done in 5 minutes per edit. (talk) 05:26, 16 July 2026 (UTC)

Thank you for the note. I understand the concern. The material I added was based on local and territorial sources that I checked myself, but I see that the speed of some edits may make the verification unclear. Teknuccio (talk) 15:26, 16 July 2026 (UTC)

Did you use an LLM, like an AI chatbot, for your edits? ‑‑gurkubondinn 16:30, 16 July 2026 (UTC)

Yes. I used AI/LLM tools mainly to translate and organize my own Italian research notes and locally sourced material into English prose. I did not use them as sources; the factual material and references were selected and checked by me. Teknuccio (talk) 18:29, 20 July 2026 (UTC)

I think I need another eye on this user to determine if they have a history of AI edits to clean up. I spotted the LLM use at this diff which added this amusingly vague, almost information-free section: . However, many of the user's other contributions, despite having obviously AI-generated edit summaries, introduce typos and ungrammatical punctuation usage (see where the edit summary ironically states that it corrected spelling errors and artist name inconsistencies while introducing no less than 5 misspellings of the subject's name). I don't know what to make of the distinction between the AI summaries and the, let's say, organically spelled content, and the potential mixing of human and LLM content within a single edit (see which edit summary boasts neutral, concise, and source-based Wikipedia-style wording and adds a recognizably AI-generated Awards & Recognition section but also adds a Television appearances section bizarrely written in the first person).

Anyway, I warned the user for the LLM usage in the first diff and tagged the article (looks like the whole thing was already AI-generated before they got there) ^purring_maggot (talk) 02:08, 17 July 2026 (UTC)

I don't know what to make of stuff like led to several more appearances that involved a close interaction with the host, that is sourced to an episode of Super Dave on Apple TV..? ‑‑gurkubondinn 12:32, 17 July 2026 (UTC)

* 1990:The Arsenio Hall Show is already mentioned in the entry, but I did not perform routines. I performed only one routine: my signature Ballon-man act. sounds like some "ask Character AI if the article is accurate" workflow, maybe? I don't know how you'd even get this otherwise, outside some kind of weird larping or "Elijah J" actually being Blumenfeld but not a SPA Gnomingstuff (talk) 13:48, 17 July 2026 (UTC)

User has now denied LLM use. Personally I find this unconvincing and I've said so; we'll see how this develops. ^purring_maggot (talk) 15:41, 17 July 2026 (UTC)

Now stating LLM was only used for "formatting" and a not-quite-promise to "be very careful with this next time": (the conversation got split up between my talk and theirs) I think it might also be worthwhile to keep an eye on this user Kappalotti who originally made the AI-generated Dwight Cuthbert Revire Gardiner page and has come out of inactivity as of yesterday. Whatever's going on here, I'm at least definitely going to stubify that page whenever I have the time. I'm supposed to be at work! ^purring_maggot (talk) 16:18, 17 July 2026 (UTC)

 Cleanup has been requested.

 Waltereg (talk · contribs · logs · block log) (16 August 2026)
Editor has been blocked from editing the mainspace (articles only) for AI/LLM violations. 

• Waltereg (talk · contribs · deleted contribs · logs · filter log · block user · block log)

Extremely rapid editing, turning redirects into LLM-generated articles. The metaphorical smoking gun is in Diff/1364752808 (one of them is even somehow smoking in three different ways).

All of their edits are contained in one subject area. The user's talk page has a litany of DAB notices, draftification notices, complaints about poor or nonexistent sourcing, and various deletion notices, about half of which even predate the public release of ChatGPT. But there is a lot to clean up. ‑‑gurkubondinn 11:04, 18 July 2026 (UTC)

For examples of large and extremely rapid edits:

• History/2025–26 Real Madrid Baloncesto season
• History/2024–25 Real Madrid Baloncesto season

‑‑gurkubondinn 11:21, 18 July 2026 (UTC)

Then there are others that I'm not entirely convinced about, for example Diff/1361974319 is a 6,4 kB edit made less than 4 minutes after the previous edit to 2026–27 CSM București women's handball season, but it seems to mostly be an empty boilerplate table. I've draftified some other articles in the same 'series' but leaving this one alone for now. ‑‑gurkubondinn 11:36, 18 July 2026 (UTC)

Sports articles are the worst. There's so much table and so little text that it's incredibly hard to know how they were written. M kuhner (talk) 18:35, 18 July 2026 (UTC)

(18 July) appears to have chatbot communication in it, <!-- The regular-season draw is scheduled for 22 July 2026. Add the fixtures after the official schedule is published. --> Kowal2701 (talk, contribs) 19:00, 18 July 2026 (UTC)

@Waltereg, can you please respond here Kowal2701 (talk, contribs) 18:54, 28 July 2026 (UTC)

Ok, What can I help you with? Waltereg (talk) 21:27, 28 July 2026 (UTC)

Did you use LLM to write some of these articles? If so, which ones? We'd hate to delete or stubify innocent articles by mistake. M kuhner (talk) 04:50, 7 August 2026 (UTC)

Waltereg: can you answer the question? ‑‑gurkubondinn 02:35, 16 August 2026 (UTC)

How can I help you? What do I need to do? Waltereg (talk) 03:51, 16 August 2026 (UTC)

Please let us know which of your articles were created with LLM assistance. If you can't do that, at least let us know approximately what date you started using LLM, and what date you stopped. (And please do stop. Otherwise you are likely to be blocked.)

A list of all articles you have created is here: . M kuhner (talk) 04:22, 16 August 2026 (UTC)

What is an LLM? Waltereg (talk) 05:07, 16 August 2026 (UTC)

Large language model, also called AI or Chatbot. Programs like ChatGPT, Copilot, Gemini, Claude, or Perplexity. Also some other programs have it built in, such as recent versions of Grammarly or DeepL. Anything that writes text as if it was a human being, but is not. M kuhner (talk) 05:35, 16 August 2026 (UTC)

They have unambiguously used ChatGPT: Diff/1364752808.

Waltereg: please answer the question without deflecting. If you continue editing without giving a satisfactory answer, I will ask admins to block you from editing until you do. ‑‑gurkubondinn 10:50, 16 August 2026 (UTC)

@Admins willing to patrol AINB: i think that communication may need to be enforced. Waltereg has continued editing, making large and rapid edits. ‑‑gurkubondinn 12:07, 16 August 2026 (UTC)

Please forgive me for upsetting you; what would you like me to do so that you forgive me? I realize I was wrong in what I did. Waltereg (talk) 14:13, 16 August 2026 (UTC)

Nobody here is upset. Can you explain why you continued if you were aware that you shouldn't do it? ‑‑gurkubondinn 14:19, 16 August 2026 (UTC)

Blocked Waltereg from article space for LLM misuse. — Newslinger talk 13:13, 16 August 2026 (UTC)

• MissBVCC (talk · contribs · deleted contribs · logs · filter log · block user · block log)

This user was warned for using LLMs on July 2.

I am worried about their participation in AfD discussions; these are some examples of what I consider suspicious: Generative AI pornography, Western Iran clashes (2026–present), 2024 Dhing gang rape case, Alternate ending, Indian influence in Bangladesh, Save the Children Jalalabad attack, Meaning-making (2nd nomination). jhvx talk contribs 18:37, 18 July 2026 (UTC)

The AfD comments are strange: a mix of LLM and human, I guess, given that almost every one has an error of some kind.

The mainspace edit I spotchecked, however, was pure LLM. I'll mark for needing a cleanup case. M kuhner (talk) 18:54, 18 July 2026 (UTC)

Generative AI pornography

This page has one weird edit filter log. ‑‑gurkubondinn 19:46, 18 July 2026 (UTC)

This is the userpage of the person who created the original (and AISIGNS-full) Generative AI pornography article, you truly cannot make this shit up: Hi, Bobby here. I am an Artificial Intelligence enthusiast. I like what's going on with AI there in the world. I like creating stuff with AI. Gnomingstuff (talk) 00:56, 19 July 2026 (UTC)

At the moment I don't think the AfD comments are LLM, within your examples there's some human variance in there such as in the Generative AI pornography comment: it feels like a high-traffic encyclopedic topic which should be deleted instead of saying "it should not be deleted" which is what the user clearly meant to say as they are voting keep. I notice they invoke a lot of policy pages, which is typical at AfD; some signs of LLM use with regards to policies would be invoking fake policies, invoking real policies but quoting statements that don't exist on the page, or invoking policies to argue something that the policies really don't say. No comment on mainspace as I have not looked at their edits there. ^purring_maggot (talk) 20:12, 18 July 2026 (UTC)

Lots of errors like that, around one per comment. Surely not pure LLM.

There's always a risk that a newcomer is copying what they see in other AfD cases, and enough of those are LLM that they may sound "off." I kind of think there's some use here, but the mainspace will be the real test. M kuhner (talk) 20:54, 18 July 2026 (UTC)

it seems like a topic prone to WP:IDONTLIKEIT/WP:ILIKEIT reasoning which may be causing comments to seem off? Gnomingstuff (talk) 16:36, 19 July 2026 (UTC)

Still on the fence over the AfD comments, but this seems like a smoking gun to me, given that the edit was made in 3 minutes. Somepinkdude (talk | contribs), in solidarity 18:03, 19 July 2026 (UTC)

Am I going crazy, or am I seeing AI signs at Football manufacturing in Pakistan? I'm seeing some small AI signs across this user's contribs like "X reported on Y", large article creations in one go, verbose headings etc, but no smoking guns. See original football article creation: 1. + some more article creations: 2, 3. Also note a copyvio warning in May- User talk:Umais Bin Sajjad/Archive 4#MAY-17-2026 -and a non-template LLM warning yesterday- User talk:Umais Bin Sajjad#LLM use. Sarsenet•he/they•(talk) 07:30, 19 July 2026 (UTC)

Nah, that v1 is obviously AI. Fermiboson (talk) 07:39, 19 July 2026 (UTC)

First paragraph doesn't quite line up with the sources in a very llm way, and that's with the Khaleej Times article feeling a bit LLMy as well. CMD (talk) 09:01, 19 July 2026 (UTC)

I have tagged as AI. Source 10, OUP/World Bank used an aggregator rather than the source OUP host. The source is fuzzily transferred to the article text, e.g. it doesn't go into the SME side. Interesting source, but summarised in the best traditions of AI. Beyond that multiple other AI signs. ChrysGalley (talk) ChrysGalley (talk) 09:27, 19 July 2026 (UTC)

I've used LLM assistance during drafting and then manually revised the prose afterwards. As WP:LLM does not permit broader generation or rewriting of article prose, I follow that restriction carefully. Regarding the large first revisions, I usually prepare substantial drafts in my userspace before moving them. I take responsibility for my edits and am reviewing the affected articles sentence by sentence against their cited sources. I will remove or rewrite anything unsupported, too closely paraphrased, or otherwise noncompliant. Football manufacturing in Pakistan was subsequently reviewed at DYK. I mention this only as evidence of the article's present condition, not as proof of how it was drafted or as an exemption from LLM. If there are specific diffs or source–text integrity concerns, point them out and I will address them. Umais🗣 09:55, 19 July 2026 (UTC)

Thanks for the candid and helpful reply @Umais Bin Sajjad, though I have to say that reply also looks AI assisted. Nevertheless, under WP:NEWLLM there are only 2 ways you can use LLM (translation, copyedits) and that does not apply here - you simply cannot use LLM to assist drafting, so I can't see how you followed that restriction carefully. Due to the DYK status I don't feel I can G15 this article (though I imagine it is a legitimate next step, given that at least some sources were not checked against the wording). The If there are specific diffs or source–text integrity concerns is standard wording we often see off AI tools when they, the tools, are challenged about AI - in reality as a human you would know where you used AI, so you are the best person to answer that point. In my view? The article needs to be rewritten, based on manual examination of the sources, and that you should resist the temptation to tinker with the existing article. ChrysGalley (talk) 10:08, 19 July 2026 (UTC)

English is not my first language, which is why I used LLM assistance while drafting. Nevertheless I checked the article manually against the sources and revised prose with MOS in mind. Now that this concern has been raised, I'll draft directly from the sources myself and use tool within the permitted WP:LLM. Umais🗣 10:33, 19 July 2026 (UTC)

Comparing the two responses from the user in this thread, my guess is that one of the replies was generated by an LLM and not the other one. ‑‑gurkubondinn 10:44, 19 July 2026 (UTC)

How did you review the draft sentence by sentence for source 2, which is an 11 minute video? CMD (talk) 10:32, 19 July 2026 (UTC)

I checked accompanying written summary and relevant opening portion of video, where it states that Sialkot makes about 70% of the world's footballs. The citation was intended to support that estimate. I will move them closer to the claims they support and remove anything that is not directly verified. Umais🗣 10:44, 19 July 2026 (UTC)

The article has been pulled from DYK: Diff/1364927197. ‑‑gurkubondinn 10:58, 19 July 2026 (UTC)

• Draftify - I think we can see what has gone on here, and the submitter has owned up to the issue and explained why, to their credit. Rather than further hole digging, I would suggest this article is now sent off to Draft status, which will then provide a good checkpoint if/when it is OK for mainspace. Other options being G15, LLMPROD, PROD, AFD. ChrysGalley (talk) 10:53, 19 July 2026 (UTC)
• Draftify is fine with me, but I won't oppose any other outcomes. Thank you to @Umais Bin Sajjad for being honest and straight forward about this, the second rule of holes must be "don't dig holes for someone else" or something like that. ‑‑gurkubondinn 11:02, 19 July 2026 (UTC)

  In order to avoid digging holes, and noting Umais Bin Sajjad stated above that they want to draft directly from the sources, a custom draftify pulling the sources and images but not the prose might be a more efficient way to move forward than bringing together the need to not only draft but also comb through text and videos. CMD (talk) 11:52, 19 July 2026 (UTC)

I draftified it and TNTed the text.--Launchballer 12:19, 19 July 2026 (UTC)

Forward Sports was expanded at the same time. The title of the Al Jazeera source as "The football" is interesting, some automated process (llm or cite generator?) pulled the second h1 tag for the title, dropping the first half ("Story of"). CMD (talk) 11:49, 19 July 2026 (UTC)

I ran GPTZero on a random portion from what appears to be the last edit made as part of the (far more piecemeal) expansion and it came back as likely human written.--Launchballer 12:19, 19 July 2026 (UTC)

The statement Forward Sports operates within the wider Forward Group. The group's related businesses include Forward Gear, a Sialkot-based bag manufacturer established in 2007 is sourced to reference nr. 15 in that revision,which just points to the www.fgear.pk landing page. Not really definitive per se, but pretty typical citing by an LLM. I hope that the user can just improve these articles without using AI tools, as they seem genuinely interested in contributing to the project. ‑‑gurkubondinn 14:03, 19 July 2026 (UTC)

EVO 2026 has a few AISIGNS like having 3 places where there's AIATTR, and AITABLE ~2026-38678-64 (talk) 23:20, 19 July 2026 (UTC)

I just queued Nusuk, whose v1 also came up 100% on GPTZero. Extra eyes appreciated.--Launchballer 10:19, 20 July 2026 (UTC)

@Launchballer (note:WP:AISIGNS is the thing which most of us at AINB use to check for signs of LLM generated drafts)~2026-38678-64 (talk) 18:19, 20 July 2026 (UTC)

so far, i'm seeing AIWEASEL, SUPERFICIAL, RO3, AITREND ~2026-38678-64 (talk) 18:40, 20 July 2026 (UTC)

 Cleanup has been requested and is being tracked at /2026-07-20 Flyingphoenixchips, 517 pages need review.

 Flyingphoenixchips (talk · contribs · logs · block log) (1 August 2026)
Editor has been blocked from editing the mainspace (articles only) for AI/LLM violations. 

Flyingphoenixchips has been submitting LLM-generated content since their first article creation in 2024, the now-deleted Wesea (Western SouthEast Asia) page. In 2025, they added ChatGPT-generated citations to a number of articles: When Blackbirds Fly, Hannah Lalhlanpuii, Shania Yan (now deleted), Ron Whittaker (musician), Mokokchung Times (now deleted), Prajal Regmi (now deleted), and Anthony Doke. Flyingphoenixchips also used ChatGPT to generate comments in deletion discussions: Arunachal Times and Wesean Times.

Their LLM misuse continued after the WP:LLM guideline was implemented. In March 2026, Flyingphoenixchips created the article Melt Fracture with a number of red links in the See also section, as well as nonexistent categories. It was draftified, and the second time the draft was declined (reviewed version), the reviewer stated: "I randomly selected five references to review, and ALL FIVE either did not exist or linked to papers with completely different titles and topics. Together with the formatting, it's pretty obvious this article was written by an LLM and posted by someone who did not care enough to verify anything."

On 30 March, another editor told Flyingphoenixchips that "articles recently created by you appear AI-generated" and asked, "Did you use AI to create the articles?". Flyingphoenixchips responded with a partial denial: "I used a script to organize bare references, yes. did not use for writing".

A new page patroller flagged four of Flyingphoenixchips's article creations as AI-generated on 9 April: 2026 Manipur bombing, Child exploitation in Arunachal Pradesh, Christian persecution in Manipur, and Rpa Ralte. In the ensuing discussion, Flyingphoenixchips denied using an LLM to write the articles, but claimed that they wrote the articles in Markdown, used "two agents to format the paper and organise the references", used "a python script to organise it all and push edits onto my wikipedia account", and then checked the articles themselves. Flyingphoenixchips was told, "Using a Python script for editing isn't allowed if you haven't gotten approval from the WP:Bot Approvals Group", and responded, "I'll try doing things myself from now on and will try to get it approved."

Without obtaining approval for an LLM-based bot or article creation script, Flyingphoenixchips mass-created 46 articles about Yushania species in 84 minutes on 1 July and 7 July. Another editor pointed out issues with the articles, including "Many of the references you have provided lack accessible URLs with which to ascertain their accuracy" and "In a number of articles, references have been duplicated, and some make no mention of the article subject." Flyingphoenixchips responded that they were "planning to revisit those pages again", but have not done so.

The timing of Flyingphoenixchips's recent article edits shows a pace consistent with submitting LLM-generated edits. For example:

• Traditional beliefs and mythology of the Tani people: Gradually added 18,832 bytes in 10 minutes, including ISBN errors, then removed nonexistent categories
• Arunachal church burnings, 1968–1978: Gradually added 17,048 bytes in 9 minutes, starting with the infobox before the rest of the article, then removed nonexistent categories
• Fetishization of Northeast Indians: Created the 35,230-byte initial revision, then deleted an LLM-generated list of sources in the same minute. Heavily rewrote the article in the next 26 minutes, reducing it to 18,029 bytes. After another editor blanked and redirected the article, Flyingphoenixchips gradually added 8,828 bytes of content in 5 minutes before deleting nonexistent categories.
• Racism against Northeast Indians: Created a 39,435-byte initial revision, which was promptly nominated for speedy deletion per WP:G15 because "Some ISBNs and DOIs go to different subjects than the reference's title". Flyingphoenixchips attempted to remove the speedy deletion tag in violation of WP:CSDCONTEST, but it was reinstated. Flyingphoenixchips then edited continuously for about an hour with a stream of small edits overlapping with three large edits (+4,146; +1,076; and +27,697 bytes) that can only plausibly be explained as LLM-generated, because there is no way those large edits could have been written by hand when Flyingphoenixchips was busy making the numerous small edits. The article was culled instead of speedily deleted.

Per the above evidence, I have indefinitely Blocked Flyingphoenixchips sitewide for repeatedly misusing LLMs, denying the full extent of their LLM misuse when asked, and mass-creating articles with a script in violation of the bot policy. An unblock request would be more likely to succeed if Flyingphoenixchips acknowledges that their LLM misuse was inappropriate, agrees to refrain from using LLMs for Wikipedia editing in any capacity as an unblock condition, and agrees to clean up all of their LLM-affected edits.

The WP:LLMPROD process would be the most efficient way to address the LLM-generated article creations. — Newslinger talk 02:47, 20 July 2026 (UTC)

• This is incredibly disheartening since they were NPP and AFC. But I can't say I am surprised. He was tackled about this by @Athanelar and I in Wikipedia talk:WikiProject Articles for creation/Archive 63#Maybe Flyingphoenixchips shouldn't be an AfC reviewer. where he said that he used an AI tool, Wispr Flow, to speed through a lot of AFC reviews, with decidedly mixed outcomes. He undertook not to use that tool again on 26 May 2026, but that was in an AFC context. Separately I think we need to think about the sticks and carrots when it comes to long-serving editors who establish their bona fides via the traditional route and then get subverted by AI's deathly embrace. This won't be the only case. Probably out of scope for this noticeboard though. ChrysGalley (talk) 07:20, 20 July 2026 (UTC)

  I'm kind of kicking myself for not looking into their contributions after the fiasco with their LLM-powered AfC reviews. Glad to see it's solved now. Athanelar (talk) 07:29, 20 July 2026 (UTC)

Also note that the box on top of their userpage is clearly AI-generated too. ChatGPT still loves those curved corners and emoji headers! Sarsenet•he/they•(talk) 08:21, 20 July 2026 (UTC)

Greetings!

Retired user here just giving my two cents on this issue. While I do agree with your block of Flyingphoenixchips, I think it's too far to block him for an indefinite period of time.

While he made a mistake, his contributions to Wikipedia are great, he's a very good Wikipedian who has a lot of knowledge in the subject he edits. I think it's best that the community gives him a firm warning to not use AI, and let him go with the warning in mind.

Articles such as Angelte Zai Ri were not written by AI buy instead by him. He's from the state of Mizoram and used the sources in the Mizo language.

Aside from AI, his edits were great. He deserved a firm warning and a temporarily block for his use of AI, aside from that, he should be let go.

Thanks, ~ AnonLionCavalier83 (talk) 17:25, 21 July 2026 (UTC)

while they meant well in trying to increase coverage of articles, they unfortunately continued to use AI even when it was mentioned by other editors. EM (talk) 19:44, 21 July 2026 (UTC)

Alright makes sense. I think he deserves a firm warning with an added temporary ban on this issue.

After that, let him continue editing as long as he's not using AI. I suggest this because the Wikipedian in question is pretty good at contributing within their respective topics so banning him would be bad at Wikipedia articles about these topics. ~ AnonLionCavalier83 (talk) 01:57, 22 July 2026 (UTC)

"Indef" is not forever: the editor can appeal the block at any time. I have no authority here, but personally would much rather see a human-written unblock request and commitment to do better, rather than allowing the editor to wait out a timed block and perhaps start right back in. M kuhner (talk) 21:04, 21 July 2026 (UTC)

thanks for the info. ~ AnonLionCavalier83 (talk) 01:58, 22 July 2026 (UTC)

I wanted to know the policy around AI-generated images. I am referring here to the image created for Template:The Hornbill Barnstar which now seems obvious to have been created using AI. Astra Travasso (talk) 14:09, 22 July 2026 (UTC)

The policy is at WP:AIGI. It only forbids their use in mainspace, so I don't believe a barnstar, meant for user pages, is actionable. M kuhner (talk) 14:19, 22 July 2026 (UTC)

Tracker created, LLMPRODs started (I started at the top, if someone else is going to join in they might start at the bottom). M kuhner (talk) 04:03, 20 July 2026 (UTC)

I've done LLMPRODs from #195 (Makum people) to #100 (Lee Hee-wan) on XTools, they've fortunately got no drafts to tag Kowal2701 (talk, contribs) 06:52, 20 July 2026 (UTC)

• They've said @Kowal2701 Hey, wanted to ping you here since I am unable to participate in the discussion? Would it perhaps be possible to userify or draftify the articles tagged for prod? I am willing to rework them (before touching anything else) and go through the AfD process. Flyingphoenixchips (talk) 08:55, 20 July 2026 (UTC), I'll leave that to admins to decide, but if we are to userfy them then we can't track {{AI-generated}} removals with the filter Kowal2701 (talk, contribs) 08:59, 20 July 2026 (UTC)

  (I'm fine with draftifying and tagging FWIW) Kowal2701 (talk, contribs) 09:07, 20 July 2026 (UTC)

Same old story: Rapid-fire article creations -- for instance, this article was created only 11 minutes after its predecessor, and this one apparently took 5 minutes -- with clear WP:AISIGNS, across the board. Just to take one example, of which there are many more:

The government and ZANU–PF presented the Amendment as a necessary refinement of the 2013 Constitution to remove practical difficulties and improve governance efficiency. The Herald, Zimbabwe's main state-owned daily newspaper, reported positively on the Bill's progress. It highlighted that the Senate passed the Amendment on 5 May 2021 by 65 votes to 10, noting that some MDC-T senators joined ZANU–PF and traditional leaders in support. State media generally framed the changes as progressive reforms that would enhance stability and align constitutional provisions with practical requirements.

So we've got AIVOCAB all over the damn place right off the bat. The same patterns, locking into place. Did the newspaper really highlight this uncontroversial fact? Good luck finding out, because the link is a 404. (The last sentence, meanwhile, is not cited at all.)

In independent outlets such as ZimLive, commentators described the Amendment as a significant step backwards that concentrated excessive power in the presidency.

Textbook AIATTR, textbook AIWEASEL (no other outlets are cited or mentioned).

Public reaction, as reported in independent media, was predominantly negative among urban populations and opposition supporters. Petitions against the Amendment circulated in May 2021. In Parliament, while the Bill passed comfortably, some opposition MPs voiced strong objections during debates, particularly regarding the implications for judicial independence and democratic accountability.

More AIATTR, cited to the same (one) source as the previous paragraph, except this time very little of this seems to be actually verified by the link. Petitions, for instance, are not mentioned anywhere in the article. Gnomingstuff (talk) 03:00, 21 July 2026 (UTC)

I resent and reject the premise of the accusations. I’ve been working on articles off Wikipedia in Word and bringing them across when I’ve been done with them. At the moment I’m working on a spreadsheet in Excel on the Constituencies of Zimbabwe. I thought it would be a better system than having lots of edits constantly changing the articles. Perhaps I don’t understand this accusation of Weasel. For years Zimbabwe-related articles have been lambasted for lacking neutrality. I try to bring in neutrality and then you accuse me of not working on the articles. Moreover, Zimbabwe-related articles have faced severe (and arguably racially-motivated) criticisms for the sourcing issues because much of what is needed to be documented is not available online and is only accessible in print or to those here in the country. Much is documented across social media, but those links are not reliable in line with WP policies because they are user generated. Therefore, the number of sources available is limited compared to our European and American counterparts, especially when we suffer such press suppression and censorship as we do. The Herald article that the user above mentions has been taken down or changed on their website since I accessed it, but you can see it existed here. My consistent issue with this - and with this user and others - is that rather than flag issues personally to someone and say “Hey - can you work on this? It’s a bit messy.” they simply revert to full-scale deletions. The one article he reverted put the table of “current ministers” to people who were in office in 2009 - some of whom are now dead. Whose encyclopaedic efforts is such a reversion serving? Let us work together, let us be constructive, not destructive. The question at hand is: is what is written true? Is it verifiable? Is it improving Wikipedia. If you find my wording clunky, then let’s improve it together. If you think the statements are “weasel” then let’s improve it against the context that many Zimbabwean articles are flagged across Wikipedia for lacking neutrality! Let us collaborate with one another and not attack one another. I don’t believe anyone here is spending all their time and effort to simply cause problems. Mangwanani ^(talk) 08:19, 21 July 2026 (UTC)

Are you using Microsoft Word's built-in revision features? Gnomingstuff (talk) 09:37, 21 July 2026 (UTC)

I certainly do make sure my spelling and grammar are correct. When you're working on long articles alone for a long time, it is not always possible to see the wood from the trees and mistakes slip through. I note that User Christhespeller has been through articles to fix curly apostrophes, for example, throughout the articles. Mangwanani ^(talk) 10:02, 21 July 2026 (UTC)

@Mangwanani - spelling and grammar mistakes are not problematic in Wikipedia, someone will tidy that up in due course. LLM usage is problematic, as you will be aware. So for me the bigger question is "are you using AI / LLM or related software to write, assist, cite, and/or phrase your drafts?" Your response above didn't really handle that core question. ChrysGalley (talk) 10:19, 21 July 2026 (UTC)

I thought it did. But for clarity's sake - no. I've been on Wikipedia for almost 20 years - the last thing I want to do is detonate my standing for something so trivial. Generally my work has been around data (ie the election results) and minor improvements to articles, especially around images and vectorisation. This is the first time that I've gone full-scale into writing whole articles from scratch because it is clear that no one else is doing it, and with the current constitutional crises going on in Zimbabwe right now, we need to make the information accessible and explicable to those who are seeking detail and information about what is happening and how things work. The issues identified above are clearly stylistic differences which we can gladly discuss. Let's go through some of the accusations for the sake of openness and collaboration:

1) Articles being published in quick succession: As stated, I've been working on articles in batches and bringing them across to Wikipedia when they're ready. As they are all on the same theme (ie Constitutional issues / organisations / agencies), there is naturally a lot of overlap and repetition, so it is easier to work on them in parallel.

2) The use of the word highlighted. Fine. Pick a different word. But the article did point out that the Opposition voted in line with the Government, a fact that in the context of the CCC recalls and capture of the opposition is a point worth stressing / emphasising / highlighting / underscoring / noting.

3) The deadlink: It wasn't dead when I first accessed it, but it is now. I have updated the link in the article and pointed out that the newspaper coverage is still accessible. This is the sort of issue that others could work on in a collaborative way, rather than an accusatory way that an editor has bad or malevolent intentions.

4) The lack of citing other outlets: As mentioned above, Zimbabwe has quite strict press censorship. Even 'independent' publications are under State scrutiny. This means that few other outlets can be cited in a meaningful way. Is the OP aware that Zimbabweans communicate news stories primarily through WhatsApp groups and through social media posts? These are not 'outlets' and can not be referenced in Wikipedia, but they are the nearest thing that you would get to alternative outlets in European and American countries.

5) Weasling. A new term to me. I'd not come across this before the OP started accusing me of things. But given how many Zimbabwean articles on Wikipedia are criticised for lack of neutrality, including even Rhodesia's 1965 UDI!, my concern was to not have anything too contentious to not be in breach of these issues.

The fact is that we need to work collaboratively here. Wikipedia is a collaborative platform. At the moment, I seem to be the only user working extensively on current Zimbabwe-related issues. There are lots of people working historically on topics around Rhodesia, but none / few on Zimbabwe. If there are issues, let's fix them together. In the meantime, the articles are being taken to states where they are either completely out of date, lacking any useful information, or both. The one on Elections in Zimbabwe for example now talks about "recent" elections being 2008. Well both the candidates in that election are dead, and one of the parties that contested that election no longer exists. It also talks about there being 270 members in the National Assembly - since 2023 there are 280. Much of what I'm working with - to try and remain factual, neutral and objective - are the laws of the land from Veritas and other constitutional platforms, so naturally there is much lifting from these official texts. If there were issues with the work I'd done - fine - I apologise sincerely and let's work together to fix it. But I can't see how leaving these articles riddled with errors is helping the Wikipedia project we all clearly care about greatly. Mangwanani ^(talk) 10:58, 21 July 2026 (UTC)

I've just spent some time now reading through the AISIGNS page that was linked to above, especially the AIVOCAB section. It is interesting to me, because many of the words listed there are ones I would naturally use. I would indeed begin a sentence with Additionally, (alongside moreover for variation), and Zimbabwean articles all talk about new laws "aligning with" the 2013 Constitution (see here and here and here. That leaves us in a difficult place now because it means any such usage now needs to be worked through a thesaurus to come up with an alternative, clunkier version if only to avoid the criticism of being the work of an LLM. Mangwanani ^(talk) 11:36, 21 July 2026 (UTC)

Well the language you have used in these posts do not seem to be AI in anyway whatsoever, but then the article text reads rather differently. Take National Prosecuting Authority of Zimbabwe and the section under Oversight and independence, we then get: The Prosecutor-General is independent and not subject to the direction or control of any person or authority. The Prosecutor-General and all officers of the NPAZ must exercise their functions impartially and without fear, favour, prejudice or bias. The trouble is that this is exactly how LLM would summarise 260 and 261 of the Constitution, however it's a fuzzy representation: 261 for officers does not specifically mention fear/favour (as it does say for many other positions such as magistrates) and that is in 260 for the Prosecutor=General. Instead 261 gives separate responsibilities for the officers, essentially not to be partisan. I think this would be pretty obvious under human oversight, at least it was to this human, whereas LLM just swallows and broadly represents, without being either accurate or inaccurate. Now I appreciate this is micro-dissecting something that is not hugely consequential, given the bigger picture, but hopefully you can see why the question arises. ChrysGalley (talk) 12:25, 21 July 2026 (UTC)

Well I'm hardly trying to write neutrally and encyclopaedically in these posts... I do agree that you are micro-dissecting the above sentence, especially when the wording "fear/favour" appear so often on the NPAZ website and across local media - oh sorry, I can't say local media any more lest it be seen as AIATTR ... ermmmm In this one article by The Herald - to this Zimbabwean, it is natural that the concept of the officers behaving in this manner is expected. But this further goes to show that that is yet another example of something being criticised for being written by an LLM, and rather than another editor going in to tweak it accordingly, the natural reaction from editors here would be to simply delete the whole article, leaving nothing at all available for researchers to look at. Again I make my appeal - let us not let the perfect be the enemy of the good. Mangwanani ^(talk) 12:45, 21 July 2026 (UTC)

• Noting the significant amount of damage done by Mangwanani and their continued evasive attitude as totally evidence from this lack of acknowledgement, I believe an indef block or block from article mainspace is warranted. Zalaraz (talk) 02:46, 22 July 2026 (UTC)

  it'd be a shame to lose such a productive long-term contributor like this, I hope they reconsider their response Kowal2701 (talk, contribs) 05:28, 22 July 2026 (UTC)

  Thank you Kowal2701. I believe I have tried to engage productively above. I have shown willingness to be collaborative with other users throughout. In all my work, my only aim and mission has been to improve the knowledge of the country I love - a mission which I seem to be tackling alone and with little to no cooperation and assistance from other users. Going forward, I now intend to stick with the data entry and beautifying aspects as I have done in the past as it is clear my own writing skills are leaving much to be desired by users who call my work “damage”.

  As an aside, I am yet to be shown something I have written that is untrue or misleading, whereas many of the pages being highlighted are full of untruths (the noted page above stating that there are 270 seats in Parliament when there are 280, for example).

  I again make my appeal - let us collaborate with one another to build up this platform we all treasure, not tear one another down and with it the good we are all here to do. Where a friend strays from the path, guide him back to the right way, rather than push him to his demise. Mangwanani ^(talk) 06:28, 22 July 2026 (UTC)

  In the "#Role and functions" and "#Impartiality and conventions" sections of Speaker of the National Assembly of Zimbabwe, the cited sources (the actual constitution) and don't support any of the content. Why? This is a very common thing for LLMs to do, and this appears to be replicated in other recent creations. You need to be honest with us or you will likely be blocked. Kowal2701 (talk, contribs) 06:40, 22 July 2026 (UTC)

  This is exactly the sort of collaboration I want! As I’ve pointed out before, it is hard to find online sources to evidence things here on Wikipedia for Zimbabwe-related topics. Much of the information is offline, or in sites that can’t be referenced on Wikipedia (for example Pindula, or various social media platforms - Facebook being the most significant for Zimbabwe-related political discussions). If you believe those sources are not strong enough for those sentences then a citation tag would be a good start to say “Hey friend - work on this”, as used to happen in the early days of Wikipedia.

  Clearly, as has consistently been pointed out, I’m biting off more than I can chew here, but I am trying to improve the space in good faith. No one has yet pointed out to me anything that I have written that has been wrong or deliberately misleading. I will work on that section now and see if I can improve the sourcing. Thank you Kowal! Mangwanani ^(talk) 06:51, 22 July 2026 (UTC)

  Don't use LLMs to edit in future, including AI chatbots like ChatGPT (please) Kowal2701 (talk, contribs) 07:12, 22 July 2026 (UTC)

  I pledge and swear to not do this. Mangwanani ^(talk) 07:24, 22 July 2026 (UTC)

  Thank you Kowal2701 (talk, contribs) 07:27, 22 July 2026 (UTC)

  I have worked on some of the sourcing issues that you have pointed out and I plan on doing more in the coming days.

  I have also started looking at some other drafts that I have worked on but not yet published. Can you help me with wording please? I see also you have flagged someone for using words like “highlighted” and “emphasised”. What can I use instead that would be acceptable as I am unsure how else you draw attention to these matters? Thank you for your help - and your collaborative, constructive (rather than destructive!) and engaging approach! Mangwanani ^(talk) 10:08, 22 July 2026 (UTC)

  WP:WTW covers this, but to answer your question, you may find an online thesaurus useful, personally I use WordHippo but there are probably better ones out there.

  Just speaking about LLM-generated content, people often underestimate just how much work is needed to make it policy-compliant; it's easier and less time-consuming to do a complete rewrite based on the sources (you probably know this but I usually recommend Google Scholar and Google Books for finding sources, WP:TWL gives access to most that are behind a paywall). Kowal2701 (talk, contribs) 10:45, 22 July 2026 (UTC)

  Just to clarify, it's not necessarily that there's a problem with those or any other specific words in and of themselves; more that they often point to one or two specific mistakes that LLM-generated text tends to make over and over when describing sources about virtually any subject, the kind of mistake that isn't really the kind of thing people do on their own or did very often prior to 2023. Gnomingstuff (talk) 14:35, 22 July 2026 (UTC)

  I would not worry too much about vocabulary. Things that are critical to fix: Any issue with sources, like dead links, wrong ISBNs, or sources that do not support the claim. Statements that "many" people are saying something sourced to just one person saying it, and variations on that (this is common and bad). Vague, empty praise, like "a significant addition to the corpus of" (made-up example, but it's the kind of thing LLMs tend to say). Unneeded lists and tables where text would be better. Made-up plot summaries--this is especially hard to check as plot summaries don't have to be sourced, but LLM cannot summarize a book that's not on the Net, so just makes things up.)

  But people are right: it's usually easier to rewrite the whole thing, speaking as someone who's fixed a few of these. M kuhner (talk) 14:48, 22 July 2026 (UTC)

Juc247 (talk · contribs), XTools

See recent creations like Economic Reporters Forum (Bangladesh) (100% AI according to GPTZero for Wikipedia), there's lots of WP:AILIST, WP:AITABLE, though weirdly not much WP:AIVOCAB (no signs of human-written content either, it's very plausibly AI). Similarly there's Rozina Islam (which GPTZero gives 90% human-written, what a useful tool), in addition there's the unlinked See alsos which is odd. At Kazi Mostak Gausul Hoq, (the subject's personal profile on a commercial site) is used for He is known for writing books on library and information science as well as fiction, history, literature and technology. which doesn't appear to support much of that, especially is known for. A lot of the biographies have/had unsourced lists like and which are typical of LLM output. Honestly I'm a bit stumped here, it may just be someone who's being clever w their prompts (ie. aware of WP:AISIGNS or using some humanising tool), I really don't know. Kowal2701 (talk, contribs) 18:17, 21 July 2026 (UTC)

I asked them Kowal2701 (talk, contribs) 18:23, 21 July 2026 (UTC)

Response:

No, I'm not using LLMs for my edits. But, I use LLMs for translation-related work. Thanks. Juc247 (talk) 18:33, 21 July 2026 (UTC)

would appreciate second/third opinions Kowal2701 (talk, contribs) 18:44, 21 July 2026 (UTC)

WP:LLMT says a bunch of things, and those things do not look like they've been respected. ~2026-38678-64 (talk) 19:04, 21 July 2026 (UTC)

"oh let me clean up my mess" "filter disallow: user blanking talk page" oh my god ~2026-38678-64 (talk) 19:16, 21 July 2026 (UTC)

I CAN'T EVEN CHANGE MY COMMENT??? ~2026-38678-64 (talk) 19:22, 21 July 2026 (UTC)

What changes do you want to make? Kowal2701 (talk, contribs) 19:39, 21 July 2026 (UTC)

change my comment to

More information Extended content ...

Extended content
::according to [[WP:LLMTRANSLATE|the translating with LLM guideline]], you cannot put your raw LLM-translated text in the article mainspace unless you've edited it a lot, could you tell us where you used LLM-translation tools on wikipedia? thanks [[Special:Contributions/~2026-38678-64|~2026-38678-64]] ([[User talk:~2026-38678-64|talk]]) 19:09, 21 July 2026 (UTC)

Close

because it's better and has less text and so on ~2026-38678-64 (talk) 19:49, 21 July 2026 (UTC)

well you know what you could do about that :p ‑‑gurkubondinn 13:48, 23 July 2026 (UTC)

Even after translating content from another language, I make manual edits to improve text.

Why do people think that inputting a translation request to an LLM does not produce LLM-generated text? Or why do they think that "a translation" means "an LLM-generated translation"? How do people think that books are translated? ‑‑gurkubondinn 13:43, 23 July 2026 (UTC)

@Kowal2701 Honestly I'm a bit stumped here, it may just be someone who's being clever w their prompts (ie. aware of WP:AISIGNS or using some humanising tool seems neither ~2026-38678-64 (talk) 06:25, 22 July 2026 (UTC)

 Cleanup has been requested.

Matete Plays (talk · contribs · deleted contribs · logs · filter log · block user · block log)

Hi, I'm relatively new to this process so I sincerely apologize if I'm doing anything incorrectly here, but I discovered this user making 5 huge edits, including creating 2 new articles with around 2000 words each (Popular Party (Albania) & Progressive Party (Albania), within the span of 40 minutes. These all contain several WP:AISIGNS, including invalid ISBNs, rule of three, and (barely edited) negative parallelisms (e.g. "The Popular Party was not a modern mass-membership party with a permanent organisation, disciplined membership or detailed platform. It was a loose alliance of parliamentarians, regional notables and intellectuals whose loyalties were often personal and changeable." in Popular Party (Albania))).

I tagged both new articles for being possibly AI-generated, but they've been making huge edits on Albanian history & politics articles, often within a small amount of time, for at least a month and a half as of writing, so this might need the attention of more experienced users here. Thank you all in advance for your help! Knypster (talk) 16:38, 22 July 2026 (UTC)

Your report is fine! We'll check out their previous edits and work out next steps. Appreciate your help. M kuhner (talk) 16:57, 22 July 2026 (UTC)

Hello editors,

I do not want to waste your time. Yes, the last pages I created were made using AI. My goal is to improve Albanian coverage on the English Wikipedia, which in some areas is extremely scarce. To achieve that goal, I used AI while personally verifying that the information it provided was accurate.

However, after reading Wikipedia's AI policy, I realize that the way I used AI was not allowed. I sincerely apologize for this mistake and will refrain from doing so in the future. Matete Plays (talk) 17:58, 22 July 2026 (UTC)

Thank you! We appreciate the quick and human response.

It would really help if you could provide a list of the LLM articles, or a date at which you started using LLM. This helps us avoid touching articles that were not LLM, which may otherwise be unnecessarily deleted. Alternatively, you could move your LLM articles to your own user space, and clean up LLM edits to existing articles yourself. That would be very much appreciated. M kuhner (talk) 18:31, 22 July 2026 (UTC)

I am sorry to say, on August 10 Matete Plays made this edit: and the first source given says nothing about a demonstration in Tirana. I am concerned that LLM use has not stopped. M kuhner (talk) 05:06, 13 August 2026 (UTC)

Sorry, I'm afraid you overlooked the last source of that paragraph: https://lapsi.al/pamjet-proteste-para-ambasades-se-ukraines-dhe-asaj-te-serbise-me-thirrje-per-uck-ne-i1091463 Matete Plays (talk) 20:14, 14 August 2026 (UTC)

Nampong123 (talk · contribs), XTools

See Wikipedia:Administrators' noticeboard/Incidents#User not stopping their mass-creation of articles despite community concerns., ~300 article creations since 13 July (not a typo). was the first, created with a maintenance tag dated to 2022. Kowal2701 (talk, contribs) 23:58, 22 July 2026 (UTC)

Oh man, those are as bad as sports articles. Hardly any text. I'm inclined to wait a bit, since the editor is responding to the ANI case. M kuhner (talk) 00:34, 23 July 2026 (UTC)

Tbf it's very plausible that the maintenance tag derived from them copying another article with that tag as a 'template'. Their denial seems genuine tbh, worth doing some spot checks of prose and results, but seems like this was done the old-fashioned way Kowal2701 (talk, contribs) 16:57, 23 July 2026 (UTC)

User is now blocked as a sockpuppet and the block appeal does not seem to be going anywhere. The contribs of the two minor socks named in the SPI don't require cleanup ^purring_maggot (talk) 01:31, 27 July 2026 (UTC)

It looks like admins have reverted nearly all of Nampong123's edits (and tons of creations have been draftified), but a few are still standing and a few have been undone by another editor in the ANI thread who stated they are selectively restoring useful edits, so if we don't strongly believe these are LLM-generated maybe cleanup is not necessary? ^purring_maggot (talk) 01:40, 27 July 2026 (UTC)

I was going to say, we need a tracker anyway, to make sure. Then I tried making one. 23K edits; over 1000 edits in the last couple days; tracker not possible. I did look at articles created: all have been deleted except 5 of the earliest 7, and those are clearly before the article-creation spree. I think given the volume, the "old-fashioned way" would have had to be some form of database dump: 238 articles in 12 days is pretty staggering, but you could do it by filling in a template. From the unblock demand (one can hardly call it a request) I think at some point they decided to Right Great Wrongs and that mass article creation was the way to do it. It's unfortunate. M kuhner (talk) 04:11, 27 July 2026 (UTC)

 Cleanup has been requested.

 Bustersidebury (talk · contribs · logs · block log) (23 July 2026)
Editor has been warned about the use of AI/LLM content and WP:NOLLM. 

Bustersidebury (talk · contribs · deleted contribs · logs · filter log · block user · block log)

Rapid implausible rate of article creations/expansions, most of which have subtle but visible AISIGNS. At wikimania rn so can't go through the usual spiel: @Admins willing to patrol AINB: - maybe some help? Fermiboson (talk) 12:07, 23 July 2026 (UTC)

I have warned Bustersidebury and notified them of this discussion. — Newslinger talk 12:40, 23 July 2026 (UTC)

I've seen this discussion and understand the concern. My expansions drew on newspaper archives and trade sources I have been gathering for some time on topics that interest me, which is why several came together quickly. I take the point about pace, and I will slow down and keep contributions smaller and more incremental. The sourcing is real and checkable. If anyone spot checks a claim or citation and finds a problem, point it out please, and I will fix it promptly. Bustersidebury (talk) 15:49, 23 July 2026 (UTC)

Hi Bustersidebury, could you please disclose the full extent to which you have used a large language model for your edits? — Newslinger talk 16:31, 23 July 2026 (UTC)

Hi. In some recent edits, I used an LLM for tasks including formatting citations, as it's not my area of expertise, standardizing markup, and catching typos. I selected and read the sources myself. I checked the factual claims against them. My focus has been on articles that lacked reliable sources, had broken references, or contained promotional language. I did not keep an edit by edit log, so I can't provide a precise breakdown. If you identify a specific claim or citation that appears problematic, I'll check it against the source and correct any error. Bustersidebury (talk) 17:02, 23 July 2026 (UTC)

@Bustersidebury: Precisely which LLM-based tools did you use? Also, could you please explicitly state whether you used an LLM to generate or rewrite article content, aside from the citation formatting and copyediting? — Newslinger talk 19:30, 23 July 2026 (UTC)

One of your recent edits sources the claim "some of his later albums" to a source that says "Ray Price's last record." This type of increase from one to several is a hallmark of recent LLMs. They will write "Several critics said" when only one did so, "According to multiple sources" when there is only one, etc.

I'm not so much concerned about this specific small error as whether it represents a pattern of LLM misrepresentation of sources. M kuhner (talk) 16:27, 24 July 2026 (UTC)

@Bustersidebury, I've draftified your two creations (Draft:Steve Bogard and Draft:Mae Estes) in case you wish to work on them (usually we delete LLM-generated articles per WP:LLMPROD). The biggest issue with LLM use is hallucinations (i.e. WP:V failures), though there are other important issues like WP:NPOV/WP:WTW and WP:COPYVIO. Usually we recommend that people stubbify or rewrite LLM-generated articles from scratch since it's often very time-consuming to attempt to preserve such content, but it's up to you. Best, Kowal2701 (talk, contribs) 22:03, 9 August 2026 (UTC)

Hades7 (talk · contribs · deleted contribs · logs · filter log · block user · block log)

User hasn't been active in some months but produced a few extremely broad pages related to Western esotericism chock full of AI hallucinations and failed verification problems. According to Gnomingstuff (Talk:Western esotericism and Eastern religions) this was admitted though I can't find where. Only six articles, Western esotericism and Eastern religions, Western esotericism and psychology, Western esotericism and arts, Western esotericism and science, but also Matrona of Perge and Jean-Marie Déchanet. Not sure if they all have this problem but I'm inclined to think the Western esotericism ones do from my knowledge of the topic. See also PARAKANYAA (talk) 13:31, 23 July 2026 (UTC)

All of those "Western esotericism and ..." articles can probably be redirected to a section in the Western esotericism article (or just LLMPROD since the user is inactive)? ‑‑gurkubondinn 13:35, 23 July 2026 (UTC)

The sections don't match up with any on that article (but they are probably notable topics) so LLMPROD is probably better. PARAKANYAA (talk) 13:42, 23 July 2026 (UTC)

LLM use probably started here (11 august 2025), i did a spot check on a few of them vs their usual edits and it strikes to me as definitely AI-assisted (subtle signs, X rather than Y, not a lot of -ing but a lot of "emphasized") (also look at the "edit source" on any of those

i do not believe that jean-marie or matrona are LLM-generated from a quick glance, but the western esotericism articles look like a mix of LLM-translation, assistance, and generated text ~2026-38678-64 (talk) 14:05, 23 July 2026 (UTC)

The comment is here, made the research of some footnotes with the clasical Google research, but in the reasearch of comparative sources for trying to discover which could be the not so well cited sources in the book I use as a template I consulted AI, so I think that give me barely related sources. Gnomingstuff (talk) 14:29, 23 July 2026 (UTC)

Also here: I'm thinking what happened is not the barely related sources as a result of an online/AI research, because I have read the sources and I'm sure they matched -months of reading these issues- but when I gave the AI the list of the sources and pages that I already had and needed to be ordered in lists and topic conections (the architecture of concordances I made before the redaction of the article, as an algorithmic system), the lines of some references were modified at some point ("hallucinations") in a "broken telephone game" style. Anyway, I'm checking the matching again. Gnomingstuff (talk) 14:30, 23 July 2026 (UTC)

Hello, Matrona of Perge and Dechanet are totally manual articles, word by word.

The ones about Western esotericism I have to complete checking to match the real sources. The problem was that AI assistence moved the pages of my references. Per example, I used Brill. Brill is scholar benchmark about Western esotericism. But when I asked to put my references in the narration the number of pages or chapters were moved. So, the scholar content is there, but the pages or the titles of the chapters where moved/modified. That's is the main mismatch. I thought that if I do by myself the architecture and give the AI the scholar data, it will organize it better than me. But I failed with that part.

So, I know the articles are good research and the scholar data is available, the information is real, but I have to check line by line the references to fix the match and that will take me several weeks. I will do it. Hades7 (talk) 09:24, 24 July 2026 (UTC)

@Hades7, I've draftified the four articles you've mentioned above in case you wish to work on them (usually we delete LLM-generated articles per WP:LLMPROD), you can see a list of your drafts here. The biggest issue with LLM use is hallucinations (i.e. WP:V failures), though there are other important issues like WP:NPOV/WP:WTW and WP:COPYVIO. Usually we recommend that people stubbify or rewrite LLM-generated articles from scratch since it's often very time-consuming to attempt to preserve such content, but it's up to you. Best, Kowal2701 (talk, contribs) 22:17, 9 August 2026 (UTC)

 Cleanup has been requested.

User has been rapidly creating and expanding articles about plants. Their edits contain AI tells for species articles. See for example the speculative "Conservation" sections at Eryngium arenosum and Boltonia caroliniana. Cicada1010 (talk) 19:18, 23 July 2026 (UTC)

Those do look like AI articles. Have our guidelines around llms been raised with the user anywhere? I don't see it on their user talk. CMD (talk) 03:08, 24 July 2026 (UTC)

Their first AfC decline mentioned it, but the generic LLM decline still reads as sort of a suggestion to not use LLMs rather than being 100% clear about the prohibition, so I've warned the editor and invited them to discuss here. The AISIGNS are pretty apparent in the two articles linked here (don't have time to look at more right now) ^purring_maggot (talk) 03:41, 24 July 2026 (UTC)

Hello! I should've looked into the guidelines more when I was warned previously, but I took a break immediately after that and had forgotten to look into it further.

I would like to run through the process I have been using involving LLMs in order to create pages. When making articles, I would copy and paste the sources and their links into ChatGPT and then ask it to generate each section. I would then put the generated text into the article and cross reference with the original sources to ensure no hallucinations or blatant plagiarism took place, along with altering some wording or linking to other pages. Images were found through appropriately copyrighted INaturalist posts and uploaded with no help of AI, along with listing categories and Wikidata creation. I just assumed this was allowed as it made article generation much more efficient, but I understand if this violates the rules of Wikipedia and needs to be reversed.

I am more than willing to aid with the cleanup, just let me know what needs to be done! Niyan Shehan (talk) 21:41, 4 August 2026 (UTC)

Hi @NShehan (sorry for the late response, this page is really busy), the relevant guideline is WP:NOLLM (also, thank you very much for your honesty). Are you near 100% sure there are no hallucinations (i.e. WP:V failures)? That's the biggest concern, though there are other important issues like WP:NPOV/WP:WTW and WP:COPYVIO. Normally I'd either draftify the creations for you to review yourself when you get the chance, or WP:LLMPROD them (i.e. delete). Regarding cleanup, usually we recommend that people stubbify or rewrite LLM-generated articles from scratch since it's often very time-consuming to attempt to preserve such content, but it's up to you. People often underestimate just how problematic LLM-generated content is, we had an editor who had professional expertise in lichen and was creating articles while systematically using AI, they're now blocked because they underestimated this and continued creating after problems were pointed out (see Wikipedia:AI noticeboard/Archive 7#Esculenta and GAs). Kowal2701 (talk, contribs) 19:38, 10 August 2026 (UTC)

Yes, using LLM makes generation of things that look like Wikipedia articles very efficient, but generating an informative, verifiable Wikipedia article seems to be pretty difficult with LLM. I haven't had time to do much AI review lately but good examples of problems are in the Conservation sections of the articles in the original report:

Eryngium arenosum has 3 paragraphs of superficial analysis and unsourced speculation. A speculative list of "potential threats" and a checklist of conditions for which "additional field surveys are needed" simply do not belong in the article. That is the LLM making up fluff because it doesn't have any solid facts to incorporate. This doesn't provide any information whatsoever. Also, Its conservation rank reflects its narrow distribution, restriction to specialized South Texas Sand Sheet habitats, and relatively small estimated number of occurrences. What source says its conservation rank "reflects" these things? What does it mean for its conservation rank to "reflect" these things? It's possible that it would be appropriate to describe sourced facts about how the conservation rank for this species is determined, but not just WP:SYNTH together a small collection of information and hint at these things being related to each other somehow.

Boltonia caroliniana has a bit of the same fluff plus aggressively attributing everything to NatureServe as if it's a controversial opinion that we have to be very careful about presenting as fact. The NatureServe conservation ranks are used in loads of Wikipedia articles so I assume we can note once that this is NatureServe's rank (to distinguish it from IUCN) and then allow the citations to speak for themselves as to where the information came from. The section goes on and on about where the species does not have a conservation status. We don't need to include a list of everything the subject is not in its Wikipedia article. If it doesn't have something, we can just not say it (unless not having that thing is remarkable in some way).

Most species articles are stubs for a reason: there simply isn't that much available information about the species. It's much better for a reader to read 3 sentences about a topic and come away with 3 verifiable facts than to read paragraphs upon paragraphs of lorem ipsum and come away with "we don't really know how it looks, we don't really know where it lives, we don't really know what the conservation status is..." etc. Humans have done a lot of weird crap on Wikipedia so I won't say we're always better at determining what goes in an article, but we have the advantage of being readers as well as editors, so when we read a source the information that stands out to us as "that should go in the Wikipedia article" is probably going to also be interesting and useful to other people who then read that article, whereas an LLM has no sense of those concepts and is just indiscriminately scraping data to fill space since it was trained that longer answers are always better. ^purring_maggot (talk) 21:21, 10 August 2026 (UTC)

 Cleanup has been requested.

• PanoptikEU (talk · contribs · deleted contribs · logs · filter log · block user · block log)

The user has good intentions and probably created all their articles before WP:LLM was prohibited but still we need to do a lot of cleanup. Most of their articles were created rapidly without any human review. Maybe convert notable Estonian articles into one-line stubs. ~2026-41256-48 (talk) 12:13, 25 July 2026 (UTC)

129 creations in 233 edits Kowal2701 (talk, contribs) 18:38, 28 July 2026 (UTC)

Their latest creations are obviously LLM-generated. Looking at their early articles, they seem decent quality compared to what we normally see and there aren't really enough meat on them for AISIGNS, but: at Sirje Olesk ref #1 doesn't support She is a recognized authority on the Finnish-Estonian writer Aino Kallas and the poet Lydia Koidula., Konrad Duden Prize has WP:AIREDCAT, and in Sven-Erik Soosaar ref #3 appears hallucinated Kowal2701 (talk, contribs) 21:22, 29 July 2026 (UTC)

Tbh I'm not 100% sure these are all AI, would appreciate a second opinion Kowal2701 (talk, contribs) 21:30, 29 July 2026 (UTC)

I disagree, don't think these three are AI - Sirje Olesk source kinda supports it thru "She has mainly dealt with various aspects of Estonian poetry, Estonian-Finnish literary relations, the life and work of Aino Kallas and Lydia Koidula, studied the history of the Estonian Writers’ Union and literary life in Soviet Estonia". Soosaar's ref #3 did in fact exist but the page went down - see https://web.archive.org/web/20190915191335/https://www.ut.ee/et/uritused/sven-erik-soosaar-eesti-keele-kultuursonavara-arengujooni-eesti-keel-piirikeel, an archived copy from 2019 (display is broken, but if you scroll down a bit you find the article).

And as for the red category, i know (especially before I got HotCat) I've occasionally put in categories assuming they were functional categories and then checked, noticed they weren't, and deleted - i find it harder to believe this is AI than that this is a user forgetting to check whether a reasonable category they inputted actually exists. Jishara (talk) 01:58, 1 August 2026 (UTC)

I've put a few into Pangram and they're consistently coming back as AI-assisted, I think this meets the threshold for LLMPROD but does anyone object? Kowal2701 (talk, contribs) 12:14, 11 August 2026 (UTC)

Cleanup efforts are ongoing and being tracked at /2026-08-11 Keechak9999, 3 pages need review.

• Keechak9999 (talk · contribs · deleted contribs · logs · filter log · block user · block log)

This user has been suspected of using LLM to review at GAN, at this discussion. Reviews can be found at Talk:Brewing/GA1 and Talk:E A Terra Nunca Me Pareceu Tão Distante/GA1, the suspicion of LLM is valid in both. My main concern, however, has been their article creation.

The following have already been deleted:

• List of bridges in Mumbai
• List of Hindu temples in Mumbai
• Sri Ahobila Mutt, Mumbai

The following is at AfD and flagged as potential AI by me:

• Lamtex Insulation - refs 2, 5, 8 and 9 are hallucinated

These are their other articles, not yet suspected of being AI generated:

• Sindoor Bridge
• Draft:List of flyovers in Mumbai

Note that I also reverted this edit at Fort (Mumbai precinct) as it looked like unreviewed LLM output to me. I issued them another warning. My guess is that this is a good faith editor, perhaps not a native English speaker. I do feel, however, that warnings are falling on deaf ears. I would appreciate other opinions. Spiderone^{(Talk to Spider)} 12:43, 25 July 2026 (UTC)

ItsMeGabeProductions (talk · contribs · deleted contribs · logs · filter log · block user · block log)

User communication found in specifically in the LLM comments in the casualties and identification section. Superficial AISIGNS also throughout. User decided, very sneakily, to simply remove the comment without addressing the fundamental issue. Presume with the long history of unknown parameter warnings that more scrutiny is required. Fermiboson (talk) 13:23, 25 July 2026 (UTC)

Hi, I am so sorry for using AI, I thought that even if I review all my articles, that they wont be removed as I followed G15, I am so sorry for using AI and I will even go as so far to accept a ban and make articles by hand, no AI, thank you if you read this ItsMeGabeProductions (talk) 13:28, 25 July 2026 (UTC)

G15 is supposed to be WP:LLM sorry ItsMeGabeProductions (talk) 13:36, 25 July 2026 (UTC)

Since March 2026 WP:LLM has prohibited using AI to generate any text that goes into an article, and prior to that (I believe since Oct 2025?) it's prohibited creating new articles using AI. Since you've committed to never using AI from now on, I assume a block is not necessary. It would be helpful in the cleanup effort if you could tell us how long you've been using AI to edit. ^purring_maggot (talk) 22:34, 25 July 2026 (UTC)

For articles, it starts with Març Llinàs, but for edits, I have never used AI, like actually, I used AI just for making pages ItsMeGabeProductions (talk) 03:05, 26 July 2026 (UTC)

Any updates? ItsMeGabeProductions (talk) 10:05, 31 July 2026 (UTC)

Sorry for the delay, this page is very busy. Thank you for your honesty. Is it okay if I draftify your creations from Març Llinàs onwards (45 in total) for you to work on before moving back to mainspace? Usually we recommend stubbifying or rewriting LLM-generated articles, rather than attempting to preserve as much content as possible, because it can be very time-consuming, but ofc that'd be up to you (otherwise we'd unfortunately have to delete them, see WP:LLMPROD). The most important thing to review for is that the sources support the content (WP:V) because of potential hallucinations. Best, Kowal2701 (talk, contribs) 19:53, 31 July 2026 (UTC)

Its okay, ill just fix all of them ItsMeGabeProductions (talk) 03:05, 1 August 2026 (UTC)

There might be a few hallucinations due to the fact that I check my sources but I still have to review them ItsMeGabeProductions (talk) 03:10, 1 August 2026 (UTC)

@ItsMeGabeProductions, can I ask why, at time of creation, Kariman Hamza had a file that didn't exist? Kowal2701 (talk, contribs) 17:59, 3 August 2026 (UTC)

I translated it from arabic wikipedia, that had a file, so I uploaded it here, it ended up deleted and I forgot to remove it ItsMeGabeProductions (talk) 00:49, 4 August 2026 (UTC)

Repuploaded* also the original file in arabic wiki was a free image and i uploaded it onto commons ItsMeGabeProductions (talk) 00:51, 4 August 2026 (UTC)

That file's never existed on commons (maybe you were thinking of ?) Kowal2701 (talk, contribs) 19:14, 5 August 2026 (UTC)

From what I can remember, I reuploaded the file from arabic wiki to wikicommons (maybe wikipedia) and it got AfDed ItsMeGabeProductions (talk) 02:19, 7 August 2026 (UTC)

@ItsMeGabeProductions, are you sure you didn't use AI for Osvaldo Jaldo? Kowal2701 (talk, contribs) 12:22, 11 August 2026 (UTC)

Every article before my return is not made with ai as they were translated using the content translation feature, then when I returned to wikipedia, I started making my articles with AI but I reviewed them for hallucinations ItsMeGabeProductions (talk) 22:48, 11 August 2026 (UTC)

 Cleanup has been requested.

Munfarid1 (talk · contribs · deleted contribs · logs · filter log · block user · block log)

Was unambiguously using ChatGPT for draft article content in 2025 as evidenced by User:Munfarid1/sandbox2 (+1 pattern, along with markdown etc). Their current edits are translations, but this appears to be a very very loose definition of the term and certainly not compliant with WP:LLMTRANSLATE.

To take one example: The original version of the article for Louisa Tounsia, created 3 days ago, helpfully includes the original French text, which obviously does not correspond to the "translated" text at all. The article has developed from there but the initial kernel appears to be straight-up LLM output, and the edits to it do not appear to have been done in accordance with guidelines or reading of the source. Again to take one example:

• Version 1 includes the uncited sentence Her recordings were issued by several of the principal recording companies operating in the Middle East, including Baidaphon, Gramophone, Odeon, and Polyphon. Baidaphon and Odeon are not mentioned in the original French text.
• By the current version it has been cosmetically changed to A prolific entertainer, Tounsia recorded 78 rpm and later 45 rpm vinyl records in Arabic and French for several of the principal recording companies operating in the region, including Baidaphon, Gramophone, Odeon, and Polyphon. It has also been cited to this source. Baidaphon and Odeon aren't mentioned in it either.

Gnomingstuff (talk) 16:49, 25 July 2026 (UTC)

(which, incidentally, also means that they have produced a ton of AI-generated DYKs, I wish people would notice this stuff earlier) Gnomingstuff (talk) 16:57, 25 July 2026 (UTC)

Hi, as you rightly point out, my draft includes very little of the French article I started out translating. The published version has been written largely on other sources, as those in the French article were nowhere enought for a standalone article in English. As far as I can see, my translations from the French are not in violation of LLMTRANSLATE except for the rule of tagging it as LLM-translated, of which I was not aware of. Further, I want to state that I am sufficiently competent in both languages to ascertain that my published translations are correct. - How shoud I go about this now regarding the published article Louisa Tounsia? Is there any need to tag this after the draft? Munfarid1 (talk) 17:27, 25 July 2026 (UTC)

There's one very key thing missing here, which is whether you used AI and to what extent. If you are using it in a way that is not a translation, then that violates WP:NOLLM. Gnomingstuff (talk) 20:33, 25 July 2026 (UTC)

I have used AI to suggest summaries or paraphrases of some of my sources. However, I have always checked the wording included in the final text and often deleted, shortened or rephrased AI-suggested text in my own words. Writing the draft has taken many hours and was done over 13 writing sessions, adding or shortening text based on different sources including translations. Therefore, I am unable to say to what extent some of the published text was originally suggested by AI. In any case, I have tried to comply with WP's basic guidelines and am confident there are no hallucinations or gross mistakes. (BTW: Thanks for pointing out the mistaken mention of Baidaphon and Odeon labels that I have corrected right away.) Munfarid1 (talk) 22:27, 25 July 2026 (UTC)

Can you tell us the precise AI model names and versions you used? And if possible, share links to the chat sessions or prompts? —ClaudineChionh (she/her · talk · email) 00:33, 26 July 2026 (UTC)

ChatGPT based on GPT-5.5. - Following are some chat sessions I used:

1) The statement that **Louisa Tounsia collaborated with Messaoud Habib** ultimately traces back to a small number of sources, with varying levels of authority.

The strongest published source is **Christopher B. Silver**, whose doctoral dissertation documents Messaoud Habib's work with Louisa Tounsia:

> "Messaoud Habib played or recorded with the Tunisian artists Asher Mizrahi, Babi Bismuth, Cheikh El Afrite, Dalila Taliana, Khailou Esseghir, **and Louisa Tounsia** –– to name but a few." ([escholarship.org][1])

Silver also identifies Habib as the **"virtuoso pianist"** accompanying Louisa Tounsia's orchestra in the Palmarium theatre in Tunis, based on Pierre Mac Orlan's contemporary account. ([escholarship.org][2])

In addition, Silver has written more specifically on his archival music-history blog:

> "As with many of Louisa Tounsia's discs, *Heukm Ennessouane* ends with an exquisite piano solo by Tunisian Jewish pianist Messaoud Habib. **Over the course of her career, Habib would accompany her often.**" ([gharamophone.com][3])

1. 1. 1. Evidence from the recordings themselves

The collaboration is also documented directly in archival record metadata. For example, the Tunisian **CMAM/Telemeta** archive lists the 1935 Polyphon recording **"Shuftak Marra Mtaʿaddiya"** with:

• **Louisa Tounsia** – vocals
• **Masʿūd Ḥabīb (Messaoud Habib)** – organ (or harmonium, depending on cataloguing terminology) ([phonotheque.cmam.tn][4])

Other CMAM catalogue entries likewise identify Messaoud Habib as part of the ensemble accompanying Louisa Tounsia on Pacific recordings. ([phonotheque.cmam.tn][5])

1. 1. 1. Origin of the wording "worked with"

The wording that Louisa Tounsia **"worked with the artists Youcef Hadjadj and Messaoud Habib"** appears in the French Wikipedia article and has been reproduced on several websites (for example Harissa.com and Last Night in Orient). ([Wikipedia][6])

However, those versions do not cite an independent published source for that exact sentence and appear to derive from the same biographical text rather than separate research.

1. 1. 1. Best source to cite

For a scholarly publication or a Wikipedia article, I would recommend citing **Christopher B. Silver** rather than the derivative websites. His dissertation and subsequent research provide the strongest evidence that:

• Messaoud Habib regularly accompanied Louisa Tounsia on recordings.
• He performed with her orchestra.
• Their collaboration spanned multiple years and record labels. ([escholarship.org][1])

[1]: https://escholarship.org/content/qt9g4352x5/qt9g4352x5_noSplash_168088494c66c408a78098897feb6ad6.pdf?utm_source=chatgpt.com "UNIVERSITY OF CALIFORNIA"

[2]: https://escholarship.org/content/qt9g4352x5/qt9g4352x5_noSplash_168088494c66c408a78098897feb6ad6.pdf?t=ota9ua&utm_source=chatgpt.com "UNIVERSITY OF CALIFORNIA"

[3]: https://gharamophone.com/2018/03/29/louisa-tounsia-heukm-ennessouane-pathe-c-1930-1931/?utm_source=chatgpt.com "Louisa Tounsia – Heukm Ennessouane – Pathé, c. 1930-1931"

[4]: https://phonotheque.cmam.tn/archives/items/D-0098/?utm_source=chatgpt.com "Item : شفتك مرة متعدية - Telemeta - Open web audio platform with semantics"

[5]: https://phonotheque.cmam.tn/archives/items/D-2589/?utm_source=chatgpt.com "Item : خوذوا الدنيا - Telemeta - Open web audio platform with semantics"

[6]: https://fr.wikipedia.org/wiki/Louisa_Tounsia?utm_source=chatgpt.com "Louisa Tounsia" Munfarid1 (talk) 11:37, 26 July 2026 (UTC)

Paraphrase: The very cover of Columbia’s Tunisian supplement, which was graced by popular

Tunisian Jewish artist Louisa Tounsia, put a fine point on the image the record labels were keen

to promote.

Answer: Here are several neutral, Wikipedia-style paraphrases:

1. **Recommended:**

> The cover of Columbia's Tunisian record supplement featured the popular Tunisian Jewish singer Louisa Tounsia, reflecting the public image that the record company sought to project.

2. **More formal:**

> Louisa Tounsia, one of Tunisia's most popular Jewish performers, appeared on the cover of Columbia's Tunisian record supplement, underscoring the image that the label aimed to cultivate.

3. **Closer to the original:**

> Columbia's Tunisian record supplement featured Louisa Tounsia on its cover, highlighting the image that the record label wished to promote through its marketing.

4. **With attribution (preferred if this is the author's interpretation):**

> According to Christopher Silver, the decision to feature the popular Tunisian Jewish artist Louisa Tounsia on the cover of Columbia's Tunisian record supplement reflected the image that record companies sought to promote.

The **fourth version** is the most appropriate for Wikipedia if the statement about the image the label "was keen to promote" is Silver's interpretation rather than an explicit claim made by Columbia itself. It attributes the analysis to the source while keeping the factual statement (that Tounsia appeared on the cover) separate from the interpretation. Munfarid1 (talk) 11:49, 26 July 2026 (UTC)

And this is what I wrote in the article: Among other aspects, he commented on a picture of her on the cover of the Columbia Records 1931 Tunisian record supplement: Dressed both in traditional and modern European clothes, Silver noted how Tounsia exemplified the image that record companies sought to convey. Munfarid1 (talk) 14:31, 26 July 2026 (UTC)

And another example from ChatGPT: A source-safe formulation for Wikipedia or academic writing. Based on the evidence actually located, I would write:

“Louisa Tounsia recorded Ala Bab Darek in 1945; the song, associated with the musical heritage of the Kerkennah Islands, was later reinterpreted by Emel Mathlouthi in 2012.”

This formulation is supported by the available Discogs documentation

And my text: "Among other well-known songs in her repertoire is her 1945 recording "Ala Bab Darek" ("At the Door of Your House"), a song from the musical heritage of the Kerkennah Islands, that was covered by Emel Mathlouthi." (Based on this statement's source: "Emel Mathlouthi, the singer of the recent Tunisian revolution, has taken to singing Louisa Tounsia’s Ala bab darek on a number of occasions." Munfarid1 (talk) 14:47, 26 July 2026 (UTC)

cleaned Habiba Msika, Music of Morocco, and Arabic Music, Asher Mizrahi seems fine ~2026-33562-67 (talk) 18:34, 26 July 2026 (UTC)

The total reversing of all my contributions to Habiba Msika seems excessive, as the infobox, my own text that was not LLLM-generated and many sources have been deleted. As I have already spent a lot of time on the subject, I will continue to work on it. Of course, I will only use my own wording and never any LLM-generated text in contradiction of the rules from now on. Munfarid1 (talk) 08:08, 27 July 2026 (UTC)

The problem with LLM text is that is much faster to paste it in than it is to take it out. Elsewhere on this noticeboard you will find a case where someone made over 200 new articles in a few days. This means that we can't afford to do a careful hand correction of every article. We would very much welcome a human-written article on this topic. If the LLM text would be helpful you could request a copy be emailed to you via WP:REFUND, however I have found that starting over clean is generally easier and safer. The LLM text tends to have subtle falsehoods that are very time-consuming to catch and fix.

Good-faith editors get caught by the LLM trap on a daily basis. We're struggling with how to prevent this. M kuhner (talk) 15:22, 27 July 2026 (UTC)

Thanks for your answer. I will do my best to rewrite this as a 100% human written article. Munfarid1 (talk) 18:36, 27 July 2026 (UTC)

@Munfarid1, can you please share which articles you had used AI for? I think it'd best to draftify them so you can work on them before moving back to mainspace (instead of removing or deleting, see WP:LLMPROD). To echo M kuhner above, people often underestimate how problematic LLM-generated content is, usually we recommend that people stubbify ot rewrite such articles, but that'd of course be up to you. The biggest problem is hallucinations (i.e. WP:V failures), though there are other concerns like WP:NPOV/WP:WTW and WP:COPYVIO. Kowal2701 (talk, contribs) 12:29, 11 August 2026 (UTC)

As stated earlier, I have used ChatGPT for researching sources and some short text for Music of Morocco, Habiba Msika, and Louisa Tounsia. They were reverted to the earlier versions. I have already carefully checked sources for Habiba Msika and Louisa Tounsia and rewritten the biographies with new text entirely of my own. Ever since, I have respected the guidelines for using AI. Munfarid1 (talk) 08:34, 12 August 2026 (UTC)

 Cleanup has been requested.

• Didarrius (talk · contribs · deleted contribs · logs · filter log · block user · block log)

Smoking gun: Special:Diff/1366062588. Has made almost 2k edits, talk page has the usual dab notices. Needs to be nipped in the bud before it goes further. ‑‑gurkubondinn 10:15, 26 July 2026 (UTC)

How is this a smoking gun? There are no messages to the user. ~2026-33562-67 (talk) 18:08, 26 July 2026 (UTC)

In the added 'Horror' section. ‑‑gurkubondinn 18:24, 26 July 2026 (UTC)

Specifically, the ":contentReference[oaicite:1]{index=1}", which is an unambiguous indicator and also suggests that the editor didn't actually read the text they were adding. Gnomingstuff (talk) 00:06, 27 July 2026 (UTC)

they appear to have ignored your warning and notice for this discussion, and their 1 August edits to their sandbox are unsourced and plausibly AI. I guess we wait for more edits to look at Kowal2701 (talk, contribs) 12:33, 11 August 2026 (UTC)

• GodswillSE (talk · contribs · deleted contribs · logs · filter log · block user · block log)

Expanding articles at an inhuman rate. WP:AIREDCAT Special:Diff/1366148547, Information ... has not been widely documented in reliable independent sources Special:Diff/1366148547, nonexistent references Special:Diff/1366147136 Special:Diff/1366102420, duplicating article content and references Special:Diff/1366140398. Jfire (talk) 00:18, 27 July 2026 (UTC)

Looks like it's from this editathon

who would have guessed that awarding prizes for total number of edits would incentivize this kind of thing Gnomingstuff (talk) 00:40, 27 July 2026 (UTC)

Good day. I am on edit-at-on which is improving Depth of article. GodswillSE (talk) 04:34, 27 July 2026 (UTC)

Have you been using an LLM to provide text for these edits? M kuhner (talk) 04:36, 27 July 2026 (UTC)

What is LLM? GodswillSE (talk) 04:58, 27 July 2026 (UTC)

Large language model, also known as AI or chatbot. ChatGPT, Perplexity, Claude, Gemini are examples. Also Grammarly and similar programs. M kuhner (talk) 05:10, 27 July 2026 (UTC)

Started using grok in the late evening of yesterday. But please, I'm going through those pages I improved, to clean it up GodswillSE (talk) 05:16, 27 July 2026 (UTC)

Thank you for your honesty. I really appreciate it.

Use of LLMs (including Grok) is forbidden on English Wikipedia since March 2026, because they put false statements and bad sources into articles much too often. So these articles will need a lot of cleanup. We can provide help and advice. Please don't use Grok here anymore (you can't fix LLM mistakes with more LLM). M kuhner (talk) 05:34, 27 July 2026 (UTC)

I started deleting some articles after cleaning up. Should I hold on or continue? GodswillSE (talk) 05:37, 27 July 2026 (UTC)

These are articles you improved, not articles you created, I think. So it wouldn't be appropriate to delete them. You could "undo" the edits, or look at checking the facts and then writing your own words. For the large number of edits involved, rewriting them all would be hard. I suggest finding a few that you care the most about and trying to fix those. That will give you an idea of the amount of work. We can help with the "undo" for the rest of them.

I know this must be very frustrating for you, and I'm sorry it happened. Thanks again for your honesty and cooperation. I hope other editors here can provide some additional advice. M kuhner (talk) 05:49, 27 July 2026 (UTC)

Hi @GodswillSE, I've draftified Draft:Ebi Yeibo, if you could review the article for WP:V (i.e. that the sources support the content) I'd be happy to move it back to mainspace for you. I'll go through your remaining edits. Best, Kowal2701 (talk, contribs) 16:50, 11 August 2026 (UTC)

Sorry, I misread. They said that they only started using Grok in the evening of 26 July, but edits before had red cats like (7 July) and (14 July) Kowal2701 (talk, contribs) 17:12, 11 August 2026 (UTC)

 Cleanup has been requested.

• Tornolac (talk · contribs · deleted contribs · logs · filter log · block user · block log)

Claims to be posting translations of their own edits to frwiki, but edits like Diff/1366272233 contain multiple OAICITES so these are machine translations. They have indeed contributed to the fr:Volney article on frwiki, but there are no OAICITES present there.

The user has categorically denied using ChatGPT in § July 2026 on their talk page, saying that no AI chatbot was used to write these edits and that they used DeepL as a translation tool and then carefully reviewed the entire text. The exact revision ID of the frwiki page has not been provided, but I could not locate any OACITES in the page history either. ‑‑gurkubondinn 10:46, 27 July 2026 (UTC)

Personally (and in my own opinion), I dont see the value in posting machine translations like this to enwiki, because someone (who does not speak French) that wants to know what the frwiki article says can just query ChatGPT themselves. Saving one ChatGPT output does't really add any value. This should only be read as my own opinion, I am fully aware of the existence of the LLMT guideline. ‑‑gurkubondinn 10:56, 27 July 2026 (UTC)

Hello,

I substantially expanded both articles (Charles-François Dupuis and Constantin François Chassebœuf, comte de Volney) by translating and adapting the corresponding French articles, which I originally wrote.

Much of the material is based on Jean Gaulmier’s biography of Volney and Céline Pauvros’s PhD thesis on Dupuis, both of which are listed in the bibliography. I also added dozens of citations to French academic books and journal articles.

My edits were not AI-generated. I used DeepL and then carefully reviewed and revised the entire text. This is a translation of an already existing, well-sourced article, not text generated by a chatbot.

If you have specific concerns, I would appreciate detailed feedback. General accusations of AI use are difficult to address, whereas concrete issues can be discussed and corrected. Tornolac (talk) 11:00, 27 July 2026 (UTC)

What do you mean when you say that you "translated and adapted" the frwiki article? Are you saying that you localised the article, or did you do any other substantive changes? How did you produce those changes? ‑‑gurkubondinn 11:04, 27 July 2026 (UTC)

I mean that I translated the French article using DeepL and then carefully reviewed the entire text myself to avoid subtle mistranslations. In fact, I did not have to make many changes. I removed or rephrased a few passages that were too specific to the French-speaking context. Otherwise, I stuck closely to the original article.

What I did is fundamentally different from publishing the output of a chatbot. You will not find AI hallucinations in the article. The value added here is making a comprehensive, well-sourced article accessible to English-speaking readers. Anyone is free to ask ChatGPT or Claude about Volney or Dupuis, but that is not the purpose of Wikipedia, which provides verifiable and collaboratively edited resource. Tornolac (talk) 11:15, 27 July 2026 (UTC)

The most time-consuming part of the work was adapting the citation and footnote formatting to conform to the conventions of the English Wikipedia. Tornolac (talk) 11:22, 27 July 2026 (UTC)

I now understand your concern. There is one OACITE tag in the edit I made to the Volney article, not “multiple” as you claimed, despite the dozens of references I added.

That single tag is there because I was trying to understand how the template worked. At one point, I asked ChatGPT to explain how to format citations and references on the English Wikipedia.

There is no OACITE tag at all in my edit to the Dupuis article. Tornolac (talk) 11:59, 27 July 2026 (UTC)

The OAICITE markers have nothing to do with references on Wikipedia (or with CS1 templates), they are artifacts from the model referencing back on itself. You previously claimed that no AI chatbot was used to write these edits, are you now confirming that was not true?

Just removing some of the tell-tale signs that draw attention to the problem does not resolve the problem, and edit warring over it does not help either. ‑‑gurkubondinn 12:32, 27 July 2026 (UTC)

Just noting that the user is now edit warring to reinstate their edits across the articles:

• Constantin François Chassebœuf, comte de Volney (edit | talk | history | links | watch | logs)
• Charles-François Dupuis (edit | talk | history | links | watch | logs)

The only difference in the edits is that the obvious OACITE marker has been removed. ‑‑gurkubondinn 12:28, 27 July 2026 (UTC)

I’m not here to wage a war. I come in peace!

Please read the version I proposed. I believe it is more comprehensive and better sourced than the previous one.

If you have identified any specific factual errors, please point them out. The concern with AI-generated content is that it can introduce factual inaccuracies or fabricated information. As I have already explained, the English text was translated from the French article using DeepL and then carefully reviewed by me. Tornolac (talk) 12:39, 27 July 2026 (UTC)

• DeepL is an AI tool and has been since 2024.
• The concern with AI-generated content is that it is not allowed. Gnomingstuff (talk) 19:54, 27 July 2026 (UTC)

  "Editors are permitted to use LLMs to translate articles from another language's Wikipedia into the English Wikipedia, but must follow the guidance laid out at Wikipedia:LLM-assisted translation. Unreviewed or poorly-reviewed LLM translations often result in similar policy violations to content generated from scratch."

  Once again, I translated the French article and then meticulously reviewed the entire translation myself. Tornolac (talk) 20:39, 27 July 2026 (UTC)

  Thanks for the clarification. In future it may help to disclose the use of LLM tools (which includes DeepL). Gnomingstuff (talk) 21:01, 27 July 2026 (UTC)

  I am just not seeing how DeepL would add an OAICITE, or how it would remain after carefully reviewing the output. ‑‑gurkubondinn 21:04, 27 July 2026 (UTC)

  I reviewed DeepL’s translation and made a few corrections, then reformatted the notes and references to comply with English Wikipedia’s system. I tried using ChatGPT to speed things up, but the results were ultimately unsatisfactory. That’s why the marker was still there. Tornolac (talk) 21:13, 27 July 2026 (UTC)

  The problem with that explanation is that you said that no AI chatbot was used to write these edits, which you later backtracked to say that you asked ChatGPT to show me an example of how to format citations (instead of just reading the documentation), and now saying that you tried using ChatGPT to speed things up. I find shifting explanations suspicious and not convincing. First categorically denying that any chatbot was used at all, then it was used to show you an example, and now you are saying that you actually did use it but didn't like it. And it also doesn't explain how the OAICITE appeared (or how you missed it when you carefully reviewed the entire text), because it is not a reference method. It is a bug in ChatGPT, it sometimes "leaks" its internal reference mechanism to the user (who then copies it into Wikipedia). These are not references like Wikipedia references are, these are most likely internal references back to it's own output, that is unrelated to referencing external sources. What would help me here though, would be to see the ChatGPT session where this was copied from. ‑‑gurkubondinn 21:19, 27 July 2026 (UTC)

  You said I used AI to generate content, whereas I used it only for translation, which is permitted, as I have already explained. DeepL is not a chatbot. I have not contradicted myself in any of my messages, and have no intention of continuing this argument indefinitely. Your excessively nitpicking approach is likely to discourage me from contributing to the English Wikipedia in the future. Tornolac (talk) 21:30, 27 July 2026 (UTC)

  I mean, you did. Your whole argument is that you used it to generate content in one specific way that is still allowed. But unless you are now claiming that this translation (the translation itself) is your own work (and not produced by DeepL), then it is still AI-generated. And you are in fact contradicting yourself, because you started out by categorically denying any chatbot use at all. You have twice walked that back now, and I am still unconvinced that they explain the presence of the OAICITE (especially when factoring in your claim to have "carefully reviewed" everything). Others may be convinced by this, which is fine, but personally I remain unconvinced. Like I said, I feel that it would help if you were to share the ChatGPT session where you copied this from. ‑‑gurkubondinn 21:35, 27 July 2026 (UTC)

  I have explained my workflow as fully as I can and have nothing further to add. At this point, I think it is best to let the other editors decide based on the explanations provided and the revisions published. Tornolac (talk) 21:57, 27 July 2026 (UTC)

  Can you show/share the ChatGPT session where you asked for "examples of how to format citations" and where the OACITE originates from? ‑‑gurkubondinn 22:00, 27 July 2026 (UTC)

  To the best of my knowledge, no English Wikipedia policy requires contributors to disclose their private conversations with AI tools. Otherwise, I would be willing to do so. Tornolac (talk) 22:03, 27 July 2026 (UTC)

  And I didn't say that was a policy. I said it would help explaining to me how an OAICITE shows up in the wikitext. ‑‑gurkubondinn 10:45, 28 July 2026 (UTC)

  You’re welcome. Best of luck in combating the misuse of AI by editors. I appreciate the work you’re doing. As for the OAICITE marker, you’ll find the explanation directly below this message. Tornolac (talk) 21:07, 27 July 2026 (UTC)

OAICITE appeared because I asked ChatGPT to show me an example of how to format citations and references using my own text. I did not use it to write the article or generate its content. I only used it to help me format the footnotes and references. In fact, it did not work very well. ChatGPT kept using Lien web instead of cite web, added italics where they did not belong, and often omitted the <ref> tags altogether. I therefore ended up formatting the citations and references myself. I simply forgot to remove that single OACITE tag afterwards.

I hope this clarifies the situation. The content itself was not AI-generated. If you compare the English article with the corresponding French version, you will see that it is a translation of that article rather than text generated by an AI. Tornolac (talk) 12:50, 27 July 2026 (UTC)

For the benefit of other contributors reading this discussion, here is a brief summary:

Gurkubondinn was right to point out that my initial explanation was inaccurate. I denied using AI because I understood the question to mean whether the article was generated from scratch. I did use DeepL to translate the French article that I had written. I then asked ChatGPT to convert the references to English Wikipedia’s citation format. The result proved unsuccessful, and I corrected it by hand. This is why a single “OAICITE” marker remained in my version after I overlooked it during my review. That was my mistake.

The remaining question is whether my actual use of AI complied with Wikipedia’s policies. I believe it did, in light of WP:NOLLM, which states: “Editors are permitted to use LLMs to translate articles from another language’s Wikipedia into the English Wikipedia, but must follow the guidance laid out at Wikipedia:LLM-assisted translation.” It would be unfortunate if my initial mistake prevented an objective assessment.

The proposed revision is available here: https://en.wikipedia.org/w/index.php?title=Constantin_François_Chassebœuf,_comte_de_Volney&oldid=1366392874 Tornolac (talk) 23:42, 27 July 2026 (UTC)

@Tornolac Per WP:TRANSLATE, If an article has sections with no citations then follow the guidance in UNSOURCED and either provide additional inline sources, or add appropriate citation needed templates as described in UNSOURCED. Content that cannot be verified must not be imported into English Wikipedia. Your proposed edit violates that guideline. Why not try somewhere like Wikipedia:WikiProject Intertranswiki/French? You can build trust with the community and hone your craft. Folks will be happy to see a native speaker offer assistance. Cedarshadow (talk) 00:15, 28 July 2026 (UTC)

The current version of the article is substantially less comprehensive and contains a number of factual errors. For instance, the opening sentence states that Volney came from a noble family, which was not the case. It also states that Thomas Jefferson and George Washington subscribed to Volney’s Christ myth theory, which they did not. There are several other examples.

The proposed revision contains dozens of academic references and over 80 explanatory notes. Unless otherwise indicated, the source is Jean Gaulmier’s 1959 biography of Volney. I have access to a digital edition, so I cannot reliably provide page numbers. I could cite Jean Gaulmier’s book at the end of each paragraph where it is the source, rather than relying solely on its inclusion in the bibliography. Tornolac (talk) 00:34, 28 July 2026 (UTC)

The user has continued making huge edits in Special:History/Constantin François Chassebœuf, comte de Volney:

• Diff/1366827615: c. 19 kB in 3 minutes
• Diff/1366828008: c. 29 kB in 3 minutes after the previous edit
• Diff/1366828327: c. 24 kB also in 3 minutes, following the last edit
• Diff/1366828592: c. 16 kB in 2 minutes

In § Translation of the French article on the talk page, it is again confirmed to be a translation from frwiki (and the article is referred to as a "draft"). The speed precludes any actual translation work and leaves machine translation as the only option. The speed also makes it impossible for any of this to have been read, much less verified, before being added to enwiki, as required by the WP:LLMT guideline.

Personal opinion: I don't see any value in just pasting LLM-translations into enwiki, if someone wants to read an LLM-translation of the frwiki article then they can easily produce it themselves and there is no benefit in saving a static copy of a generation here. ‑‑gurkubondinn 09:34, 5 August 2026 (UTC)

You know, back in June there was a pride month contest. Spanish speakers identified articles about LGBT-themed books that were ready for the rote work of translation. An editor from India was kind enough to do some of it. Everyone was happy. The bottom line is that translated text is held to a high standard and consensus is to use LLMs within that standard. This user is still inserting lines like By then, his last illusions about Bonaparte had vanished without attribution. Would it be worth asking a WikiProject to help them? They seem interested in making it acceptable, as far as I can tell. To date they're close but not there yet. Cedarshadow (talk) 04:40, 6 August 2026 (UTC)

The citation should be added close to the material it supports. It is usually sufficient to add the citation to the end of paragraph, so long as it’s clear which source supports which part of the text. Material repeated multiple times in a section does not require an inline citation for every mention. Tornolac (talk) 06:13, 6 August 2026 (UTC)

It was the same translation that you had removed, I simply added references. As I explained on the discussion page, to avoid making a single edit that was too large, I split it into sections. The 29 or 24 kB were obviously not produced in three minutes. If there is still anything you do not understand, despite our lengthy discussion above, I remain at your disposal. Tornolac (talk) 05:52, 6 August 2026 (UTC)

Immanuelle (talk · contribs), XTools

See Wikipedia:Administrators' noticeboard/Archive357#Request for review of editing restrictions (and all the reports before it), apparently they were rapidly translating articles from languages they didn't speak. ~230 post-Nov 2022 creations (3,800 now-deleted drafts btw). It's unclear to me when they started using LLMs, but it's safe to say that practically all their translations are unreviewed machine translations (see their 1st proper creation, Silken Painting of Emperor Go-Daigo (Jan 2022)) and ought to be tagged with {{Rough translation}}. Kowal2701 (talk, contribs) 13:02, 27 July 2026 (UTC)

We need a proper {{machine translation}} template. ‑‑gurkubondinn 15:56, 28 July 2026 (UTC)

I've asked which translator they used, I assume we can then use the date that translator switched to LLMs as the cut-off point Kowal2701 (talk, contribs) 18:35, 28 July 2026 (UTC)

I do not believe that would be a good assumption to make per a past ANI, the draft rsjaffe seems to be talking about is Draft:Aspirational Rationality per , and it is not a translation. Also, Immanuelle made extensive use of LLMs at Abstract Wikipedia as recently as three months ago, they started early and never stopped. fifteen thousand two hundred twenty four (talk) 19:06, 28 July 2026 (UTC)

Thanks, Takuhadachiji-hime (17 Jan 2023) was created from scratch and has content not in the source that appears hallucinated, and they actually removed some content saying not in the source . There's 2 or 3 creations before that which could plausibly be LLM-generated, I'll just tag those with {{AI-generated}}. Kowal2701 (talk, contribs) 19:27, 28 July 2026 (UTC)

The edit summary of (March 2023) is AI generated starting point Kowal2701 (talk, contribs) 20:34, 28 July 2026 (UTC)

This user made a lot of articles on proto-indoeuropean gods. I took some to deletion, but left others. The subject is largely synthetic, although there are some sources that at least speculate about some of them. These were popping up in 2023 I think. Please don't let my edits on some of these pages disuade anyone from starting a deletion or merge discussion on these. Sirfurboy🏄 (talk) 20:49, 28 July 2026 (UTC)

I've done the LLMPRODs. I was quite conservative with them, she did a lot of merging/splitting and I just tagged some with AI-generated without determining the origins of everything. Some also only had signif. edits from an IP which could have been her, idk. So to the poor sod that gets to cleanup up her remaining contribs, you may want to consider LLMPRODing. Translations pre-Jan 2023 still need tagging with {{Rough translation}} Kowal2701 (talk, contribs) 21:28, 28 July 2026 (UTC)

I've been working on checking/retranslating their translations of Japanese articles in the Shinto field for years and still have a long, long way to go, but let me know if there's any way I can help. A lot of their articles/contributions still aren't tagged with rough translation tags, partially because I haven't had time to actually confirm that all of them are translations and not just bad articles originally made in English.

Also, I'm going to be honest, I don't think a lot of these articles were necessarily LLM generated. They have very poor English in many cases, which you would not expect of LLMs. Kamuo Ichihime for example is a horrible article and on my list to do something about, but it doesn't read like an LLM to me.

Though, if the solution is to stubbify some of these with like one line of decent, verified content, then I'm fine with that too (but I'm not sure I can get to all these in 5 days, especially since I haven't "found" all of them yet...) Erynamrod (talk) 09:42, 29 July 2026 (UTC)

Yeah honestly, going through them, some didn't seem like out-and-out generations, but I assume they habitually used LLMs to at least assist them (also given their self-identification as a transhumanist). The quality is terrible regardless, and presumptive removal probably makes sense even if it's not because of LLM use. Yeah, stubbifying is probably safest and easiest, feel free to userfy any articles you fancy if the timeframe is too short. AFAICT they usually disclosed whether an article was a translation. You probably know, but you can see a list of their creations by clicking the XTools link above. We can notify WP:JAPAN to see if anyone wants to help (I'll probably do this for WP:MYTH as well) Kowal2701 (talk, contribs) 10:33, 29 July 2026 (UTC)

Sounds good, I'll start on that. Erynamrod (talk) 15:57, 29 July 2026 (UTC)

If I could ask a question since I'm not quite sure of the process, we know that Immanuelle used a lot of machine translation (likely LLM translation when it became available) and we also know that she did not have the Japanese language skills necessary to review those translations. Assuming she didn't create the article originally, is our only option to tag those as rough translations, or is there a basis for removing that content since WP:LLMT requires LLM translations be reviewed? I've been working on fixing her translations for years and, while stubbifying all these articles will seriously lighten the load, I'm still looking at years worth of work. Being able to cut out obviously bad information (especially considering another skill she lacked was choosing well-written Japanese articles to translate) would be a big help. Erynamrod (talk) 16:11, 31 July 2026 (UTC)

Yes, unreviewed or poorly-reviewed LLM translations are still subject to all the same problems as content generated from scratch and can be presumptively removed. I'd support presumptive removal even if they weren't using LLMs (see WP:LUGSTUBS for some precedence on this) Kowal2701 (talk, contribs) 16:19, 31 July 2026 (UTC)

User:Haoreima

Haoreima (talk · contribs · deleted contribs · logs · filter log · block user · block log)

Moved from parent thread § Immanuelle

 – ‑‑gurkubondinn 16:29, 30 July 2026 (UTC)

Can someone look at the contribs of Haoreima (talk · contribs) (XTools), they worked a lot with Immanuelle and have creations like Kowal2701 (talk, contribs) 20:18, 28 July 2026 (UTC)

Not certain, but I'll say likely per the contents of and the way they were cited later. fifteen thousand two hundred twenty four (talk) 17:53, 29 July 2026 (UTC)

they've also removed a lot of the AI-generated tags . Would create a new report but this page is already too big Kowal2701 (talk, contribs) 14:54, 30 July 2026 (UTC)

from March 2024 triggered filter 1030 for having fbclid (meaning it came from Facebook? Won't lie, idk what this means) Kowal2701 (talk, contribs) 16:52, 30 July 2026 (UTC)

Yes, fbclid is a facebook tracking id, it doesn't indicate LLM use. fifteen thousand two hundred twenty four (talk) 17:15, 30 July 2026 (UTC)

btw Haoreima is the author of those SEWP articles that were copied over Kowal2701 (talk, contribs) 20:57, 30 July 2026 (UTC)

Manipur State Kala Akademi, Thougallon, Luwang Huiningsumba, Toibi Leima etc. are 100% AI according to GPTZero for Wikipedia, but the scores are generally pretty inconsistent Kowal2701 (talk, contribs) 21:15, 30 July 2026 (UTC)

Also check Meitei novel, Thangyi Taobung Khongdouba, Akong Langmeisu, Loibi Tharangamba, Thongngak Lakpa, Thongpu Shamuknucha, Yaithing Konu, Tengleimaka Amba, Khoiri Tonjaosu, Akangsu, Chimarangba, Haotak Laiba, Yaiphurakpa, Khayoiron Tangja Samphabi. Even if they are transferred from Simple English pages, the creators are the same. Astra Travasso (talk) 15:34, 6 August 2026 (UTC)

Also see, Haoba Athouba, Haokhong Sinaikhu, Kangchin (Khuman warrior), Khuman Tongbu, Thongbu Wainucha, Thongpu Shamuknucha, Namun Chaobi, Khuman Kwakpa Litonpanba. Astra Travasso (talk) 18:13, 6 August 2026 (UTC)

Special:Diff/1322357604 from Nov 2025 appears to have chatbot communication in it. It's at this point we should message them and ask, but I don't fancy that as I doubt it'll go well Kowal2701 (talk, contribs) 19:44, 31 July 2026 (UTC)

I reviewed that diff and saw no evidence of chatbot communication. The comments are copied from {{Infobox recurring event}}, and the reason it was flagged for AI-generated citations was the presence of example.com which 1346 flagged at the time. fifteen thousand two hundred twenty four (talk) 19:55, 31 July 2026 (UTC)

they've been warned for AI use and had autopatrolled revoked at SEWP (simple:User talk:Haoreima#Making articles with AI), the most obvious articles are slated for speedy deletion. They were all created between June and July 2025, and were copied to enwiki in February and March 2026.

• Manipur State Kala Akademi
• Luwang Huiningsumba
• Toibi Leima
• Khayoiron Tangja Samphabi
• Yaiphurakpa
• Akong Langmeisu
• Loibi Tharangamba
• Khoiri Tonjaosu

(July 2026) has what appears to be chatbot communication (the editorial notes are different to what's at simple:Template:Infobox organization) Kowal2701 (talk, contribs) 14:39, 12 August 2026 (UTC)

Also removed some LLMPROD tags, see User talk:gurkubondinn § A misunderstanding. ‑‑gurkubondinn 14:55, 12 August 2026 (UTC)

(worth waiting to see how they respond to the warning on SEWP before contacting them) Kowal2701 (talk, contribs) 17:03, 12 August 2026 (UTC)

 Cleanup has been requested.

• Haseeb_Manj (talk · contribs)

Repeated warnings for LLM use, including one that rose to the level of G15, since 2025, the latest being 9 days ago. They only seem to have acknowledged the first warning, to which they gave what appears to be an AI-generated response, and a) do not appear to have stopped at all, and b) are actively and contentiously restoring their AI-generated content. Gnomingstuff (talk) 04:05, 28 July 2026 (UTC)

I've turned two of their articles into redirects:

• University of Kamalia: had a source that didn't exist and was never saved on archive.org
• Khyber Pakhtunkhwa Football Association: had lots of visually obvious AISIGNS like AIBOLD and AILIST.

Also restored Cadet College SwatCadet College Swat to the redirect by BookishReader, who has taken another one of their articles to AfD for redirection. ‑‑gurkubondinn 15:21, 28 July 2026 (UTC)

from August 2025 has markdown, not that that's terribly helpful now Kowal2701 (talk, contribs) 19:38, 31 July 2026 (UTC)

Support LLMPRV, their first creation was G15'd and the next one looked like with obvious chatbot output in "Initiatives and vision" section. They haven't edited for a few weeks but doesn't meet the def of inactive (1 month) Kowal2701 (talk, contribs) 19:55, 11 August 2026 (UTC)

Support LLMPRV. M kuhner (talk) 19:41, 12 August 2026 (UTC)

• ABehjat (talk · contribs · deleted contribs · logs · filter log · block user · block log)

Posted an AI generated question about their draft to the Teahouse. A look at their contribs revealed this edit which contains communication intended for the user (they added a comment sfter one of their own refs saying "URL unverified; confirm before publishing". Athanelar (talk) 08:41, 28 July 2026 (UTC)

See also. I believe "Yes" would have sufficed. ChompyTheGogoat (talk) 18:45, 28 July 2026 (UTC)

@Athanelar, as you requested:

• Gordon_Myse
• Ric_Wake - which you reverted, but I can vouch for the sources as I've reviewed them

These three were my biggest efforts so far because it required a lot of newspaper sourcing and journal sourcing:

• Charles_Malfray
• Ladang Geddes
• Jesus in the Baháʼí Faith

The rest were primarily manual efforts with LLM used for grammatical fix. ABehjat (talk) 19:32, 28 July 2026 (UTC)

If it helps, I'm willing to zip all my source documents as well (both in original and .txt format) if it helps with the audit/review process. Sorry again for this :( ABehjat (talk) 19:37, 28 July 2026 (UTC)

ABehjat, I don't know how that would help; this isn't about signaling a problem with sources. Drmies (talk) 19:50, 28 July 2026 (UTC)

For sure - @Athanelar provided few examples of other AI articles and I saw in their Talk conversation about difficulty of confirming the sources and validating the article. Thought it would be helpful perhaps to share the raw sources if it comes to that point. ABehjat (talk) 19:59, 28 July 2026 (UTC)

That is not the point where we are at. The point is you were using AI, unwisely but also inattentively. Drmies (talk) 20:17, 28 July 2026 (UTC)

I believe it's their good faith attempt to assist with cleanup. ChompyTheGogoat (talk) 00:48, 29 July 2026 (UTC)

Hi @ABehjat, I've draftified Draft:No Mana in case you wish to work on it, and Jesus in the Baháʼí Faith looks like it will get draftified as well (usually we delete such articles, see WP:LLMPROD). The biggest issue with LLM use is hallucinations (i.e. WP:V failures), though there are other important issues like WP:NPOV/WP:WTW and WP:COPYVIO. Usually we recommend that people stubbify or rewrite LLM-generated articles from scratch since it's often very time-consuming to attempt to preserve such content, but it's up to you. Let us know if you have any questions or need any help with anything, thanks. Kowal2701 (talk, contribs) 17:11, 12 August 2026 (UTC)

Thanks for the notice.

I wasn't aware of WP:LLMPROD. I can see it started in April 2026, but at this stage I guess I've seen the implementation of it across all my articles. I'm waiting for Jesus in the Baháʼí Faith to be dratified before I create a stub from it. I'll do the same for Draft:No Mana.

For what I've done so far, I'm not concerned about hallucinations and I wrote how I've used LLM here: Talk:Jesus_in_the_Baháʼí_Faith: I sourced all citations/sources, OCR'd them, package them into a Database, and then only query against this database which returns sentences from the sources. The only thing the LLM is basically doing after that is paraphrasing what the article is saying, and that's where I spend time reviewing to ensure interpretation is aligned with that the citation was describing. I've also built my equivalent of Earwig/COPYVIO checker as well to ensure it does not plagiarize the content. NPOV/WTW is a fair point and I've noticed from articles like Jesus in the Baha'i Faith and Social Justice in the Baha'i Faith how it took on a different tone. Anyway, not that it matters anyway because LLM is strictly not allowed regardless of how it's used.

I'm not opposed to stubbifying, but is there a way I can work with someone who is doing AI Article cleanup (not the AfD/AI check/part, but like how do you 'de-AI' an article into stub). In some of the past discussions/Village Pumps I've read it was mentioned there were cleanup efforts before the current WP:LLMPROD policy was in place who worked on fixing articles. ABehjat (talk) 19:49, 12 August 2026 (UTC)

Thanks, that sounds good. If you comment here when it's finished, someone can review it, if that's what you're asking? If you want someone to collaborate with on the article, you can try Wikipedia talk:WikiProject Bahá'í Faith or WP:MNB? Sorry but I think people here might be too busy to help out with article writing.

Before WP:LLMPROD was passed, AI-generated articles were usually draftified (and drafts get deleted after 6 months, so it was functionally the same). Normal edits were also usually reverted, but people had to review each edit rather than assume they violated WP:PAGS. We've basically never had someone responsibly and habitually use LLMs to generate content that didn't violate PAGs as far as we know (Esculenta was seen as that until they were blocked) Kowal2701 (talk, contribs) 20:14, 12 August 2026 (UTC)

Thank you and I appreciate the help! Athanelar has helped review one of my articles in AfC that I stubbified AfC Help Desk, but I haven't gotten any feedback since or whether it's good to go.

Once Jesus in the Baha'i Faith is in draft space, I'll re-write it and give an update here. Same for No Mana. I tried WP:BAHAI, I found a few (two only) who helped make edits in my sandbox (including for the one in AfC queue and Jesus in the Baha'i Faith). ABehjat (talk) 21:26, 12 August 2026 (UTC)

Thanks again for sharing Esculenta's experience, and I found the outcome sad. I reviewed some of their work and they're quite well written. The issues raised for Esculenta's case is similar to that in the Village Pump discussion I found here describing how the AI rulings are incredibly restrictive for scenarios where available experts are far and few (OP calls out the lack of sufficient professional medical experts who wish to improve medical works and documents in Wiki and how the editor wishes to use AI to scale their efforts).

To follow up, I found this statement below interesting: If errors continue to slip through at this rate, LLM use doesn't actually save labor, it just transfers labor from you to other editors, as any time savings for you are offset by the many hours the rest of us have to spend reviewing your content. One of my activities here is clicking on Random Article button and hoping to see old articles and references that I can fix or edit. And this reality of work being handed off to others is already the case whether one is using AI or not where I see articles that had full sections without citations, are clear WP:PROMO violations, or get vandalized. Those are all efforts and errors made by previous editors/passer-bys that eventually got fixed/addressed by me. It would've been ideal if it were addressed when the editor wrote them, but this is how Wikipedia works - iteration and iteration of revisions and edits to bring the best out of knowledge.

Anyway, the stick used against AI is strict and painful for contributing Wiki editors (I think you've heard this boring refrain from every person here 😅). But perhaps the fifth pillar of Wikipedia would be brought into account when considering these realities shared above + points from Village Pump and more (perhaps CIR enforcement like how some features of Wikipedia are limited to extended editors).

Sorry for this long message, but seeing that you contributed LLMPROD, I didn't know who else I can discuss this with, and perhaps I'm missing something in all this. ABehjat (talk) 23:35, 12 August 2026 (UTC)

The idea of giving experienced editors permission to use LLMs, on the grounds that they will consistently produce acceptable articles, has already been tried. Results have not been good. In the archives of this noticeboard is a case where an experienced admin with excellent LLM skills thought they could make articles this way...but the articles still had source/text mismatches and bad sources, and had to be deleted.

There are two articles often touted as successes (they are often mentioned at the Village Pump, which is where people discuss this) but two out of tens of thousands of LLM articles we've seen is not good enough.

I have fixed some LLM articles and have come to the conclusion that more hours of work must be spent fixing them than it would have taken to write a human article in the first place. I spent much longer trying to save my first LLM article (written by someone else) than writing my first human article. LLMs only seem like labor-savers because they can transfer the labor from the original author to the clean-up team. The clean-up team understandably does not appreciate this. (My first human article is so far my only one, because I keep getting involved in cleanup.)

I remember with pain a very experienced, autopatrolled editor who admitted the LLM use but asked for an evaluation of his proudest, most carefully checked article. I found a source/text mismatch in the second source I looked at. This cost us what could have been a very interesting article (if it wasn't full of small falsehoods) as well as a very interesting editor (he quit). Really tragic; I wish I knew how to prevent it. M kuhner (talk) 05:20, 13 August 2026 (UTC)

Thank you for your input!

I agree LLM out-of-the-box botches up sources greatly. Recent evolution of practices around information retrieval and recursive self-improvement in AI has done a good job in terms of patching these issues in my case. My primary work area is within the realm of music taxonomy and metadata, so I've been building tools within my domain to ensure to return actual facts (and verify facts). From what I've seen so far, the source information retrieval capabilities are good, but what I'm seeing problematic remarkably is the tone and voice. When the LLM is bound by the 'world' of the sources it's given, it tends to absorb its voice. So a heavily editorialized news about Shakira for example, despite it pulling core facts, tends to weave peacock-like/biased-like voice into the article/information retrieved.

Drawing on your experience, perhaps the best outcomes is a randomized buddy system. In this case, let's say if there's a collab on an article creation, it would minimize some of the challenges you raised because it can be flagged before it gets worse. Apologies for being nosy, but I saw that one of the articles you're working on is Daily Monitor that you shared you wanted to clean. I'd be happy to give a walkthrough via Sandbox to validate against those sources that are present, and then rebuild the article based on confirmed sources/validated sources. ABehjat (talk) 20:42, 13 August 2026 (UTC)

 Cleanup has been requested and is being tracked at /2026-07-31 Re000searchist, 290 pages need review.

Re000searchist: reporting for undisclosed LLM usage. It's especially concerning that they've appeared to use LLM-generated content to WP:GAME the edit count requirements for extended confirmation, allowing them to dump this LLM content into the Arab-Israeli conflict CTOP, among others.

Their edits do not have a single smoking-gun giveaway like an unremoved LLM instruction or a ChatGPT utm source. (Update: I missed it, they actually do; see below discussion). What they do have, on the other hand, is an edit history over 1000+ edits that is almost exclusively of the format produced by Claude.

Major tell, citation clustering and synthesized paragraphs with machine-written edit summaries. LLMs love to write a 2-5 sentences paragraph, in a style of "X source said Y" (because it thinks it needs attribution when a human would just be direct) describing reports about something that happened, rather than the thing itself; then cite the entire paragraph at once with an ambiguous reference at the end, and an overly descriptive edit summary. Selected examples:

• This edit to Tariffs in the second Trump administration added a five-sentence paragraph concerning a lawsuit by two small businesses. The paragraph combines the identities of the plaintiffs, the court, several legal arguments, and the litigation history of the plaintiffs' lawyers. Two fully formed citations, to The Guardian and The New York Times, are placed together only at the end of the paragraph rather than against the particular statements each supports. The edit summary was:

  Added information about the lawsuit filed by two U.S. small businesses against the July 2026 tariffs, citing The New York Times and The Guardian sources.

• Again done here: This edit to the same article.
• This May edit to Military applications of artificial intelligence included the following indirect attribution:

  Some reports stated that AI-assisted systems were used to assist in identifying and tracking senior Iranian officials during Israeli military operations in the 2026 Iran conflict.

  . This is how LLMs write an attributed statement, not how humans do. Again,

  Researchers and human rights organizations described Israeli AI-assisted targeting systems as relying on large-scale analysis of surveillance footage, drones, mobile devices, and communications metadata...

  -- a human would indicate which researchers or human rights organizations, and quote the description directly. LLMs try to broadly genericize from a specific example in exactly this manner. The edit summary was:

  Added sourced information on reported AI-assisted targeting systems and related human rights concerns regarding potential misidentification risks.

• This 19 June 2026 edit summary was:

  Added 2026 reporting on the use of Anthropic’s Claude in the Iran–US conflict, including its role in military intelligence and targeting support.

  Again, humans wouldn't be writing about "2026 reporting" in 2026. That's an indication that the LLM was told to search for reporting from a certain date. A human would have named the sources (notwithstanding this, the sources included unreliable sources like "Cryptobriefing" that would never have passed human review.)

More examples of indirect attribution rather than directly stating:

• Some reports stated that...: diff
• Researchers and human rights organizations described...: diff
• critics questioned the justification..., without identifying the critics in the prose diff
• based on Western media reports in an edit summary diff -- implying that this is *not* a Westerner writing this. A human, and a Westerner in particular, would not need to specify "Western media".
• According to official statements... and similar generalized framing in additions to Special Headquarters for the Development of Nanotechnology.

This one was so egregious, I reverted it and warned them about LLM usage. The response was as predictable as you'd expect: an LLM-generated reply that denies that they were an LLM, in overly official language that no human would actually use. Thank you for your message. The content was based on reliable published sources, and my intention was to summarise those sources in a neutral way. If the wording resembled AI-generated text or did not meet Wikipedia's standards, I appreciate the feedback and will revise my editing approach accordingly.

It appears obvious to me that this is a non-English speaker using LLM to bulk translate articles.

• Verse of the Mustad'afin was created as a 330-byte translated article in revision 1364948893. After several small edits, revision 1364951195 added 6,750 bytes in one edit. The resulting material had a complete section structure, uniformly composed prose, indirect formulations such as According to some reports, and multiple paragraphs in which references were grouped at paragraph endings.

• Special Headquarters for the Development of Nanotechnology was created by translation in revision 1336005599. Revision 1336549587 added 3,561 bytes to Key Events in the Development of Nanotechnology in Iran in one edit. Revision 1336552285 added another 1,702 bytes to Key Figures in the Development of Nanotechnology in Iran. The additions use the same tells as above: a uniform section-by-section structure, generalized attribution, and paragraph-ending citation clusters. But additionally, they also show that the user is somehow technically proficient and familiar with Arabic to be able to paste complex foreign language citations in -- but somehow not familiar enough to realize that WP:TASNIM is a deprecated source, or that most of the sources would never pass a reliable source review.

Seeking a consensus here that the user is, despite their denial, in fact using LLMs. I'm concerned this may be a hefty amount of cleanup. ⇒SWATJester ^{Shoot Blues, Tell VileRat!} 15:54, 28 July 2026 (UTC)

Their edits do not have a single smoking-gun giveaway like an unremoved LLM instruction or a ChatGPT utm source – courtesy of Special:AbuseFilter/1346. Trump’s social media use has been credited with transforming how political figures communicate, emphasizing direct access to millions of followers and bypassing traditional editorial filters. His outspoken style and rapid posting rhythms on Truth Social have drawn attention ... , yeah that's an LLM user. fifteen thousand two hundred twenty four (talk) 16:58, 28 July 2026 (UTC)

Lol well there you go, thanks! I probably should have qualified what I wrote above as being "that I found in my sampling".⇒SWATJester ^{Shoot Blues, Tell VileRat!} 21:06, 28 July 2026 (UTC)

I must have looked into this user at some point, because one of the links to the filter logs was purple for me. Don't remember why I was looking into the user though. ‑‑gurkubondinn 17:25, 28 July 2026 (UTC)

This is an LLM user, and has been making substantial use of LLM for at least the last six months, in my view. ChrysGalley (talk) 08:33, 30 July 2026 (UTC)

See for their edits which have triggered relevant edit filters Kowal2701 (talk, contribs) 19:35, 31 July 2026 (UTC)

They're articles created in August appear human-written, I'll draftify ones before then Kowal2701 (talk, contribs) 19:51, 13 August 2026 (UTC)

They're all translations from Farsi Wikipedia (Special Headquarters for the Development of Nanotechnology, Fatima in the Quran and Hadiths, Verse of the Mustad'afin) so we need to scrutinise them against WP:LLMTRANSLATE Kowal2701 (talk, contribs) 19:57, 13 August 2026 (UTC)

Cleanup efforts are ongoing.

• 004bvan (talk · contribs · deleted contribs · logs · filter log · block user · block log)

Draftified two articles from this user that had noticable AISIGNS:

• Draft:Elastic membrane: markdown in AbuseLog/44748340
• Draft:Academy of Sciences of Turin: created with AIREDCAT in Diff/1359377933

They seem to have tried to create Draft:Pope Leo XIV at some point. There are plenty of AISIGNS in the filter logs for it, like bracketed numbers as faux-references and {{citation needed}} tags, and utm_source=chatgpt.com on reference URLs: AbuseLog/44307449. It was also partially in Italian, and does not seem to have been saved because there is no entry in the deletion log.

Had apparently previously run into this user, because I turned Principle of actionPrinciple of action into a redirect last month. ‑‑gurkubondinn 17:10, 28 July 2026 (UTC)

• The user is adamant (on their talk page) that they did the writing themselves. I do not believe that for a moment. Drmies (talk) 19:36, 28 July 2026 (UTC)

'''''Editor's note:''' The following is a hypothetical citation used to demonstrate the standard of sourcing required by Wikipedia. It does not represent a real publication.'' [hallucinated ref omitted] from 31 July 2025. Unambiguous cut-and-paste LLM user.

More **markdown** from early 2025 etc. The original version of their userpage even contained a hallucinated alternate account. I would suggest applying WP:LLMPRV to their contributions if there is consensus here to do so. fifteen thousand two hundred twenty four (talk) 20:09, 28 July 2026 (UTC)

Thanks. I was wondering about the HistoricRx user earlier, but I forgot to mention or follow up on it. ‑‑gurkubondinn 20:19, 28 July 2026 (UTC)

Is there a template for an LLM block? I think it's time. Drmies (talk) 20:20, 28 July 2026 (UTC)

Most of my LLM blocks are partial blocks that restrict editing in mainspace, where LLM usage is the biggest problem. For that I use {{uw-pblock|area=Article}}. We do have {{uw-aiblock}}, however. I usually find a sitewide block to be overkill, though, unless the rampant LLM usage is taking place in draft space. ~Anachronist (who / me) (talk) 23:30, 28 July 2026 (UTC)

(btw {{uw-llmpblock}} exists), they haven't made any edits since being notified of this discussion, but consensus is still needed for LLMPRODs (or we just wait a couple weeks) Kowal2701 (talk, contribs) 20:02, 13 August 2026 (UTC)

Thanks. I wasn't aware of that template. ~Anachronist (who / me) (talk) 23:37, 13 August 2026 (UTC)

Support LLMPROD and LLMPRV now. M kuhner (talk) 19:45, 14 August 2026 (UTC)

To be clear, I support LLMPRV as well. fifteen thousand two hundred twenty four (talk) 00:17, 15 August 2026 (UTC)

Thanks guys, LLMPRODs are done, there's nothing else needed, just waiting for them to go through Kowal2701 (talk, contribs) 00:46, 15 August 2026 (UTC)

 Cleanup has been requested.

Jarvis20001122222 (talk · contribs · deleted contribs · logs · filter log · block user · block log) Account created a month ago. Very prolific, many articles have been draftified, and they were told off for recreating them in parallel in mainspace. They hit edit filter 1,371 for adding {{lang-el}}, which has been phased out since 2024. Have used sources not matching statements, have put in redcats, have left in HTML comments like this and like that, and have used em-dashes. Oftentimes the draft version and the mainspace version have completely different text, as if they told an LLM "didn't work, try again". I have asked them about LLM use but all they gave me was a thanks. Many articles are at AfD already; would appreciate some more eyes on this. YuniToumei (talk) 21:37, 28 July 2026 (UTC)