Wikipedia:WikiProject AI Cleanup/Noticeboard

From Wikipedia, the free encyclopedia

WikiProject AI Cleanup/Noticeboard

Shortcuts

This page is for reporting issues of AI misuse.

For issues concerning a single article, please refer to the guide on how to deal with AI content before posting here.
You can inform an editor of a discussion involving them with {{AINB-notice}}, but this is not a requirement.
More general discussion about AI use and policies should be brought up at Wikipedia talk:WikiProject AI Cleanup.
{{AIC status}} can be used to create a hatnote keeping track of a case's status.
To view all open cases (including those whose reports have been archived), see Category:WikiProject AI Cleanup open cases.

Start a new discussion

Archives

This page has archives. Topics inactive for 90 days are automatically archived 1 or more at a time by Lowercase sigmabot III if there are more than 4.

Potential LLM generations by Bechamel

Bechamel (talk · contribs) recently submitted the article 'English afternoon tea' for FA review. I came across this article in a happenstance manner, but I noticed numerous problems immediately. I suspect that an LLM was used to create this article. The smattering of dashes, the strange definitive language like Afternoon tea is served at nearly all important official, social, and sporting events – without five o’clock tea, none of these occasions would be truly British, and the presence of an odd, bulleted list concern me. A look at Bechamel's other contributions show a strange variation in writing style from article to article. I asked Bechamel about this, but he denied using an LLM. I am hoping that the veteran editors here may be able provide assistance in confirming if my assessment is correct. Yours, &c. RGloucester — ☎ 23:42, 17 November 2025 (UTC)

Discussing with editor at Talk:English afternoon tea#Problems with this article, requesting other editors here to hold off on disclosing any assessments they've made before Bechamel has a chance to respond, thank you. fifteen thousand two hundred twenty four (talk) 03:06, 18 November 2025 (UTC)

**Markdown** added at English afternoon tea and Venetian window , and removed .

In discussion editor stated they use MS Word but not copilot. After being informed of the markdown issues they stated they couldn't rule out the possibility that Word might be using copilot mechanisms despite copilot otherwise not functioning on their Windows install.

Without evidence of greater disruptive activity, I don't think there is much else to productively do at this time. fifteen thousand two hundred twenty four (talk) 18:41, 18 November 2025 (UTC)

Thank your hard work. I have gone through the article, line by line. A combination of WP:OR of primary sources, grandiose assertions that were not supported by their citations, mistaken page numbers, strange phrasings, and close paraphrasing. This is the first time I've bothered to enter the 'AI cleanup' line of work, and I can't say it was very enjoyable. I appreciate the work everyone does here. Yours, &c. RGloucester — ☎ 08:23, 19 November 2025 (UTC)

They've created more articles with AI signs and had one G15'd a couple days ago. I've left a 'final' warning Kowal2701 (talk) 17:14, 7 February 2026 (UTC)

Stylophore ought to be nominated for deletion or restarted from scratch. Even if it cannot be proven to violate WP:NEWLLM, the entirety of the references (save for a single monograph) is inappropriate. Einsof (talk) 17:54, 7 February 2026 (UTC)

This has not been written using LLM. Rather, it has been translated from ru:Стилофор where it had previously been developed normally. See User talk:Bechamel#Stylophore. I expect a translation tool has been involved. See Wikipedia:Articles for deletion/Stylophore. Thincat (talk) 14:32, 8 February 2026 (UTC)

User:Adamsecretxx

Adamsecretxx (talk · contribs)

Most of the edits that have been made show signs of LLM (most significantly the bolding, and some sources not linking to what they say they link to), but not enough for G15; additionally, most of the articles that have been created seem to be non-notable, but in that frustrating boundary region where it's not obvious whether it seems non-notable because of the LLM or because it is. (Or maybe I'm being too lenient here.) User has also made significant changes to Victor Mayer, a preexisting article, and made half of it look AI generated. Would appreciate more eyes on the contribs. Fermiboson (talk) 22:00, 3 December 2025 (UTC)

Agreed that most/all of their substantive mainspace contributions seem to be unreviewed LLM. Albert Mayer (Davos) is G15-able (looks like you tagged it for A7). I think Marie Fabergé can be G15'd as well, and it is teeming with WP:AISIGNS. Also see Special:Diff/1322820628: I wonder about this editor's English level. I will drop them a note on their talk page. We should create a tracking page as well NicheSports (talk) 23:10, 3 December 2025 (UTC)

Thank you for the comments. I´ll be more mindful firm now on to explain my edits and new page proposals. Adam Adamsecretxx (talk) 00:44, 6 December 2025 (UTC)

+1, as the kids say. I draftified one of their creations a couple of weeks ago for a possible combination of LLM and CIR issues, and now their talk page has lit up with many similar notices and no response from this user. ClaudineChionh (she/her · talk · email · global) 13:48, 4 December 2025 (UTC)

Hi, thank you for the messages.

I´m admitting that I had LLM check up on my texts. I did not know thats not allowed. the research and the original texts were mine. I¨m also not sure what is noticeabale and what not.

MARIE FABERGE For example Marie Fabergé is the wife of one of the most prominent jewellers ...so is she noticeable or not? The entire lineage of Fabergé is written about in basically every monograph about carl Fabergé. It is not up to me to decide.

I do have knack to research topics that are NOT extremely published, so that might be the problem, plus that I used ai to check up on my form and content, I´ll update how I work.

REUCHLINHAUS I did edit one page now. Reuchlinhaus, I hope it looks clean. The German wikipedia already has a page on that subject, so at least in germany it seems to be noticeable, and I kept it very close to what was in the German wikipedia. Thanks for checking my writings and for being lenient with me as an on and off user. Adam Adamsecretxx (talk) 23:27, 4 December 2025 (UTC)

I¨m also not sure what is noticeabale and what not – The relevant policy is WP:NOTABILITY, and the section to pay the most attention to is the one on the general notability guidelines. Do note that the English and German Wikipedias have different standards for notability, so a topic that is notable on one project may not be on the other. (You may wish to visit the Introduction, which can help get you up-to-speed on what is expected from editors on enwiki.)

As a more general note about large language models, they are ill-suited to assist with editing. LLMs are prone to hallucinations, have issues maintaining a neutral tone, and often produce prose incompatible with the manual of style. As these models are predictive, they can predict incorrectly even for simple prompts or requests. Thank you for discussing this with us. fifteen thousand two hundred twenty four (talk) 00:31, 5 December 2025 (UTC)

Yes. I study the guidelines now for the different countries and it becomes more clear. It helps to be sure about what generally all the rules are, albeit jn detail it iS about a group effort. Adamsecretxx (talk) 00:43, 6 December 2025 (UTC)

Your article on Maria Fabergé contains sources that are entirely hallucinated. The ancestry link goes nowhere and the book "Gustav Fabergé and his Dresden Years" doesn't even seem to exist exist; the only results for it on google are the Wikipedia articles for the Fabergés, and gbooks turns up nothing. Your fourth {{cite web}} reference doesn't even have a URL specified. Athanelar (talk) 01:38, 10 December 2025 (UTC)

Hello, Yes the text is largely based on that book with the recent research. It¨s not hallucinated. the link does not go to a real page and I will see that I can find a link that works. Adamsecretxx (talk) 02:06, 10 December 2025 (UTC)

The text is based on what book? The book "Gustav Fabergé and his Dresden Years" doesn't seem to exist. I can't find any evidence of it whatsoever. Can you show some indication that it exists?

How did you end up adding a link to an Ancestry page that doesn't exist? Athanelar (talk) 02:10, 10 December 2025 (UTC)

For future reference, the citation in question is:

<ref name="Skurlov">{{Cite book |last=Skurlov |first=Valentin |title=Gustav Fabergé and his Dresden Years |publisher=Igor Carl Fabergé Foundation |location=Moscow |year=2018}}</ref>

Could you please provide an ISBN or other similar identifier? The title does not appear to exist. fifteen thousand two hundred twenty four (talk) 02:15, 10 December 2025 (UTC)

I suggested above that that article could probably be G15'd, but after seeing this source analysis that probably is now definitely. You want to take it? NicheSports (talk) 02:09, 10 December 2025 (UTC)

Take a look, I've already tagged it. Athanelar (talk) 02:10, 10 December 2025 (UTC)

Nice. Thanks! NicheSports (talk) 02:12, 10 December 2025 (UTC)

It looks like they're still at it (see these edits to Jardines del Bosque and Jardines del Pedregal) despite stating they would stop using LLMs on their talk page. Should this be brought to ANI? Zygmeyer (talk) 21:39, 16 December 2025 (UTC)

I think so, yes NicheSports (talk) 21:42, 16 December 2025 (UTC)

I've never done this before, is there a guide for the procedures I should follow or should I just go for it? Zygmeyer (talk) 22:16, 16 December 2025 (UTC)

There are instructions at the top of the noticeboard. The most important point is to present relevant diffs, which have already been compiled in this thread, the tracking page, and the user talk page. You also need to ensure that you notify the user you're reporting. Twinkle can help you follow all the necessary steps. ClaudineChionh (she/her · talk · email · global) 22:22, 16 December 2025 (UTC)

I misspoke slightly – Twinkle won't fill out a whole report, but it can simplify the process of notifying users. ClaudineChionh (she/her · talk · email · global) 22:44, 16 December 2025 (UTC)

Hi, I see your conversations about my account. What exactly did I do at the pages Jardines del Pedregal and Jardines del Bosque that goes against Wikipedia rules? - Adamsecretxx (talk) 22:49, 16 December 2025 (UTC)

@Adamsecretxx, did you use a large language model/AI chatbot to generate new text for those two articles? I can't speak for @Zygmeyer, but those two edits have a different style and level of English fluency compared to your responses on talk pages, which makes me suspect AI use. ClaudineChionh (she/her · talk · email · global) 23:00, 16 December 2025 (UTC)

Thanks for asking directly. Yes, I did use an AI/LLM tool at times to help rephrase or polish text, especially because English is not my first language. I now understand that this is discouraged for mainspace editing because it can create problems with tone, accuracy, and sourcing.

I’m stopping that practice for Wikipedia edits. I’m reviewing my recent contributions, reverting or trimming anything questionable, and moving any further work to draft or sandbox so it can be checked carefully before publication. Adamsecretxx (talk) 23:13, 16 December 2025 (UTC)

I was going to respond to this, but now that a report has been made on Administrators' noticeboard/Incidents, let's continue the discussion there. ClaudineChionh (she/her · talk · email · global) 02:20, 17 December 2025 (UTC)

Ekabhishek

I was surprised to find all nonexistent links in this large addition. The editor Ekabhishek (talk · contribs) reverted themselves after some nagging and have stopped editing for the last couple of weeks. I found more instances of llm generated text added wholesale by them.

Going back in their contributions list up to July 2025 and looking at large edits, I've boldly (they are an admin) removed their edits which weren't backed up by the refs, from Meenakshi Jain, Mrs (film) and Dastak (1970 film) as original research. But I keep finding more - like [this edit] where the last ref link is a page not found - as I go further back. Legospy (talk) 06:41, 7 December 2025 (UTC)

Yikes, that first example is textbook AI text. I've seen NPPs and autopatrolled users throw LLM text around, but admins? That's new. I'm going to see how deep is this. AlphaBetaGamma (Talk/report any mistakes here) 03:43, 8 December 2025 (UTC)

Yeah that sure has a few WP:AISIGNS. And it had a bunch of broken links? That is a G15-level of proof of unreviewed LLM content. I also see that Ekabhishek chose to not answer three questions on their talk page from Legospy about LLM usage. @AlphaBetaGamma I will add an AINB notice to their talk page and tell them we need to hear from them. Do you want me to wait until you've dug more? NicheSports (talk) 04:04, 8 December 2025 (UTC)

Looks like they haven't edited since November 18 so no rush on the AINB notice. I'll wait until you've looked into this. NicheSports (talk) 04:23, 8 December 2025 (UTC)

Would it not be wise to bring it up at WP:AN? I would imagine an admin editing disruptively like this should at least warrant a reconsidering of their access to the tools. Athanelar (talk) 04:44, 8 December 2025 (UTC)

They haven't edited in weeks so there's no rush imo. Let's look into their edits first? Also too early to talk about recall. NicheSports (talk) 05:04, 8 December 2025 (UTC)

I know this is a blatant assumption of bad faith that should usually be avoided, but if they simply return to editing and fly under the radar with their edits again, wouldn't the damage be severe? I can't feel safe until they give us a non-dodgy explanation/declaration. I really do not want an another recall drama (it has already depressed me a lot) unless the admin in question causes a WP:ANI level of disruption and refuses to communicate.

Speaking of digging through, I was seeing if I needed to blast off any faulty articles, but I don't think there is an urgent need compared to their edits. they have made around 11 articles in 2025, and all of their other creations predate ChatGPT. Stuff like Bina Ramani had some eye-raising issues with promotional words (pionneering with no cited source actually saying that word). However, their 11 creations in 2025 probably needs some duct taping, so cleanup is requested. AlphaBetaGamma (Talk/report any mistakes here) 05:45, 8 December 2025 (UTC)

Judging by their creation activity, we might need to scan every one of their mainspace edits after the admin's return from inactivity in 2024ish, which is easily over a thousand. Holy hell... AlphaBetaGamma (Talk/report any mistakes here) 05:52, 8 December 2025 (UTC)

Sigh. Thanks for looking into this. I agree with Athanelar that this probably needs to go to AN if Ekabhishek doesn't return and address these issues soon. If we go there, we should try to quantify the extent of the LLM misuse and bring representative diffs - any chance you can keep working on that? Added: Fifteen's diffs below look compelling. We could also ping an admin to get advice. Newslinger comes to mind. And if at any point Ekabhishek returns to editing without engaging with us I think we should go to AN right away. NicheSports (talk) 06:34, 8 December 2025 (UTC)

I've notified Ekabhishek of this discussion at User talk:Ekabhishek § AI cleanup noticeboard discussion. — Newslinger talk 06:40, 11 December 2025 (UTC)

"Pioneering" is probably a rendition of descriptions like this: "What really makes Bina Ramani's transformation of Hauz Khas Village so remarkable and memorable is that she did much more than just reviving a neighbourhood, she birthed the vibe of the place. She turned ruins into rendezvous spots, history into heritage, and then made the heritage turn into a social movement. It wasn’t long before the area became synonymous with indie fashion, experimental art, and late-night conversations under fairy lights." (The Godmother of Hauz Khas And Its Transformation Of The '80s (cited in th Bina Ramani article))

Another article that uses the word "pioneer" to describe Bina Ramani is this one.

The first case Legospy cited and the examples fifteen thousand two hundred twenty four analyzed sound like llm/hallucination though. Deamonpen (talk) 08:16, 8 December 2025 (UTC)

Special:Diff/1318901237 – both quote=s appear to be hallucinations
Special:Diff/1307138307 – contains markdown-style footnote refs: [^3][^4]
Special:Diff/1295888253 – The Sabarmati Report And Hindi Journalism: A Political Narrative Disguised As Cinema does not exist, The Sabarmati Report movie review: Refreshing and fast-paced does not exist, The Sabarmati Report Review: Vikrant Massey shines in this ideological drama does not exist, The Sabarmati Report Review: A riveting but one-sided narrative does not exist (here is what the homepage of "thebrew.in" looked like in May 2024 when the supposed article was published).

This was a surface-level look. I've not included any snippets of LLM prose which often overstep the bounds of WP:NPOV, nor utm_source indicators. fifteen thousand two hundred twenty four (talk) 04:36, 8 December 2025 (UTC)

They're unambiguously using AI for some things -- here is an early sandbox edit that contains attributableIndex json all over the place Gnomingstuff (talk) 22:13, 8 December 2025 (UTC)

In some recent additions, I may have used ai for copy editing and improvement, but rewritten to avoid such a scenario, though I admit some mistakes would have crept in. So if you feel like it is too much in some places, please go ahead and remove or rewrite, no issues. Since we are all here to BUILD wikipedia and in good faith. I will also try to rewrite whatever it is possible next. Thanks --Ekabhishek^talk 14:02, 17 December 2025 (UTC)

I may have – Why the uncertainty? An LLM has definitively performed more than copy editing as evident by multiple hallucinations. Regardless, LLMs are unsuited for copyediting, they are predictive models and can predict wrong even with very simple prompts, leading to the introduction of hallucinations, biases, unencyclopedic prose, and non-neutral constructions.

Acting in good faith is half the equation, competence is the other half and requires editors understand the tools they use to ensure that, at a minimum, outright falsifications are not blindly copied-and-pasted into the project. fifteen thousand two hundred twenty four (talk) 00:09, 18 December 2025 (UTC)

User:MaynardClark

Fairly prolific with edits, having the usual WP:AISIGNS all over the place going back to 2023; but what elevates this to an actual noticeboard post is the talk page comment here, User_talk:MaynardClark#ChatGPT, specifically the last paragraph which suggests that review isn't really being done. Gnomingstuff (talk) 23:20, 16 December 2025 (UTC)

User:Feizan

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

Many telltale signs of AI. Huge edits using curly quotes and awkward tone, while talk page responses are sloppy. Thank you! Magnolia677 (talk) 15:31, 18 December 2025 (UTC)

I had to deal with some rough edits to Porcupine but left the user alone due to inactivity past the end of November. More attention would be helpful now that they are editing more again. -- Recon rabbit 18:52, 18 December 2025 (UTC)

User:UrusHyby

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

This editor first began adding AI-generated text to Litvinism with all the obvious signs. See for example this edit which is obviously AI-generated. Despite the warnings on their talk page, they have continued to make large additions to a range of articles with signs of AI use. It seems possible that they have now added AI-generated text along with fictitious references that do not directly support the statement. This is evidenced by some of the references having no page numbers or a long range.

See for example this edit to Battle of Orsha where they included a link to a page that does not exist (I could not find any such article that exists). After this was reverted, they restored the changes but added more sources that do not directly support what is being said. Similar issue at Marc Chagall. In this edit they added more sources and another editor tagged one of the statements with the following reason: Editor, who added this and two other refs, provided a non-existent quote in one of them, while twisting some other already present one in spite of quotation marks around it. Mellk (talk) 11:38, 23 December 2025 (UTC)

As another example, UrusHyby made this edit to Kievan Rus' and one of the statements they added was: It was a multi-tribal, loose federation where the Principality of Polotsk (in modern Belarus), the Principality of Kiev (Ukraine), and the Novgorod Republic (Russia) acted as key, often competing, centers of power. I looked at the cited source (they specified pages 10 to 15) and I could not find those specific states being mentioned as centers of power. For example, the Polotsk principality is only mentioned in the context of post-Soviet Belarusian intellectuals turning to it in "their search for the origins of their nation in the same historical period" (p. 12). Mellk (talk) 12:33, 23 December 2025 (UTC)

They've been using an LLM since they resumed editing in September, is LLM nonsense. All references at their most recent creation, Mikita Melkazioraŭ, don't exist.

I've left them a message on their talk page that will hopefully convince them to desist. Cleanup is necessary. fifteen thousand two hundred twenty four (talk) 13:21, 23 December 2025 (UTC)

Hey everyone, thanks for your attention and your strive of having wikipedia the best source of corect information. That i share as well. I will copy here the comment i've made on my personal page, stating that I do believe that i follow the core concept of WP that AI is a proper tool, is it used carefully/. That is exactly what i was doing. Similarly the approach of WP:AICLEAN states "The purpose of this project is not to restrict or ban the use of AI in articles, but to verify that its output is acceptable and constructive, and to fix or remove it otherwise.". All the changes i do are goodwill based and supported by sources that i personally handpick.

But thank you for the notice I will be paying even more attention that the links remain intact after the automated proofreading\grammar corrections that i do. Preferences (talk) 14:08, 23 December 2025 (UTC)

All the changes i do are goodwill based and supported by sources that i personally handpick – All of the references you added to Mikita Melkazioraŭ are hallucinated and do not exist, this edit to Bicycle-sharing system added a hallucinated reference, as did this edit to ERM Telematics, as did this edit to Package tracking, as did this edit to Device tracking software, as did...

Nowhere in Wikipedia policy will you find it stated that AI is a proper tool, it is simply not a core concept. Two policies and guidelines which do exist are WP:VERIFIABILITY and WP:DISRUPT. Your editing is in clear continual violation of both. Your automated proofreading\grammar corrections are simply disruptive, and if you continue to perform them I will open an WP:ANI report myself. Stop.

If you lack the ability to proofread and correct grammar yourself, and instead must rely on copying and pasting output from an LLM, then you should not be making those edits at all. fifteen thousand two hundred twenty four (talk) 14:48, 23 December 2025 (UTC)

Even though, I am still confident about the sources I use. In all of the edits you've mentioned, there are proper links 1 for the Bicycle-sharing system, or this in Device tracking software edit I did det your point, as you're right, that some of the links get broken, unfortunately.

Which is a pity, as many of the articles that I improved were in a really poor shape, including a politically-hot one Litvinism that was a subject for deletion. And having good intentions and some understanding of subjects I tried to improve it as well as some others.

I do doubt that your revert of my edit of ERM Telematics article that had WP:PROMO since 2016 made any good, and brought it any to a better state, comparing with my version, even keeping in mind usage of AI tools.

But, I got the point. Thanks again. I will be more careful. Preferences (talk) 15:19, 23 December 2025 (UTC)

Are you sure you are confident about the sources you use? What about this edit, where you provided a non-existent quote "Born Moishe Shagal in Liozna, near Vitebsk, in what is now Belarus" - allegedly from the Jackie Wullschlager's book? There is no such a quote in the book! So where did you find that quote? Alexschneider250 (talk) 20:25, 23 December 2025 (UTC)

In the very same edit you added a ref to the book "Minsk - Wilna - Stationen einer undenkbaren Freundschaft", allegedly by Thomas M. Bohn. Are you sure that book exists? If so, could you provide a link to a web page saying anything about that book? Alexschneider250 (talk) 22:43, 28 December 2025 (UTC)

As one more example, UrusHyby made this edit to Russification adding a quote (with quotation marks around it): "What the Russian bayonet didn't accomplish, the Russian school will.", allegedly by the Russian Governor-General Mikhail Muravyov, from the book "Belarus: A Perpetual Borderland" by Andrew Savchenko, p. 55. But I read that entire page (and neighboring pages too) in that book and I could not find any sentence like that there, not even slightly close in meaning to that quote.

Moreover, that quote is a fake! Mikhail Muravyov never said that. Actually, Ivan Petrovich Kornilov, a Russian official in the 19th century, said this:

Русское образованiе сильнѣе русскаго штыка. Въ какiе-нибудь 4 года русскiя школы сдѣлали болѣе для образованiя народа и ослабленiя полонизма, чѣм войска въ десятки лѣтъ. [Russian education is stronger than the Russian bayonet. In about 4 years Russian schools did more for the people's education and weakening of the polonization than troops in tens of years.]

— Ivan Petrovich Kornilov, Kornilov, I. P.: Russian Cause in the Northwestern Krai (St. Petersburg, 1901), A.P. Lopukhin's Press, p. 147.

This quote is being twisted and spread all around Wikipedia by politically-engaged editors, who are also trying to make it look like it was said about the oppression of the Belarusian language. Obviously, the quote is about using education to oppose the imposition of the Polish language among inhabitants of the Northwestern Krai. Alexschneider250 (talk) 00:56, 29 December 2025 (UTC)

Tracking subpage created, 92 articles appear to warrant further review. While compiling the list, I encountered a multitude of hallucinated references. fifteen thousand two hundred twenty four (talk) 18:35, 23 December 2025 (UTC)

This is continuing, see Special:Diff/1333034795. Most of this does not verify in the source , and whatever does is SYNTH. Very likely LLM-generated. NicheSports (talk) 16:38, 16 January 2026 (UTC)

They have now reinstated this content . I left them a final warning on their talk page NicheSports (talk) 09:56, 20 January 2026 (UTC)

Talk:Arthur Katalayi

Large amounts of likely AI-generated uncivil talk page messages left on this page. I left a comment under the first one, but I decided not to further on, since we shouldn't feed the trolls. ~2025-31416-56 (talk) 13:20, 23 December 2025 (UTC)

Lehrling18 creating AI-generated pages

Hi all,

The user I mentioned above has been creating drafts and revising existing articles (380 edits, account created 22 November) about French WW2 military units (typically French Army corps). Interestingly, he has also participated in talk pages (, , ), with AI-generated responses.. I haven't done a deep dive into whether his sources are hallucinated, but none of them have URLs or are CS1. He has also been adding navboxes to the articles, but they are made from tables + CSS (they won't uniformly update) and occasionally are named differently depending on the article.

Here are some of the articles that he has created. While it isn't as obvious compared to other AI-generated pages, the signs are still there and his talk page participation is damning. In draftspace, when his content gets denied for being AI-generated, is appears that he just regenerates it then resubmits it.

EatingCarBatteries _{(contributions, talk)} 21:07, 24 December 2025 (UTC)

This LLM-generated text has been collapsed and should be excluded from assessments of consensus.

The following discussion has been closed. Please do not modify it.

I'm the user in question. I'm a programmer, which explains my workflow: I use structural skeletons and batch-processing to handle large datasets efficiently. It's not AI.

Regarding the feedback:

Templates & Collapse: I've already started replacing the CSS tables with proper templates and removed the "click to expand" feature to meet MoS standards.
Sources: I use physical archives and specialized literature for 1940 OOBs. They don't have URLs, but I'll use CS1 formatting going forward.
Experience: I've done extensive work in other Wikis before; I'm not a bot.

Note on timing: I received a message on my talk page at 20:41 (UTC) on Dec 24. I was reported here at 21:07 (UTC) the same day—only 26 minutes later, on Christmas Eve. I would have appreciated more than half an hour to respond to technical feedback before being publicly accused of using AI. I'm currently building a template library on my user page to fix all remaining issues. Lehrling18 (talk) 01:07, 25 December 2025 (UTC)

It is generally a bad idea to respond to allegations of AI use with AI.

I can tell because of the bolded list formatting. SuperPianoMan9167 (talk) 01:19, 25 December 2025 (UTC)

It's not AI. – You are using an LLM, if you continue to lie and edit in this manner using an LLM, an ANI report will be opened which will most likely result in you being indefinitely blocked or cbanned. Stop. fifteen thousand two hundred twenty four (talk) 06:57, 25 December 2025 (UTC)

Weirdly, they seem to have vanished immediately after they were reported here. Prior to this, they were making multiple edits per day. ~2025-31416-56 (talk) 19:05, 9 January 2026 (UTC)

Their userpage also appears totally AI generated. Sarsenet•he/they•(talk) 09:08, 25 December 2025 (UTC)

Hallucinating sources

Started finding what appear to be hallucinated citations – here, here, and here. (Even if the individual titles correspond to actual books, other details like authors and ISBNs did not match at all.)

Flagged to the editor who is experimenting with LLMs. They have now reverted a bunch of their own edits but think this series of articles is worth a closer look to make sure there aren't any other garbled or made-up citations. Cielquiparle (talk) 13:03, 30 December 2025 (UTC)

Tracking subpage created and is available at the top of this report. 158 pages need review. I've done my best to exclude obviously unproblematic edits. fifteen thousand two hundred twenty four (talk) 13:44, 30 December 2025 (UTC)

I've given them a bespoke final warning, if they misuse an LLM again please ping me. fifteen thousand two hundred twenty four (talk) 14:10, 30 December 2025 (UTC)

@Fifteen thousand two hundred twenty four They continue to edit without checking their own work or taking responsibility per their Talk page. Seems like they are WP:NOTHERE to build an encyclopedia. Cielquiparle (talk) 19:26, 25 January 2026 (UTC)

WikiNoia is now indefinitely blocked from articlespace following an ANI report at WP:ANI#Another continually unconstructive LLM editor. fifteen thousand two hundred twenty four (talk) 01:02, 26 January 2026 (UTC)

There exists unanimous consensus to allow for presumptive reverts of all WikiNoia's past articlespace, non-minor edits, see WP:ANI#Presumptive revert proposal. When making a presumptive revert, I would suggest leaving a link to that consensus in the edit summary. fifteen thousand two hundred twenty four (talk) 23:43, 26 January 2026 (UTC)

Flemza

Flemza (talk · contribs) seems to be using LLMs with minimal subsequent review to create questionably notable articles on individual songs. I just draftified everything I could but many others were created from redirects and aren't eligible. The user acknowledged LLM use at User talk:Flemza § LLM use and claims they were reviewing the outputs , but I am unconvinced by their responses at Draft talk:No Apologies (Papa Roach song) § LLM generated content where I documented several examples of problematic and likely LLM-generated content. They also moved an article back to mainspace after a rapid rewrite that introduced a material copyright violation. They have not answered my question about whether this violation was caused by LLM or by them. NicheSports (talk) 20:53, 31 December 2025 (UTC)

They're still creating new articles that have a few AI signs, Shotgun Blues (song) in the last few days. This probably needs admin attention if we're certain Kowal2701 (talk) 18:01, 6 February 2026 (UTC)

Thanks for checking Kowal2701. I can do a source verification analysis next weekend if no one else has time before NicheSports (talk) 15:41, 7 February 2026 (UTC)

NicheSports, are you able to have a look at this? They created Holy Water (The Funeral Portrait song) a couple days ago (seems better quality than their previous ones) Kowal2701 (talk, contribs) 22:52, 2 March 2026 (UTC)

Frontier models are unbelievably good and are absolutely learning from our efforts here. In many cases it is now difficult to know if something was LLM generated when there aren't talk page comments providing examples of the user's own writing for comparison. In this case there isn't much, basically only . There is some CLOP in the article - is one of his most personal and cathartic songs, about confronting dark moments and finding the strength to overcome them is mostly lifted from . So yeah most likely this editor is still using LLMs but the only way to determine with any certainty is a detailed source analysis of the entire article. NicheSports (talk) 06:14, 3 March 2026 (UTC)

Likely AI translations by User:Leeanah with some issues

The edit summary for this initial version of Transport in New Caledonia claims it's a translation from the French Wikipedia article. I don't speak fluent French... but you don't have to speak French to notice that in the French article, at time of creation, the "Transport fluvial" section was empty besides a template to expand it. That section of the English article has a lot of text there, and it sure reads like AI text.

I don't know whether the LLM is doing this unprompted or if it was prompted to fix the "expand" tag -- the rest of the article isn't that obvious as AI, but I don't know French well enough to say how faithful it is. Either way though that's not great, and they have several other translations. Someone asked previously on their talk page a while back about whether they used AI, but they didn't answer. Gnomingstuff (talk) 18:57, 2 January 2026 (UTC)

When creating Correction of the Rhône upstream of Lake Geneva they added a utm_source=chatgpt.com ref which was absent from the corresponding frwiki article. Same with History of Suresnes. They've used chatgpt outside of translation tasks as well. I don't speak French either, but it is clear that the rich in preserve a rich and varied folklore from Acadian folklore isn't in the original frwiki text (and neither is the *relevailles* markdown). Not looking good.

This is the second time an OKA member has been the subject of a noticeboard post in recent days due to possible translation issues and lacking communication, see this ANI report. In that report the OKA founder, 7804j, was very responsive, so pinging them may be an option if Leeanah doesn't engage, and Leeanah really needs to engage. fifteen thousand two hundred twenty four (talk) 19:29, 2 January 2026 (UTC)

Thanks for the info. (to be clear I stumbled across this article by searching for keywords, not seeking out anything from that project specifically) Gnomingstuff (talk) 19:35, 2 January 2026 (UTC)

It seems too faithful in parts. For example, translating "route transversale" very literally into "transverse road". No idea if that is a manual translation issue or a machine translation issue, I have seen it occur in both. The presence of "<---- second part ---->" is quite odd though, and does suggest multiple chunks were translated separately and stuck together. CMD (talk) 10:19, 3 January 2026 (UTC)

Thank you for taking the time to look into this and for laying out the concerns in detail.

This edit was made back in August 2025, and I want to be clear that the “River transport” section should not have been added the way it was. At the time, I misunderstood the boundaries of a translation task and treated an empty section in the source article as something that could be developed further. That was an error on my part.

I did consult external sources then, but after reading the discussion here, it’s clear that parts of the content were not properly supported by those sources and should not have been included. Regardless of my intent, the end result does not meet Wikipedia’s standards for verifiability or for faithful translation.

As a first step, I will remove the “River transport” section entirely, as well as any other content identified as unsupported, so that the article reflects only what is present and sourced in the French version.

I’ve also changed how I work since then. I no longer expand or supplement content during translation. My role is strictly limited to translating existing, sourced text, with AI tools used only to assist with revising language and editing formatting, followed by manual review against the original article.

I also recognize that I should have responded sooner on the talk page, for that I apologize. While I’m a professional translator, I’m not the most experienced when it comes to navigating talk pages and Wikipedia’s internal processes more broadly. That’s something I’m actively working to improve, so I can engage more effectively and avoid situations like this in the future.

I appreciate the feedback from everyone here and take responsibility for fixing this and making sure it doesn’t happen again. Leeanah (talk) 15:59, 11 January 2026 (UTC)

Is it valid to use "AI tools used only to assist with revising language and editing formatting, followed by manual review against the original article"? I suppose the answer depends on how much "manual review" is really done. (What is "editing formatting"?) David10244 (talk) 03:52, 18 January 2026 (UTC)

with AI tools used only to assist with revising language and editing formatting, followed by manual review against the original article. This is evidently untrue, as evidenced by the fact that there are diffs above showing you added utm_source=chatGPT parameters, which only occur when source links are copypasted directly from ChatGPT. Either ChatGPT inserted a source into its 'edited formatting' which you copypasted without realising you were doing so, or you are not being entirely truthful about the scope of your AI usage.

It is also worth noting that your response here reads as at least partially AI generated in itself. In particular, language like As a first step... and the insistence that AI was only used for copyediting are both hallmarks of AI-generated responses.

@7804j I'm having serious concerns about your standards for the editors you employ via OKA. If you want your project to be taken seriously as a constructive part of Wikipedia, this kind of conduct by one of your translators is hugely problematic. Athanelar (talk) 12:56, 24 January 2026 (UTC)

Proposed deletion of History of U.S. involvement in Venezuela's petroleum industry

This page was seemingly created with an LLM (has ?utm_source=chatgpt strings and self-citations to other Wikipedia articles), but has since been edited by humans. I proposed it for deletion since it falls afoul of WP:NEWLLM, but if someone wants to take a crack at cleaning it up, please go ahead. There should likely be an article on this subject, just not contaminated with LLM slop. Einsof (talk) 23:39, 3 January 2026 (UTC)

If it has been substantially edited by humans then it does not violate NEWLLM. SuperPianoMan9167 (talk) 00:07, 4 January 2026 (UTC)

The instruction from NEWLLM reads, in its entirety, "Large language models should not be used to generate new Wikipedia articles from scratch". It makes no exception for what happens afterward. Einsof (talk) 00:10, 4 January 2026 (UTC)

Seems better suited for AfD than for prod, since this is very much not an uncontroversial topic nor of uncontroversial notability Gnomingstuff (talk) 00:32, 4 January 2026 (UTC)

Quite possible. I figured that since there are so many eyes on this topic right now, there is plenty of opportunity over the next week for anyone who objects to PROD to convert it to AfD. Einsof (talk) 00:39, 4 January 2026 (UTC)

I agree that there should be an article and will be trying to my best to rewrite it over the course of the week! I've rewritten the lead and the first section already.

Seems like there's already been a few edits to remove all the Wikipedia sources, although many unverified claims remain. I'm hoping to replace almost all of the prose in my changes. Altoids0 (talk) 22:05, 4 January 2026 (UTC)

Communitylover

Communitylover (talk · contribs · deleted contribs · logs · filter log · block user · block log) says they are a teacher working with their class to create an article on a local business. The draft they created tripped 1346 (hist · log) for a URL with a chatgpt parameter and I think is mostly raw LLM output - see Draft:Hyde Park Hair Salon Barbershop of Hyde Park. Over-attribution everywhere. I asked them about it on their talk page and they told me, in a message I am about 98% sure was LLM-generated (gptzero says 100%), that they had not used an LLM. They then doubled down on this in their own words - note the difference in writing style. I've never filed here for an editor based on a single draft, but this user seems sufficiently non-collaborative that I'd like a second opinion and (potentially) some help managing the situation. NicheSports (talk) 04:58, 4 January 2026 (UTC)

I've left them a comment also. They've been provided with relevant guidance, I'd wait and see what they do with it. fifteen thousand two hundred twenty four (talk) 05:44, 4 January 2026 (UTC)

Thanks fifteen. I saw your message on their talk page - very helpful NicheSports (talk) 05:46, 4 January 2026 (UTC)

User:CZmarlin and promotional edits

This user has made, and continues to make, a lot of articles and substantial rewrites of articles, mostly regarding cars and colleges. Going back to at least early 2025, they appear to be AI. They have denied it, sort of (although without directly denying that they used AI, as opposed to the books/articles cited) but I don't believe this, based on the content at hand.

Most of their substantive edits claim to be "CE" (copyedit I assume), but they are not just "copyedits" as they add substantive content. That content generally shows the bog standard WP:AISIGNS. A sampling:

Rambler Rebel: like tacking promotional participials onto sentences (, offering a compelling alternative to consumers., , showcasing AMC's technological initiatives., etc), adding editorializing not present in the original text, e.g., turning "Development of AMC's new overhead-valve V8 engine began..." into A crucial component of this new direction was the development... (also note crucial) I assume it has been hand edited somewhat -- how else does one get This change marked a pivotal for AMC -- but still.
Siedlce University: Promotional AI stuff all over the place, the most glaring being Its history, from its origins as a teacher-training college to its current status as a comprehensive university, underscores its enduring dedication to education and its impact on the Siedlce region and beyond.
Concord University: Undue emphasis on symbolism (its journey reflects the broader development, superficial analyses (reflecting its evolving mission, underscored the state's growing commitment to public education, signaling a new chapter in its long-standing commitment...), promotional tone and AI vocab all over the place, etc.
AMC Matador: The same stuff: The new Premier represented a new direction for the former AMC, showcasing a more modern, European-influenced design and engineering, reflecting the evolving automotive landscape of the late 1980s..

I haven't left a new comment because based on the above response I do not expect it to go well. Gnomingstuff (talk) 16:43, 5 January 2026 (UTC)

AMC Hornet: The AMC Hornet emerged as a pivotal vehicle in AMC's strategy, directly replacing the venerable ..., However, its robust and versatile rear-wheel-drive platform continued to serve as ... Furthermore, demonstrating the enduring adaptability of its original design, the Hornet's chassis underpinned the innovative ...
Nash 600 The 1949 Nash 600, with its daring aerodynamic styling, spacious interior, and innovative comfort features, represented a bold step for Nash Motors in the post-war automotive landscape, establishing a distinct identity in the competitive economy car segment, This innovative placement aimed to bring essential gauges closer to the driver's line of sight, a concept ahead of its time ... Nash also introduced a groundbreaking "Twin Bed" feature ...
Jerrari: These unique hybrids combine the rugged utility of a Jeep Wagoneer with the exotic power of a Ferrari, The 1977 Jerrari Wagoneer adopted an understated approach ... Modifications to the exterior were subtle yet functional, The 1977 Jerrari stands as a testament to Harrah's unique vision and ... where it continues to captivate with its blend of utilitarian ruggedness and supercar heart.

They've added a fair amount of unencyclopedic prose infused with common LLM-isms. I'd be curious to hear about their process in crafting the above samples. fifteen thousand two hundred twenty four (talk) 20:06, 5 January 2026 (UTC)

Gnomingstuff & Fifteen thousand two hundred twenty four, any further thoughts on this one? The edits are ongoing. Thebiguglyalien (talk) 🛸 01:44, 19 January 2026 (UTC)

I've notified them of this discussion, hopefully they can enlighten us a bit as to what their process was when creating the above text. The prose is less than encyclopedic, LLM or not. fifteen thousand two hundred twenty four (talk) 03:31, 20 January 2026 (UTC)

apologies, have been working on other backlogs, looks like not much to do at this point in time Gnomingstuff (talk) 20:29, 22 January 2026 (UTC)

Preye douglas

Preye douglas (talk · contribs) is a non-native English speaker who is seemingly using LLMs for all of their major mainspace edits, including many article creations , as well as using LLM for comments on talk pages . They have been warned about this multiple times on their talk page. They have wikilawyered about their use of LLMs . I've been following this for a month but both the LLM use and wikilawyering about it (bonus, also LLM-generated) are continuing, so here we are. A few examples:

The year began with Ruger's "Asiwaju" holding the top position for four consecutive weeks in January. Early 2023 saw collaborative success, with "Gwagwalada" by Bnxn, Kizz Daniel, and Seyi Vibez spending four weeks at number one, marking one of the year's most successful multi-artist collaborations.: The editorializing about collaborative success is unsupported by the source
Together, these metrics reflect the overall consumption and popularity of music within Nigeria's contemporary music landscape. has no in-line citation at all
In accompanying press materials, the track was introduced as a mid-tempo afropop song that combines affirmational lyrics with a reflective tone I'll let you guess whether the bolded bit verifies in the source

NicheSports (talk) 03:13, 6 January 2026 (UTC)

Gwagwalada (song) is considered a collaborative success as the song charted on multiple charts. It reached number one in Nigeria and it's certified 2x platinum. If that doesn't sound like success to you I don't know what would. Not many collaborations achieved that in 2023. As for the Turntable chart methodology there is a source for that added right after "radio and television plays across Nigeria."

The line from Commas was basically paraphrasing of what was said in the sources.

Over a gentle rhythmic beat with swelling strings; ‘Commas’ exudes breezy confidence with a hint of reflection from the 21-year old who, in just three years, is already growing to become one of the most vital African artists of her generation. - Music-News

“Commas” offers fans an intimate look into the vocal growth and journey Starr has embarked on over the last few years. Staying true to her positive and affirming lyricism, Starr’s latest single spotlights the singer’s continued focus on her growth as she sings, “I carry God so I fear nothing, Steady increasing the commas” on the single’s chorus. – OkayAfrica Preye douglas (talk) 03:47, 6 January 2026 (UTC)

This is pure WP:OR
The lead had no citations at all, and given there was no body in the article (only a table) all claims in the lead lack in line citation support
You did not provide the OkayAfrica article as a source for the third claim. Your statement above misrepresents the state of the article, whether you intended to or not. Please do not do that again

Did you use LLMs to assist with any of the edits I linked to in my post above? NicheSports (talk) 04:33, 6 January 2026 (UTC)

@NicheSports It's not WP:OR because "hint of reflection + affirmative lyricism" can be paraphrased as "affirmational lyrics with a reflective tone." Putting it that way, you are basically still saying the same without copying and pasting.

Typically, lead sections don't really require citations per WP:CITELEAD and that's just a list article. Apart from the chart methodology, whatever you see on there are facts derived from the table. The table has sources from the TurnTable magazine website.

The OkayAfrica source was cited in other parts of the article. I may not have added it as an inline citation after that line but hey, it's among the references on the page. Since you have a problem with that, you should have just added ^{[citation needed]} template after that line. But anyways, I will be adding that OkayAfrica reference now.

There is no denying that Artificial Intelligence has become a part of our daily lives, as mainly assistants. So do I use artificial intelligence in my daily life and work? Yes. Do I sheepishly follow whatever AI says without verifying if they are actually credible? No. Do I use LLMs in creating Wikipedia articles? Not so much just basically supportive tasks such as copyediting, and improving grammar but of course with caution as I am well aware of the tendency of these bots hallucinating at times and spreading misinformation.

Are those lines you listed written using LLMs? No. Those are basically the kind of lines you would find on any Wikipedia article about a record chart or song. Long before LLMs editors have always drawn inspiration from existing Wikipedia articles. Preye douglas (talk) 05:30, 6 January 2026 (UTC)

My comment about WP:OR was in relation to the first example in my original post, namely the phrases "Early 2023 saw collaborative success" and "marking one of the year's most successful multi-artist collaborations" NicheSports (talk) 05:51, 6 January 2026 (UTC)

That's basic English. It isn't that hard, really. The song was released in February 2023, that's considered an early part of the year. Still Q1. Now this collaboration goes on to dominate airplay and television charts in a country of 230 million people, topping the national chart and even getting plays in the UK to extent that it made top 10 on one of the country's official chart and still went ahead to shine on a billboard chart while being certified 2x platinum and nominated for Best Collaboration at Africa's version of Grammy Awards. That's huge success. It don't need to be on Hot 100 to be called a success :) Preye douglas (talk) 06:10, 6 January 2026 (UTC)

Do not combine material from multiple sources to state or imply a conclusion not explicitly stated by any of the sources. Similarly, do not combine different parts of one source to state or imply a conclusion not explicitly stated by the source. (emphasis mine)

Have you read the no original research policy? fifteen thousand two hundred twenty four (talk) 15:05, 6 January 2026 (UTC)

I have edited out that part, leaving the main quote from the artist and I have read WP:OR Preye douglas (talk) 19:55, 6 January 2026 (UTC)

Note: I forgot to link to the articles for the three examples I provided above. I added those above, permalinked to their most recent version at the time of my initial post NicheSports (talk) 04:08, 6 January 2026 (UTC)
There you go. No Wikipedia editor is perfect. Everyone makes mistake sometimes. The important thing is admitting it and fixing it or correcting whatever that was wrong. I make sure to double check my articles before publishing. I use sandbox to create. I cannot guarantee that there won't be mistakes but as a community why don't we work together to call out these mistakes when we see them or try to fix them ourselves. Some of the issues like not citing sources properly are things you can point out on the article talk pages and tag the editor to it so they can fix it if you don't want to do it yourself. And these kinds of mistakes are human error, some are caused by the editor not having enough time or probably they have a busy schedule so they weren't able to add all references. These kind of things happen on Wikipedia not trying to say it's a good practice but just saying that calling me out and attributing it to LLM is just unfounded because prior to the evolution of LLMs or proliferation of Artificial Intelligence, Wikipedia has always had citation issues this is just human error, as a matter of fact LLMs I believe have the capability to also cite sources so if truly I was using LLMs I'd probably just program it or train it to write and include citations as well and I bet you they would do it better than most editors here. Preye douglas (talk) 04:35, 6 January 2026 (UTC)
Do you use tools such as Grammarly, Quillbot, or DeepL? Those are AI-based which is not always clear in their advertising. Gnomingstuff (talk) 20:09, 6 January 2026 (UTC)
No, I don't use any of those tools. Preye douglas (talk) 23:29, 6 January 2026 (UTC)

Going 2x platinum and reaching number one in a country with over 230 million people is success. Not to mention the song penetrated a UK official chart and US billboard chart. Very few Nigerian songs get to that level so that's success judging from where the artists are coming from. Preye douglas (talk) 04:16, 6 January 2026 (UTC)

Since when did it become a good practice to copy and paste exactly what sources say. Experienced Wikipedia editors know that first you have to read what's in the source and deliver it in your own words while not straying very far from what the sources are saying. Preye douglas (talk) 04:21, 6 January 2026 (UTC)

has some AI signs Kowal2701 (talk, contribs) 18:14, 11 February 2026 (UTC)

Can you be more specific? What was the issue? Is it the citations? Did any phrase sound like puffery? Preye douglas (talk) 00:22, 12 February 2026 (UTC)

It was a language thing, but if the sources support everything there then it’s alright. AI signs aren’t the problem, it’s the WP:V failures etc. that they indicate Kowal2701 (talk, contribs) 00:26, 12 February 2026 (UTC)

Thanks for your response. I checked the portions you highlighted and can't find where WP:V wasn't met. Preye douglas (talk) 00:30, 12 February 2026 (UTC)

More LLM translations, more issues

This is another set of translations from the OKA project (which has disclosed LLM use), but I found it via an unrelated search.

There are dozens of these but I'm just going to focus on Caciquism. This is the French article at that time. There is AI text that does not have any counterpart in it, most of which are synthesis tacked on to paragraphs, such as:

This linguistic borrowing highlights the historical and cultural connections between these various groups.

* This made them a crucial link during the era. (see below)

This anecdote depicts the workings of caciquism and the seizure of power by the two dynastic parties.
Additionally, the formation of important mass parties, such as the CEDA, marked a pivotal moment in political history. (CEDA is mentioned and called an important party but not the "pivotal moment" stuff)

This seems to be most of the obvious ones; my knowledge of French isn't good enough to assess the faithfulness of the actually translated stuff well. (It seems like some direct quotes are being paraphrased and/or "translated" into AI-isms, which seem somewhat anachronistic given the period of history we're talking, but that could well be from the French quotes.) Gnomingstuff (talk) 21:48, 7 January 2026 (UTC)

I know the head of the project has come forward to disclose use, but it looks like it’s just the tip of the iceberg regarding how much AI this project contains. The instructions on meta direct people to translate using Grok, then use other LLMs to improve the language. I can’t imagine the result being anything more than nonsense soup. By their own admission, the bulk of the work is done via LLM. ExtantRotations (talk) 02:40, 9 January 2026 (UTC)

Luckily, I'm a native French speaker so I should be able to help! The lead section is an accurate translation with sentences being split for readability, with the only three actual changes being Dans l'historiographie, ainsi que dans la presse ou les milieux intellectuels de l'époque (In historiography, as well as in the press or intellectual circles of the era) becoming In historiography, journalism, and intellectual circles of the era (removing the implication that it was the press of that era), très influent essai (very influential essay) being translated to influential essay, and that same sentence moving around leading en raison du ... (due to the...) becoming ... popularized the term. So, pretty okay lead translation all things considered.

In the #Concept of "cacique" section, the very first paragraph adds changes in meaning. Besides the hallucinated sentence pointed out by @Gnomingstuff, it removes the fact that the word in other Western European languages is a borrowing of the Spanish one (while that could be something not verified by the source, there is no indication of that being the case). It also adds a specifically (in It referred specifically to...) which doesn't carry any meaning or equivalent in the French original. The second paragraph is a very accurate translation, and I don't have anything to say about it. In the third paragraph, I'm confused by "seigneur" not being translated – the word doesn't have any specifically French meaning as we're talking about Spanish society primarily, and translating it as "lord" (while keeping the Spanish original "señor") would've made more sense. What leaves me more puzzled is the choice to move half of the later quote outside of the quotation marks while leaving it closely paraphrased.

The fourth paragraph is where we start to see more inaccurate translations, especially in quotes: personne puissante, qui jouit d’une influence par la crainte qu’il inspire dans une localité (a powerful person who holds influence through the fear they instill in a locality) becomes a dominating individual who instills fear and holds influence in a locality, which, while approximate, obscures the link made in the original quote between fear and influence. We also see qui a plus de commandement et de pouvoir, et veut par sa superbe être craint et obéi des inférieurs (who wields more command and power, and wants through his splendor to be feared and obeyed by inferiors) become who wields more power and commands more respect by being feared and obeyed by those beneath them, which completely reverses the causal link. The fifth paragraph and the two associated quotes are comparatively quite accurate. The last paragraph of this first subsection is also okay, although we again see the issue of quotes being very closely paraphrased without being marked as quotes. The other two very short subsections (three sentences in total) are fine too.

In the beginning of the second section, This made them a crucial link during the era. is surprisingly not an AI-ism, but a translation of Il devient toutefois un maillon véritablement essentiel au cours de... that got moved at the end of the paragraph. I've only briefly scrolled through the rest of the article, but I'm especially concerned at two long paragraphs, marked as unsourced in the French original, being imported without sources and with the tags removed. Chaotic Enby (talk · contribs) 11:24, 9 January 2026 (UTC)

oh good catch on the latter, I was mostly going sentence by sentence. For some reason LLMs seem to really hate direct quotations when used for editing, they often replace them with paraphrases, so that part isn't surprising. Gnomingstuff (talk) 14:22, 9 January 2026 (UTC)

The more worrying part is that the direct quotations are often very closely paraphrased (if at all) while not being marked as quotes in the output. Chaotic Enby (talk · contribs) 14:33, 9 January 2026 (UTC)

Camoz87

Camoz87 (talk · contribs) I've been following this one for a few weeks. This user is a non-native English speaker who is relying on machine translation (at a minimum) for their mainspace edits. They have added transvio to articles and have been warned about LLM use 5+ times, by 3 editors, on their talk page. They have created many articles, which are often then draftified by other editors . This machine-generated text is insufficiently reviewed, sometimes including references with broken URLs . Here is an example of a material source-to-text integrity issue that I identified . This is even worse: it sure looks like they ran a few hundred poems through an LLM and asked it to summarize them in English. The user shows no signs of changing their approach, continuing to make machine-assisted edits today. NicheSports (talk) 16:04, 9 January 2026 (UTC)

There's got to be a quicker way to sort cases like these out rather than a lengthy ANI thread followed by a CBAN Kowal2701 (talk) 17:07, 9 January 2026 (UTC)

Agreed. Although this one doesn't merit a CBAN, rather a mainspace block. It could have gone to ANI at any point in the last week I've just had ANI LLM fatigue. NicheSports (talk) 17:20, 9 January 2026 (UTC)

As an it.wiki sysop please consider that this user has been indef banned on it.wiki and is constantly evading the block creating hundreds of articles with LLMs even in his native language. --Friniate ✉ 20:18, 12 January 2026 (UTC)

Thanks for the heads up. Just filed at ANI where I imagine this will be quickly handled. Pinged you there out of courtesy in case you want to add anything re: block evasion NicheSports (talk) 21:03, 12 January 2026 (UTC)

Or not so quickly. Confused by this one NicheSports (talk) 23:57, 13 January 2026 (UTC)

I assume people are waiting either for them to respond or to start editing again, they haven’t been active since the 9th Kowal2701 (talk) 00:03, 14 January 2026 (UTC)

On a related issue, Is there a quick way to get the list of articles edited by them needing cleanup? (I mean, without creating it manually) --Friniate ✉ 00:31, 14 January 2026 (UTC)

For the enwiki pages they've created, see here . I was going to AfD/prod some of these once the ANI thread was resolved.

Also noting their global contribs they've been active more recently at es.wiki, porting over many of the same LLM-generated articles. Does this merit a global lock? NicheSports (talk) 00:36, 14 January 2026 (UTC)

@NicheSports Thanks, but I meant a list of all edited articles like the subpage created by @Fifteen thousand two hundred twenty four. Is there a quick way to have it or must it be done manually?

I'm going to ask to a steward if it's possible to apply a global lock, but I'm not sure if it will be done as an ANI thread on enwiki is still ongoing... --Friniate ✉ 15:28, 14 January 2026 (UTC)

Yeah I think requesting a global lock is a good idea, they're active at es.wiki and wikidata (though I personally can't tell if their edits are unconstructive). The tables are created by using User:DVRTed/AINB-helper script Kowal2701 (talk) 15:34, 14 January 2026 (UTC)

@Kowal2701 Thank you, I'll have a look at it! --Friniate ✉ 16:10, 14 January 2026 (UTC)

This Spanish language article they created on 11 January is a word-for-word translation of the likely LLM-generated Letter from Leonardo da Vinci to Ludovico Sforza. Given their extensive use of machine-translation on it.wiki, it is probably safe to conclude this was a machine translation. NicheSports (talk) 16:14, 14 January 2026 (UTC)

(As I thought, the steward I reached confirmed that it's preferable to wait for the conclusion of the ANI procedure here, in order not to overcome local community consensus). --Friniate ✉ 16:23, 14 January 2026 (UTC)

I didn't use AINB-helper for this one, though that can be a useful tool. The list was fetched using CCI's contribution surveyor (uncheck minor edits, set a relevant date range and a large negative bytes value), then formatted. If an editor wants a list of edited articles without creating a new subpage, then contribution surveyor is a more useful tool. fifteen thousand two hundred twenty four (talk) 17:40, 14 January 2026 (UTC)

Tracking subpage created and is accessible at the top of this report, 38 articles appear to warrant further review. fifteen thousand two hundred twenty four (talk) 02:01, 14 January 2026 (UTC)

They have now returned to enwiki and are attempting to G7 several of their article creations, with the edit summary "Italian LTA", which I suppose they are per Friniate, and trying to link to this thread. See . They are also updating the tracking page (sometimes incorrectly). Not sure how to interpret this. NicheSports (talk) 13:54, 14 January 2026 (UTC)

Have asked them to engage at ANI, but they appear to be trying to aid in the clean-up? Kowal2701 (talk) 15:05, 14 January 2026 (UTC)

Seems like it? Thanks for reaching out to them. Some communication would be helpful. NicheSports (talk) 15:13, 14 January 2026 (UTC)

Speaking from experience I tend to have a much more malignant interpretation of his behaviour, he has an history of meddling with cleanups also of other users' edits. I think it's also likely that he's continuing to edit with temporary accounts... --Friniate ✉ 15:24, 14 January 2026 (UTC)

For example like this one.... --Friniate ✉ 15:30, 14 January 2026 (UTC)

Or this one... --Friniate ✉ 15:32, 14 January 2026 (UTC)

As I thought, they for example tagged Giovanni Battista Ramusio as completed, without reverting themeselves... --Friniate ✉ 15:36, 14 January 2026 (UTC)

@Friniate Anachronist took care of things on enwiki . If you want to do anything on the cross-wiki side I'll leave it to you; re-requesting a global lock makes sense to me but I'm out of my depth here. I looked further into es.wiki and since January 7th they've made ~100 edits there , many but not all of which seem to be machine translations of their recent edits to enwiki. Examples: :: :: , etc. NicheSports (talk) 07:02, 16 January 2026 (UTC)

~~It's only a partial block from article creation, so not entirely remedied.~~ (misread the block log) fifteen thousand two hundred twenty four (talk) 19:24, 16 January 2026 (UTC)

They are blocked from creating or editing articles. PackMecEng (talk) 21:58, 16 January 2026 (UTC)

@Friniate pinging you here per our discussion at a different talk page. I saw this comment from @Ca yesterday and thought they might be able to help delete this LTA's LLM creations. From xtools Camoz created 36 articles in the LLM-era (a dozen of which are already deleted); from what I have seen, all of these are likely unreviewed LLM-generated output and many can probably be deleted at AfD per WP:NEWLLM as long as they have not been subsequently materially edited by another user. Some of them are pretty egregious, such as List of Carmina Burana. If I weren't traveling I would help AfD some of these but I thought Ca might be willing to take a look. NicheSports (talk) 03:57, 25 January 2026 (UTC)

More information TBC elsewhere ...

TBC elsewhere

Could start a thread at WP:VPI for brainstorming a better process? Though this might need to wait until we get a proper policy at WP:NEWLLM. You might be better off messaging an admin on their talk page, outlining the warnings etc. and asking for a formal warning from them? Kowal2701 (talk) 17:27, 9 January 2026 (UTC)

I've been thinking about this for a while and have a few ideas. My two basic ideas are 1) establishing that denials of unambiguous LLM use (a high bar, which will mostly apply when there are multiple talk page comments establishing that a user is a non-native English speaker) should be an automatic, typically indefinite, mainspace block, which the user can appeal by acknowledging the LLM use and accepting a complete ban on LLM editing assistance of any kind including LLM-powered "grammar checkers" 2) asking a few admins who are willing to sanction for LLM use to be more active at AINB. Maybe we do the first round at WT:AIC or even on someone's talk page to keep things from going off the rails? I would like to get 1-2 admins' input as well, Chaotic Enby for sure and I was thinking Star Mississippi given Star has handled a lot of the LLM blocks at ANI. Where do you think is best? NicheSports (talk) 17:44, 9 January 2026 (UTC)

That sounds great, especially the second one. Either ping people to WT:AIC under the premise that you’ve seen them be active on this, or message one of them asking if it’s a good idea (I’m not the best person to ask tbh). If the responses are positive, then a message at WP:AN asking admins to be a little active here probably isn’t a bad idea. Kowal2701 (talk) 18:25, 9 January 2026 (UTC)

Not really comfortable with a block criterion applying mostly to non-native English speakers, at the very least it is bad optics and may easily seem discriminatory. Gnomingstuff (talk) 20:03, 10 January 2026 (UTC)

The point isn't about non-native English speakers but identifying denials of unambiguous LLM use. There are other criteria that would apply. I'm also not looking for this to be a PAG as much as general guidance, with the goal of reducing time at ANI. Anyways, I'll open up a separate conversation, likely at WT:AIC, in the next few days. NicheSports (talk) 23:57, 10 January 2026 (UTC)

(NicheSports—Kowal2701—Fifteen thousand two hundred twenty four—PackMecEng) I started to ask for G15 on two articles created by Camoz (List of Carmina Burana and Bernardino Vázquez de Tapia, but since they are a lot, I prefer to write here before finishing the job: do we all agree to ask for G15 for all the articles created by them? --Friniate ✉ 18:54, 22 January 2026 (UTC)

@NicheSports sorry, I made a mistake with the ping so it probably didn't arrive to you. --Friniate ✉ 18:56, 22 January 2026 (UTC)

One of the three specific G15 criteria must be met before G15 can be requested, this means there needs to present either LLM communication, non-plausible hallucinated references, or nonsensical citations. Even if an editor is blocked following a discussion, unless there was consensus in that discussion, blanket deletion via G15 is not a default option.

Other remedies to consider in addition to G15 would be to draftify the article and leave a descriptive talk page message as to why, AfD, tagging the article, or fixing the article. fifteen thousand two hundred twenty four (talk) 19:14, 22 January 2026 (UTC)

@Fifteen thousand two hundred twenty four So do I have to write in every G15 request why I think that the article was created by an LLM? I'm certainly not going to fix them, as for the tagging I think that it'd just postpone the problem... Maybe G15 for clear-cut cases and draftification where we have reasonable doubts?

PS: I didn't suggest a blanket deletion for all the articles created by them, but only for those already listed in the clean-up page. --Friniate ✉ 20:44, 22 January 2026 (UTC)

Yes, the page must exhibit one of the three strict G15 criteria before being tagged, and which of those three criteria are best indicated when tagging. Performing a descriptive draftification, either by leaving an appropriate tag or talk page message along with a good edit summary, is often a valid option when G15 is not met.

To clarify, the tracking list contains articles which should be reviewed further, some of the listed pages may need no action taken at all, in which case the unnecessary status should be set. fifteen thousand two hundred twenty four (talk) 20:56, 22 January 2026 (UTC)

Agreed with fifteen's explanation. I try not to G15 nominate anything that is marginal. I was planning on AfD-ing List of Carmina Burana, as I'm pretty confident that one would be deleted at AfD. I am traveling for work but can look through their other creations for PROD/AfD candidates when I'm back. Anything that isn't deleted can safely be draftified per the machine-generated clause of WP:DRAFTREASON, which is much looser than the G15 criteria, as long as the article also meets the basic rules outlined in WP:DRAFTNO. NicheSports (talk) 22:09, 22 January 2026 (UTC)

@NicheSports Ok, you have much more experience than me, if you are going to open AfDs I'll leave it to you then, I'll only check the articles which don't need AfDs or to be draftified.. --Friniate ✉ 22:53, 22 January 2026 (UTC)

By the way, I think that TA as this were most likely always them, so we should pay attention to possible socks. --Friniate ✉ 23:24, 22 January 2026 (UTC)

Agreed that looks like them. Probably worth adding a few pages to my watchlist... fifteen often looks out for socks as well. Thanks again for letting us know about the cross wiki behavior btw. NicheSports (talk) 23:29, 22 January 2026 (UTC)

No issue, thank to you for all the work you've done, a wikiproject like this is really cool XD --Friniate ✉ 23:35, 22 January 2026 (UTC)

They've done a lot of logged out editing as TAs, more than I care to list. For those with TAIV the majority seem to be under the /20 of ~2026-92236, and most recently there's ~2026-29058-7 which is on a different network. Will monitor for a month or so. fifteen thousand two hundred twenty four (talk) 00:22, 23 January 2026 (UTC)

Edits from User:OrlovskyAtlas

This account was created in November and since then has completely re-written (or attempted to re-write) dozens of articles, mostly in the financial space and on contentious topic WP:CT/IRP. They mostly do not include edit summaries, but when they do they seem obviously auto-generated. Despite receiving numerous warnings for their edits, they have yet to reply to any of them, which makes me further suspicious. There do appear to be minor human-generated edits in-between these massive rewrites, but WP:AISIGNS abound in the larger edits. Most suspicious of all is that several of these extensive re-writes occur just minutes apart (example: 1 and 2); a completely impossible timeline for a human. RunningOnBrains^(talk) 17:53, 9 January 2026 (UTC)

The user is still engaging in implausible rewrites. Reported at ANIBoynamedsue (talk) 07:07, 12 January 2026 (UTC)

Courtesy link: WP:ANI § OrlovskyAtlas: Constant use of AI despite warning. fifteen thousand two hundred twenty four (talk) 16:00, 12 January 2026 (UTC)

I've reverted a handful. Sarsenet•he/they•(talk) 10:45, 12 January 2026 (UTC)

Suspected LLM use in articles created by MisawaSakura

Came across a few new articles by an extended confirmed user MisawaSakura (talk · contribs · deleted contribs · logs · filter log · block user · block log) that showed some signs of LLM use. I draftified them and asked the creator on their user talk page if they'd used LLMs. They repeatedly dodged the question and eventually ragequit. More background here. Based on Lee Vilenski's suggestion I'm bringing this up here since who knows how many more LLM-generated pages there could be that I didn't scroll down far enough to catch. —pythoncoder (talk | contribs) 18:53, 9 January 2026 (UTC)

That's quite a lot of mainspace page creations. ~Anachronist (who / me) (talk) 05:31, 16 January 2026 (UTC)

@Pythoncoder: Kristina Zlateva, that you moved to draftspace on account of its likelihood of being LLM-generated, had several sentences removed but otherwise underwent no fundamental textual rewrite. User:MisawaSakura then submitted it to AfC where it was accepted the following day by User:MurielMary. So unfortunately it still contains the original LLM writing (which a couple of checks have confirmed). —Fortuna, imperatrix 13:05, 17 January 2026 (UTC)

RfC of possible interest to this project

Image use policy - prioritizing images used by reliable sources when using AI-generated imagery? — Rhododendrites ^talk \\ 20:21, 9 January 2026 (UTC)

Jībanmṛta and Esyms

Jībanmṛta (talk · contribs) made a large number of LLM-generated edits before they were globally locked as a sockpuppet. Most of these edits fall within the South Asia contentious topic. They had a tendency to add large volumes of densely referenced text in quick succession at a pace that would not be possible if researched and written manually, which was frequently followed by the addition of wikilinks in a separate edit afterward. Examples include:

Terrorism in Bangladesh: 4 November 2025 (18:06, 18:41, 18:57, 19:24, 19:35); 5 November 2025 (05:43, 06:14, 06:51, 06:59)
Sindoor: 15 November 2025 (17:49, 17:58, 18:14, 18:23, 18:25, 18:38)
Ram Mandir Dhwajarohan: 28 November 2025 (09:54, 10:05, 10:17, 10:36, 10:43, 11:25)

Jībanmṛta also participated in the GOCE December 2025 Copy Editing Blitz, in which they claimed to have copyedited articles that are thousands of words long in a matter of minutes. For example, in Special:Diff/1327817371, they removed 4,054 characters from the Sichuan opera article 13 minutes after placing the {{GOCEinuse}} tag to indicate that they had started the copyediting process.

As a sockpuppet of a previously blocked and banned account (CosmLearner), Jībanmṛta's edits can be reversed without needing to provide any further justification per the block evasion (WP:BE) and ban evasion (WP:BRV) policies, as long as at least one of the policies is cited in the edit summary. — Newslinger talk 13:11, 14 January 2026 (UTC)

Esyms (talk · contribs) has also been globally locked as a sockpuppet of CosmLearner (talk · contribs), the same account that Jībanmṛta is a sockpuppet of. The edit history of User:Esyms/sandbox shows Esyms starting with unsourced LLM-generated content and adding citations to it afterward. — Newslinger talk 15:08, 14 January 2026 (UTC)

I see that Esyms had been participating in the GOCE January 2026 backlog elimination drive prior to being identified as a sockpuppet, and supposedly made this copyedit to the TV Brasília article in seven minutes. Dhtwiki, as the lead project coordinator of GOCE, is there any action you would like to take in response to this issue? — Newslinger talk 04:35, 15 January 2026 (UTC)

After looking at a couple of Esyms's copy edited articles, it looks like they were doing good work for the most part. Their "TV Brasilia" edits, which you have reverted, were not all to my liking, but it seemed to be helpful work done in good faith. If you are required, without exception, to undo all of their work, I will probably have to spend some time rescuing much of it. Dhtwiki (talk) 09:48, 15 January 2026 (UTC)

Thank you for taking a look. Esyms is the 11th sockpuppet listed in Category:Wikipedia sockpuppets of CosmLearner, so I will have to disagree with you on the presence of good faith. Personally, I don't think a sockpuppet who asked an AI chatbot to copyedit articles, and then pasted the contents of the output into Wikipedia, should be listed on the leaderboard, but that is a decision for the GOCE WikiProject and I am simply relaying the information to you. This particular incident is a dual cleanup case because it involves both LLMs and sockpuppetry: not all of Jībanmṛta's and Esyms's edits need to be reverted, but any edits that are not obvious improvements are likely to be reversed. — Newslinger talk 11:55, 15 January 2026 (UTC)

How can you tell that Esyms really did use AI to do their copy edits? The edits were both observant of MOS (placing dashes in date ranges) and then surprisingly unobservant (rewriting within a quote), but that could be true of purely human copy editing. Of course, Esyms is not going to receive barnstars, nor long remain on the January drive leaderboard. Dhtwiki (talk) 22:51, 15 January 2026 (UTC)

In Special:Diff/1332237323, Esyms created a user script with the comment "Grok AI been used in most of the places". In their next edit five minutes later, they made significant revisions to the script with the edit summary "AI rewritten". These two edits show that Esyms has access to an LLM (Grok) that would be capable of making the copyedits they submitted, and that Esyms has used this LLM to make edits on Wikipedia. The Jībanmṛta and Esyms accounts were operated by the same individual (CosmLearner), and as I mentioned in my initial comment and my second comment, the timing and order of these accounts' edits to article content are indicative of LLM use.

CosmLearner's other sockpuppets also have a long history of LLM abuse in Wikipedia articles. For example, the most recent edits in Special:Contributions/B'Desh-In Outlook show the addition of tens of thousands of characters of article content within minutes, which is even more ridiculous considering that it was done with the mobile web interface. When we look at the comments that were actually written by the same individual without LLM use, such as Special:Diff/1260009187, Special:Diff/1260433296 (edit summary), and Special:Diff/1256582596, we see a level of English proficiency that would be insufficient for copyediting articles on the English Wikipedia. I don't believe that Esyms/CosmLearner has the ability to perform the GOCE-related copyedits by hand.

Thanks for confirming that Esyms won't receive recognition for the January drive. That was my main concern when I pinged you, because it's not my place to make this type of decision. I recognize that the TV Brasília article does need to be copyedited again, and that many of the changes in Special:Diff/1331321872 were valid, so I will take some time to copyedit the article myself. — Newslinger talk 12:30, 16 January 2026 (UTC)

Thank you for the lengthy explanation and for seeing that a blanket rollback of Esyms's edits is unnecessary. Dhtwiki (talk) 05:54, 17 January 2026 (UTC)

I have checked the editor's 11 other articles, other than TV Brasilia, that Esyms copy edited for our drive and found very little that was wrong. If their work was done with an AI tool, it seems to be a very good tool for that kind of work. Dhtwiki (talk) 07:35, 24 January 2026 (UTC)

Hmlarson

Hmlarson (talk · contribs) is an experienced and autopatrolled editor who tripped 1325 (hist · log) with their creation of PWHL Takeover Tour. If they are using LLMs, the use is occasional, they may subsequently edit some of the content, and I do not think it predates November 2025. See this edit from 2023 which was definitely not LLM-generated; compare to the style of the recently-created Kimbra Walter. What seems clearer is that Hmlarson is creating articles with material source-to-text integrity issues; pretty much the entire Purpose and objectives section fails verification, and is the type of corporate bulleted promo you'd expect from an LLM. Unfortunately, the conversations at Talk:PWHL Takeover Tour about these issues did not go well, with aspersions I was operating a multiple accounts and no engagement with the potential issues raised . I placed an ai-generated tag on the article, which they removed . I ran the article through Earwig and found some CLOP (nothing remotely worthy of CCI), but 945 girls' hockey players participated in clinics and meet-and-greets with PWHL athletes is copied word for word from .

More information Snippets from recently-created articles ...

Snippets from recently-created articles

PWHL Takeover Tour

The Purpose and objectives section almost entirely fails verification as noted above
Talk:PWHL Takeover Tour § Potential WP:V issues documents additional issues
The initiative serves multiple purposes: growing the sport's fanbase, testing potential expansion markets... the bolded bit is SYNTH at best; the article neither mentions nor implies testing markets. This is sourced to a primary source, with support for these claims coming from a quote from a PWHL exec
The rising interest in the PWHL extends beyond its current cities, evidenced by a new single-day ticket sales record... this does verify, but to a primary and promotional source

Kimbra Walter

Large and undue Philanthropy section with LLM-style formatting. Some promo: The organization operates on core values of equity, opportunity, and sustainability.
Walter is a practicing attorney who has dedicated much of her career to philanthropic endeavors. fails verification and is sourced to a primary source
The trophy, designed by Tiffany & Co., is made of sterling silver, weighs approximately 35 pounds, and stands 24 inches tall. completely fails verification, sourced to

NicheSports (talk) 04:44, 16 January 2026 (UTC)

I did notify Hmlarson about this thread. I'm hoping we can discuss here; I respect their commitment to the articles they work on and they have clearly done a lot of great work on Wikipedia. I would have preferred to do this at a talk page but after the interaction at Talk:PWHL_Takeover_Tour I figured it was best to bring here. NicheSports (talk) 04:49, 16 January 2026 (UTC)

From a quick glance PWHL Takeover Tour#Format also fails verification.

Given that their only response has been to call you a sock the odds of interaction here seem slim, but hope springs eternal. Zygmeyer (talk) 05:12, 16 January 2026 (UTC)

the conversations at Talk:PWHL Takeover Tour about these issues did not go well – No kidding, the immediate and repeat followup condescension and bad faith musings are egregious.

I've looked at another one of their recent creations, 2021–22 PWHPA season, and it's so full of LLMisms and unsupported OR it's difficult to convey in a concise manner.

The Season statistics and results section is baffling from a human editor perspective, it's an unsourced section saying there is no source for the information, but makes sense if returned from an LLM.
The unsourced Notable players section contains a list with no notable players, just model bulletpoints of the types of notable players like International players from various countries and Younger players developing their skills.
The existance of an Impact and significance section itself is characteristic of model output, and the mostly-unsourced and editorial contents do not dissuade this notion: The season demonstrated the PWHPA's resilience and adaptability ... The successful navigation of the Olympic season and the continued growth of partnerships proved that the association's ...

This goes on and on. I've not bothered checking for source-text integrity, so much of the article is unsourced OR that it's already well past the threshold of the standards an autopatrolled editor should be holding themselves to. Given this and their recent unconstructive responses to good-faith concerns about their creations, I don't think they should have the perm, but that's something a different venue would have to address. fifteen thousand two hundred twenty four (talk) 05:30, 16 January 2026 (UTC)

Thanks for looking into this. Agreed on the perm, although I have no appetite for a different venue right now. As for the cleanup, I'm optimistic that it won't be that bad (I was worried when I saw 30k+ contribs). I am pretty confident the apparent LLM use started recently. Now that I have the AINB tracker installed I'll create the tracking page tomorrow. NicheSports (talk) 07:43, 16 January 2026 (UTC)

Yikes, this is a bummer. I agree there should be more women's sports pages condidering how extensive our men's sports coverage is, but this is not the way to do it. I'm also concerned that prior to AI, Hmlarson copy-pasted text from press releases and organization websites and such. Thanks for making the cleanup page. I might try to help chip away at that list soon elchupacabra (talk) 04:00, 26 February 2026 (UTC)

Is anyone able to check their recent edits? I see a few m dashes, and 22 Feb seems LLM-generated Kowal2701 (talk, contribs) 23:01, 2 March 2026 (UTC)

The prose of that edit seems likely, I fixed an issue with an incorrect ref, but aside from that the edit appears accurate. Spot checks of other recent edits don't show any obvious problems, certainly nothing on the level of the now-deleted Draft:PWHL Takeover Tour. fifteen thousand two hundred twenty four (talk) 15:38, 3 March 2026 (UTC)

S-Aura

S-Aura (talk · contribs · deleted contribs · logs · filter log · block user · block log)

I've tagged several galaxy articles created in December 2024 as AI-generated: NGC 1162, NGC 1163, NGC 1164, NGC 1166. They were full of incorrect information including hallucinated references, I've done some work to clean them up but they should probably just be nuked and rewritten. Other pages created by the same user (S-Aura, now blocked) should be looked into. SevenSpheres (talk) 19:01, 16 January 2026 (UTC)

SevenSpheres, you can request speedy deletion using WP:G15 if you don't fancy rewriting/stubifying them Kowal2701 (talk) 13:57, 17 January 2026 (UTC)

Sugar Laden

Sugar Laden (talk · contribs · deleted contribs · logs · filter log · block user · block log)

User is creating multiple articles incredibly quickly with non-existent categories and markdown. Need these to either be cleaned-up or deleted. – LuniZunie_(talk) 15:01, 18 January 2026 (UTC)

sorry maam Sugar Laden (talk) 15:02, 18 January 2026 (UTC)

Seems like User:~2026-37658-6 is also them per WP:DUCK, as the redirects made by Sugar Laden were converted into AI generated articles by the temp user. – LuniZunie_(talk) 15:13, 18 January 2026 (UTC)

Brought to ANI but still need cleanup. – LuniZunie_(talk) 15:19, 18 January 2026 (UTC)

Editor blocked, three pages remain: Keiko Necesario, Gaano Ko Ikaw Kamahal (deleted) and ~~Draft:Pasko Na, Sinta Ko~~ (G15 tag removed for now) are all G15 tagged. fifteen thousand two hundred twenty four (talk) 17:39, 18 January 2026 (UTC)

Gaano Ko Ikaw Kamahal has been G15'd. SuperPianoMan9167 (talk) 17:57, 18 January 2026 (UTC)

BassiStone cleanup

BassiStone (talk · contribs) was community banned for using AI/LLM on various articles over the past few months. To make a long story very short, I only got involved in this because, as a regular on the Britney Spears article, I checked their contributions to Spears's 1999-2001 personal life section and on Swan Lake and some of these edits were in fact created via AI-generated information.

Given that, along with the fact that any edit by a banned user (good or bad) would be reverted (I have reverted some of them myself), I have been considering a potential cleanup project to weed out any AI/LLM issues on the articles the user has edited as a whole. sjones23 (talk - contributions) 17:25, 19 January 2026 (UTC)

along with the fact that any edit by a banned user (good or bad) would be reverted – The shortcut name of WP:BANREVERT is a little misleading. Banreverts only apply to edits made in violation of a ban or block, BassiStone's edits do not qualify since they were made prior to any ban or block. fifteen thousand two hundred twenty four (talk) 20:10, 19 January 2026 (UTC)

If there was a ban evader in general, we would revert them on sight and report it.

However, as what you pointed out, BassiStone's edits don't count since he’s not currently a ban evader. With the ban evader statement out of the way, one of my main points in this cleanup suggestion is that we should consider resolving the AI/LLM issues. At this point, should we file a cleanup request? sjones23 (talk - contributions) 22:21, 19 January 2026 (UTC)

Tracking subpage created, 48 articles await closer review. Only edits after 23 June 2025 were considered as this appears to be when more obvious LLM use began . fifteen thousand two hundred twenty four (talk) 01:59, 20 January 2026 (UTC)
Thanks. As a side note, the Britney Spears article AI issue was already dealt with a few weeks ago. sjones23 (talk - contributions) 06:05, 20 January 2026 (UTC)

Order of the Knights of the Fiery Cross

Just removed an ChatGPT-generated image from this article; a lot of the text seems likely to be AI-generated as well. Omphalographer (talk) 03:59, 29 January 2026 (UTC)

The article prose largely seems human-written, with a couple of exceptions. I removed some of it, other parts I wouldn't want to touch without a full source review. I would say the article is largely fine, or at least doesn't give any glowing red flags, but I'll leave this tagged as cleanup requested in case anyone with a little more time wants to do a full source discrepancy check. Athanelar (talk) 17:46, 29 January 2026 (UTC)

Pretty sure it's all AI. From late 2025 on AI text presents differently, although this does have a few straggler signs -- emphasizing which is one of the last vocab holdouts from the GPT-4o era; the See Also section of random high-level crap, some stuff that seems like faint vestiges of the "undue emphasis on coverage" smell, etc. Plus just common sense that if you're already adding an AI-generated image you're more likely to use AI for text as well. GPTZero agrees with me on this fwiw.

Sourcing is weird. Haven't found any major hallucinations in the public ones, but it's kind of all of the place. Source #7 is a blog post with screenshots of the newspaper articles from #8, #9, and #10, but which gets cited for which claim seems arbitrary. contemporary reports in the United States variously described "thirty or forty" arrests is cited to #8 and #9, but it's a direct quote from #8. Which is only one article and not "variously"... but source #7, which isn't cited, does mention that "similar articles appeared in newspapers from throughout the region." (Although it doesn't mention what they are, their contents, or whether it's the same AP article syndicated.) There's some stuff that seems like it could be synthesizing the various articles, but without access to the books/journals, I can't confirm that. Gnomingstuff (talk) 00:30, 30 January 2026 (UTC)

AI-generated articles about Ugandan dances

I've come across several articles about Ugandan dances that seem AI-generated: Dingi dingi, Edonga, Kadodi dance, Ekizino, Amaggunju, and Akogo. Especially the sections about performance and significance show a lot of WP:AISIGNS and some of the sources the articles use don't exist. Cicada1010 (talk) 16:11, 29 January 2026 (UTC)

Just read through these. They are likely LLM-assisted but they do not seem to be wholly LLM-generated. Either way, they aren't encyclopedic and frequently violate WP:NPOV in addition to the sourcing issues you identified. Many of the editors who have significantly contributed to these articles seem connected to the very unfortunate #MEUG25 drive, which led to thousands of often LLM-assisted edits in a short time period. See this ANI thread WP:ANI § Disruptive editing using #MEUG25. We could create a cleanup case for that entire drive, although there are a huge number of edits to review. NicheSports (talk) 19:20, 30 January 2026 (UTC)

Yeah agreed - it seems like part of them may be translations and part original generation, and maybe worked on by multiple people in tandem, given the sudden swerves from generic AI-sounding paragraphs to more specifically grounded details. Most of the original versions are ~2024 by MichealKal it looks like. Gnomingstuff (talk) 22:24, 30 January 2026 (UTC)

Should we just tag all the affected articles and open a cleanup tracker? –LaundryPizza03 (d c̄) 01:17, 6 February 2026 (UTC)

AI-generated articles on Alaskan islands by Infinitywiki2

Hello! This user seems to be writing articles using LLMs. Examples include Aghik Island (Alaska), Abalone Island (Alaska) or, more egregiously, this article, which is fully devoid of content: Aalus (Alaska). It does seem, however, that they started generating their content alphabetically, meaning there is still ample time to intervene. Take care and thanks for the project! --BasicWriting (talk) 15:27, 30 January 2026 (UTC)

These are odd, they don't even say where the islands are. CMD (talk) 15:44, 30 January 2026 (UTC)

I reached out on their talk page and they responded favorably. Thanks for catching this before it became a bigger problem. -- LWG ^talk ^(VOPOV) 15:50, 30 January 2026 (UTC)

Awesome, and thank you! BasicWriting (talk) 15:55, 30 January 2026 (UTC)

All of these have been moved to draft. I removed the AI generated fluff. There is useful information on cebuano wikipedia generated by Lsjbot, but nothing more than coordinates. -- Recon rabbit 21:11, 30 January 2026 (UTC)

One article was not draftified, but has an open AfD at Wikipedia:Articles for deletion/Alice Island (Alaska). fifteen thousand two hundred twenty four (talk) 01:54, 31 January 2026 (UTC)

Wiki Education's report on LLM use by students

Came across this post on WP:ENB: Generative AI and Wikipedia Editing: What We Learned in 2025. It's worth a look. Main takeaways appear to be:

Copy/pasting LLM output bad
Problem is much more about real citations that fail verification than fake citations
They're using Pangram to do LLM detection (the post includes some evaluation thereof) and sending out alerts to students/staff when student contributions test positive (staff removes content if students don't)
They encouraged students to use LLMs just to identify content gaps and identify sources
Include some suggestions for Wikipedians at the end.

— Rhododendrites ^talk \\ 18:15, 30 January 2026 (UTC)

Back in my day, submitting work you didn't write yourself got you a zero on the assignment, if not more serious disciplinary measures. Thebiguglyalien (talk) 🛸 19:16, 30 January 2026 (UTC)

There's a topic about this on the main project talk page, probably better to continue the discussion there than on the noticeboard Gnomingstuff (talk) 22:25, 30 January 2026 (UTC)

Skyerise AI Cleanup?

Back in April 2025, Skyerise (talk · contribs) (formerly known as Yworo), a longtime and prolific user for nearly 20 years, was indefinitely blocked for civility issues per Wikipedia:Administrators' noticeboard/IncidentArchive1185#Skyerise and civility. In October, some concerns were raised on Gnomingstuff's talk page about Skyerise's potential AI/LLM issues.

In any case, is a cleanup necessary at this point? sjones23 (talk - contributions) 07:44, 2 February 2026 (UTC)

Yeah they disclosed using AI at one point, so it's not "potential", but I don't have the diff handy. I feel like I've run across many of their articles independently just through looking through AI tells but have not gone through their whole history because I do not have the stomach for it.

Obviously this only applies to edits they made after LLMs, oftentimes they returned to the same article in 2024 after editing it in 2019 or something. Gnomingstuff (talk) 23:42, 2 February 2026 (UTC)

I saw the diff buried in the discussion on their talk page: Special:Diff/1233685221. Thebiguglyalien (talk) 🛸 23:58, 2 February 2026 (UTC)

Burn it all and start over again. I was initially apprehensive about accepting their AI use but it's far beyond reasonable doubt. As previously discussed, it's not like I'm attached to their work in particular. wound theology ◈ 03:12, 3 February 2026 (UTC)

That's about 20,000 edits since 2023 Kowal2701 (talk) 17:47, 6 February 2026 (UTC)

Maybe we can start the cleanup effort? sjones23 (talk - contributions) 11:37, 8 February 2026 (UTC)

User:BookOfDisquiet

Serial AI user, silently removes tags from articles despite the AI issues being obvious, the usual stuff Gnomingstuff (talk) 23:34, 3 February 2026 (UTC)

I've left them a message Kowal2701 (talk) 20:20, 4 February 2026 (UTC)

@Gnomingstuff I am editing that article, and noticed that "riparian" was used a number of times in the article. Is that a known "LLM word" or whatnot? Katzrockso (talk) 23:45, 7 February 2026 (UTC)

not to my knowledge either way Gnomingstuff (talk) 07:46, 8 February 2026 (UTC)

Riparian is a word with a specific meaning typically used in ecology, so I would say no. "LLM words" are mostly related to editorialization or attribution. Sure looks like this editor is misusing LLMs though, including for edit summaries NicheSports (talk) 08:47, 8 February 2026 (UTC)

User:Cdkvsdph

This user reverted draftification on some LLM-generated articles and removed template on one, so bringing them here as AfD is not likely the solution for all of them.

Capture of 20 Jewish Youths by Khalid ibn al-Walid (634 CE): Only one source is cited for central claim, and it's an Arabic book not accessible for machine translation. Extensive WP:OR and WP:SYNTH. Several passages, especially "Significance", bear hallmarks of LLM-generated prose.
2026 Boko Haram Sabon Gari killings. Numerous claims fail verification.
Apakolu Massacre. As with the "20 Jewish youths" article, most of the article and sources are background to puff out an article about a lightly sourced event. The source used to cite the central claim fails verification on several points; it's an article about an attack on Musenge, not Apakolu, and it does not mention the other place names the article does. The only source that does mention these attacks attributes the news to a human rights group.
2026 Chabad car ramming attack. Likely a notable topic, but prose seems AI-inflected and the citations don't line up with the material they're supposed to source. For example, the source for The attack occurred amid a broader rise in reported antisemitic incidents in the United States, prompting renewed calls from community leaders for increased security at houses of worship doesn't say anything about anti-semitic incidents or increased security needs.
Draft:Kharab_Ashk_shelling. Once again, the central claim of an attack on Kharab Ashk does not appear in two of the four the sources used to make it (, ), and the "Significance" section jumps out as AI-generated.

The difference between the smooth prose of the user's article contributions and the rocky prose of talk page comments and edit summaries also suggests there's some LLM use. Thanks for taking a look (and letting me know if I'm off-base in my assessment of AI's role here). Dclemens1971 (talk) 03:34, 4 February 2026 (UTC)

This LLM-generated text has been collapsed and should be excluded from assessments of consensus.

The following discussion has been closed. Please do not modify it.

Capture of 20 Jewish Youths by Khalid ibn al-Walid (634 CE).

The concern raised here conflates sourcing limitations with policy violations. Wikipedia explicitly permits the use of non-English and non-digitized sources (WP:RS), and machine-translation accessibility is not a requirement. Early Islamic history frequently relies on single surviving Arabic chronicles, and the presence of one reliable historical source for a narrowly described event does not constitute WP:OR or WP:SYNTH absent the introduction of novel analysis, which has not been demonstrated here. If particular passages—such as the “Significance” section—are felt to overstate conclusions, that is a matter of weight and phrasing suitable for trimming or qualification, not draftification or deletion. Allegations of “LLM-generated prose” are not actionable under policy and do not establish unreliability.

2026 Boko Haram Sabon Gari killings.

The claim that “numerous statements fail verification” indicates citation gaps rather than non-notability or fabrication. Wikipedia policy requires that challenged material be supported by reliable sources (WP:V), not that all details be independently confirmed at the time of article creation. Reporting on Boko Haram violence often emerges through fragmented coverage, NGO reporting, or evolving casualty figures, all of which are routinely handled through attribution and subsequent cleanup. The appropriate remedy is to tag, source, or remove individual statements, not to treat the article as inherently invalid.

Apakolu Massacre.

The criticism here centers on alleged location mismatches and reliance on a human rights group for attribution. Neither issue is disqualifying. Attribution to human rights organizations is explicitly acceptable for conflict reporting, particularly in regions with limited on-the-ground journalism. If a cited source discusses nearby or related attacks rather than the precise village name used in the article, that suggests a need for clarification or renaming, not an inference of fabrication. Background context is permitted under WP:SUMMARYSTYLE and does not constitute “puffing” when clearly distinguished from the event itself.

2026 Chabad car ramming attack.

This topic appears facially notable, and the issues identified relate to citation alignment rather than substance. A sentence whose citation does not support a broader claim about antisemitic trends or security concerns should be corrected or removed, which is a routine editorial matter. Such mismatches are common across Wikipedia and are addressed through normal cleanup processes. They do not undermine the validity of the article as a whole nor suggest bad faith or automation.

Draft: Kharab Ashk shelling.

The concern that some sources do not explicitly mention Kharab Ashk again reflects citation precision issues, not hoaxing or original research. Conflict reporting often references clusters of villages or shifting frontlines, and early coverage may be imprecise. The solution is to tighten sourcing, adjust wording, or remove unsupported claims. Assertions that a “Significance” section appears AI-generated are stylistic judgments and not policy-based grounds for draftification absent demonstrable factual errors.

General allegation of LLM use.

Finally, the suggestion that differences between article prose and talk-page comments imply LLM usage is speculative and not probative. Wikipedia policy does not prohibit AI-assisted drafting, and writing style is not evidence of unreliability or misconduct. Editors routinely write more formally in articles than in discussions, and WP:AGF cautions against inferring intent from stylistic differences. Content should be evaluated on verifiability and sourcing, not presumed authorship method. Cdkvsdph (talk) 23:48, 5 February 2026 (UTC)

I've reported this at Wikipedia:Administrators' noticeboard/Incidents#Using an LLM to justify inappropriate LLM use. Thebiguglyalien (talk) 🛸 00:39, 6 February 2026 (UTC)

Is anything you disagree with LLM? Cdkvsdph (talk) 00:49, 6 February 2026 (UTC)

Editor is now blocked for Repeated use of LLM/AI to improperly create/write articles and then denying its very obvious usage. fifteen thousand two hundred twenty four (talk) 02:17, 6 February 2026 (UTC)

Cleanup page created HurricaneZeta^C 01:28, 6 February 2026 (UTC)

FYI, Jklkjkj has moved most of these back to mainspace. SPI has been opened. Dclemens1971 (talk) 03:53, 8 February 2026 (UTC)

Courtesy link: Wikipedia:Sockpuppet investigations/Jklkjkj. fifteen thousand two hundred twenty four (talk) 04:14, 8 February 2026 (UTC)

Blocked as a confirmed sock of Famous editor123456. fifteen thousand two hundred twenty four (talk) 00:47, 9 February 2026 (UTC)

Tttg2

Tttg2 (talk · contribs)

Suspect LLM usage based on erroneous, plausible-sounding additions:

Created a hoax draft about a fictional species, Draft:Erica Krogseter.
At : Information does not match source. Don't have the time to correct this addition.
At : Matched to , which does not contain any of the information except height.
At : Added a claim, not supported by sources, that this is a Nazi propaganda song. No support for the claim that the flower in the song is specifically Erica tetralix. Reverting user suspects LLM use in .

–LaundryPizza03 (d c̄) 05:38, 5 February 2026 (UTC)

Correction: This is Tttg2 (talk · contribs), not Ttg2 (talk · contribs). –LaundryPizza03 (d c̄) 07:07, 5 February 2026 (UTC)

User blocked due to cross-wiki sockpuppetry discovered at Wikipedia:Sockpuppet investigations/Tttg2. –LaundryPizza03 (d c̄) 01:15, 6 February 2026 (UTC)

Presumably the contributions of their sock Blu sweet roses (talk · contribs) also need to be scrutinized for LLM content? Hemiauchenia (talk) 01:23, 6 February 2026 (UTC)

Yes, including possibly on other wikis, where they likely used machine translation to post the hoaxes (Erica krogseter and a fictional philosopher named Yuri Giliboie that didn't have an enwiki draft made) cross-wiki. –LaundryPizza03 (d c̄) 01:28, 6 February 2026 (UTC)

Their other sock, Akayluhu (talk · contribs), has only one extant global edit: at it:Università della Lorena. –LaundryPizza03 (d c̄) 01:30, 6 February 2026 (UTC)

Akayluhu created an article about the philosopher at no-Wiki, which was deleted. I have already cleaned up all edits across all language versions where Yuri Glilibloie and/or Mr. Krogseter and/or Erica Krogseter (also as a person, not just a plant species) was mentioned, and a couple of vandalism edits were reverted as well.

A lot of the edits were "newcomer tasks" by adding links in articles. Some edits could very well be AI content, I didn’t look further into those edits earlier this week. Probably something that should be done. 1000mm (talk) 02:39, 6 February 2026 (UTC)

Mohamed0612

This user has created 37 articles, seen here, all of which appear to have been created by an LLM. Additionally, there are AI-generated talk page messages assessing the article's class, like here. There have also been some major changes to other articles in the name of WP:NPP. I already warned them on their talk page with a {{Uw-ai1}} and on my own talk page here, but the issues are widespread enough that I think they warrant cleanup as it stands even if they don't continue to use AI. HurricaneZeta^C 17:39, 6 February 2026 (UTC)

Mohamed0612 (talk · contribs) Kowal2701 (talk) 17:42, 6 February 2026 (UTC)

Hi, I am Mohamed0612. I am sorry for the confusion. I am a real person, a researcher from Fez, Morocco.

I admit I used language tools to help me with English because it is not my first language, and I wanted the articles to look professional. I see now that this was a mistake and made my work look like AI.

However, please check the sources. I did all the research myself using Moroccan official documents (SGG) and local news. An AI cannot find these specific PDF decrees or know the latest 2025/2026 details about our local faculties in Fez.

I promise to stop using any AI or language refinement tools for my future edits and Talk page messages. I will write in my own simple English. Please don't delete the articles; they provide important information about Morocco that was missing here. I am happy to show the manual research process for any article.Mohamed Filali (talk) 17:53, 6 February 2026 (UTC)

Would you commit to go back through your articles and ensure that every sentence matches what the source says @Mohamed0612? Katzrockso (talk) 23:26, 7 February 2026 (UTC)

@Katzrockso: Yes, I commit to this. I am already checking my articles one by one manually. I am a researcher and I am sure about the facts from SGG and official sources, but I will fix the English phrasing to match the sources exactly. I am glad that experienced editors like MPGuy2824, Rodw, R Prazeres, and Mariamnei already helped or reviewed my work. I will do all the cleanup myself from now on. Thank you for the chance. Mohamed Filali (talk) 09:19, 8 February 2026 (UTC)

User:NiseEdits

Has made many edits with the usual WP:AISIGNS. I asked them about this back in September but they didn't answer and are continuing to edit. Gnomingstuff (talk) 19:25, 6 February 2026 (UTC)

trying something new, apologies if this is dumb In their 8,600 edits, they've only made 34 edits to article talk pages, and 24 to user talk pages. They've been warned about LLM-use twice (with good explanations) to which they didn't respond. Is a mainspace block appropriate (to at the very least get them to communicate)? Kowal2701 (talk) 20:55, 6 February 2026 (UTC)

was thinking {{Admin help}} could save time and energy from lengthy ANI threads Kowal2701 (talk) 20:59, 6 February 2026 (UTC)

Answering {{Admin help}}: The last messages were in September and I am seeing little AI misuse in the user's subsequent edits, of which there are quite a lot. Please could you try re-engaging with the user? Arcticocean ■ 18:24, 7 February 2026 (UTC)

Can do, thanks Kowal2701 (talk) 18:32, 7 February 2026 (UTC)

They appear to have ignored my message, I guess we keep an eye on it Kowal2701 (talk, contribs) 18:01, 11 February 2026 (UTC)

The edit that prompted posting this was from less than a month ago, I had just happened to stumble across it independently. Gnomingstuff (talk) 06:23, 8 February 2026 (UTC)

Discussion at Wikipedia:Administrators' noticeboard/Incidents § Eastmain's use of generative AI in mainspace and draftspace

You are invited to join the discussion at Wikipedia:Administrators' noticeboard/Incidents § Eastmain's use of generative AI in mainspace and draftspace. –Novem Linguae (talk) 13:44, 7 February 2026 (UTC)

User:Guile's Theme

User has been accused of bludgeoning an rfc. They left a punchy thread on my talk page that attracted the attention of two admins. They said it was pejorative and demeaning to be accused of using AI, but then they posted a source with a ChatGPT utm in the rfc and then deleted the utm from their comment. This looks not great to me, but I’m not very experienced with this stuff. Mikewem (talk) 16:38, 7 February 2026 (UTC)

Their next comment in the thread is also sourced to ChatGPT, but maybe they haven’t noticed that one. Mikewem (talk) 17:08, 7 February 2026 (UTC)

There is always the possibility that one is simply using an LLM to search for sources, which would result in those UTM parameters. Such LLM use is not widely seen as a transgression, as far as I know. If there are WP:AISIGNS that an LLM is being used to generate article text or talk page comments, then more people might take issue with it, but it usually requires accumulating enough signs to overcome a plausible deniability threshold. And, yes, I have also found that probing possible AI use makes people defensive, sometimes to the point of being accused of violating AGF or civility guidelines. Einsof (talk) 18:15, 7 February 2026 (UTC)

It was an admin who was doing the probing, but my best assumption is that GT did not know they were an admin or did not know how to check or did not know that they could check. Mikewem (talk) 18:42, 7 February 2026 (UTC)

User:Semangathidup45

Semangathidup45 (talk · contribs) has been around for 5 months and in that time has accumulated 20 templated or custom warnings on their talk page. They have responded to one . 4 of these warnings were for LLM use (examples here ) and 2 were for unattributed translations . I warned them about both two days ago after seeing this . They have had multiple articles draftified, including one for potential LLM generation. Despite all of this they continue to use machine assistance for their editing, see here for some typical LLM prose Their wedding reception was held according to Gorontaloan customs, reflecting the heritage of the Habibie family's ancestral home. Yesterday they created Draft:Alwi Abdul Jalil Habibie in mainspace which contains the phrase He was a respected figure who owned rice fields, coconut plantations, and cattle and horse ranches in Batudaa Village. This was draftified on sourcing/notability grounds. I'm not sure how well this user understands English , so they may not understand the warnings and talk messages they are receiving. NicheSports (talk) 09:19, 8 February 2026 (UTC)

I started cleaning up their mess but was chastised by admin GiantSnowman for my wholesale reverts at Talk:Arnór Ingvi Traustason#AI-generated text. That article still cites a nonexistent source that Semangathidup45's AI chatbot completely hallucinated into existence. I have unfortunately not bothered to keep up with the AI text Semangathidup45 has added to Wikipedia since then. I'll state what I said there: They were adding AI-generated text into wildly unrelated articles at rapid speed. There's no chance they have the knowledge required to understand each of the subjects and ensure what they were adding was correct. If the person making the AI additons has proven through their edits that they are not reviewing the AI-generated text, which Semangathidup45 has (check the edits), then it is not incumbent upon me to review it. Nor is it my responsibility to clean it up to standard so that it can be retained in the given article. There is no place for unreviewed AI on Wikipedia. This isn't Grokipedia. Οἶδα (talk) 19:21, 8 February 2026 (UTC)

Sitewide consensus seems to agree with you (see Wikipedia:Administrators' noticeboard/IncidentArchive1213#Presumptive revert proposal for example). Also, any editor restoring the content assumes responsibility for its verifiability and needs consensus if the restoration is challenged. Thebiguglyalien (talk) 🛸 00:09, 9 February 2026 (UTC)

Sorry that happened Οἶδα. Presumptive reversion of such edits typically works but I've mostly used it after admin action, which seems needed here NicheSports (talk) 04:29, 9 February 2026 (UTC)

@NicheSports: Am I hallucinating or is this user now socking with PMGR1968 (talk · contribs)? There are obvious similarities and, after making edits exclusively to football topics, of all the film articles they could have chosen to edit on Wikipedia they start with an obscure Japanese film article that I also just so happened to create? If true, they're getting better at paraphrasing the AI. The article in question is I, the Executioner (1968 film), and it appears the AI pulled from the corresponding article on jawiki. Οἶδα (talk) 07:00, 27 February 2026 (UTC)

Hmmm. Per they had a previous account, so it is possible. However I think these are not the same person. Semangathidup45 edits many Indonesian-related topics in addition to obscure European football articles. Their username also suggests a language from that part of the world. PMGR1968 has not edited anything related to Indonesia. PMGR1968 also responded on their talk page which Semangathidup45 rarely if ever does. These don't match either. I would not file at SPI in this case. However we should ask PMGR1968 about their previous account. I am also very curious about how they generated this , which entirely fails verification but has essentially zero AISIGNS. If this was actually AI-generated or assisted it would be evidence for one of my fears, that the latest LLM model releases are capable of producing essentially undetectable article prose. This also explains why I have been less active recently. NicheSports (talk) 14:17, 27 February 2026 (UTC)

At a certain point, the responsibility for these problems is shared by the people who add the umpteenth warning or are otherwise aware of the situation without seeking administrative action. Which I suppose now includes me as well. Thebiguglyalien (talk) 🛸 00:10, 9 February 2026 (UTC)

User:DP-Author

request a review of AI slop?. Relevant diffs:

ButterflyCat (talk) 16:04, 10 February 2026 (UTC)

It's a strange editing pattern but I can't say it immediately strikes me as LLM output (particularly a diff that drops in new text already with a backdated {{cn}} tag). I've dropped a note on their talk page. Einsof (talk) 16:29, 10 February 2026 (UTC)

Thanks for taking a look(: i agree it’s an unusual pattern; I raised it only for review. Appreciate you leaving a note on the user’s talk page. ButterflyCat (talk) 16:34, 10 February 2026 (UTC)

RfC arguments by DangerousEagles

Concern was expressed that User:DangerousEagles may have used an LLM to generate arguments in an ongoing RfC. Another argument was posted to an ongoing high-profile RfC. Some previous writing samples for comparison: . I'd appreciate the attention of uninvolved eyes to determine whether these arguments should remain. Uhoj (talk) 01:28, 11 February 2026 (UTC)

While my writing does seem at a glance like AI, I can assure you that I am not AI. With this oversaturation of AI content, it is incredibly saddening that human writing is unintelligible from AI writing. I, simpl put, am a human being.

Thanks and regards,

@DangerousEagles DangerousEagles (talk) 18:02, 11 February 2026 (UTC)

typo: simpl - simply. DangerousEagles (talk) 18:03, 11 February 2026 (UTC)

Hi DangerousEagles, please read WP:BLUDGEON. In an RfC it is obviously fine to weigh in and to engage with other editors' opinions, but not to the extent of posting long responses to every editor you disagree with. Einsof (talk) 18:43, 11 February 2026 (UTC)

I'm sorry, I haven't read WP:BLUDGEON before, thank you for showing me this. It's unfortunate that my human-written text is classified as AI, even though it was purely written by me. DangerousEagles (talk) 15:38, 13 February 2026 (UTC)

@Uhoj can you please provide diffs for the RfC comments you'd like a second opinion on? NicheSports (talk) 18:18, 11 February 2026 (UTC)

@NicheSports Thanks for looking into it. Here you go:

And here are diffs of previous writing for comparison: --Uhoj (talk) 20:13, 11 February 2026 (UTC)

yeah this is pretty blatant, sorry Gnomingstuff (talk) 23:06, 11 February 2026 (UTC)

I'd concur with the specific difs listed likely being LLM generated. WinstonDewey (talk) 17:09, 12 February 2026 (UTC)

I actually believe DangerousEagles on this one... I was going to agree LLM at first but reading more of their contributions they are just extremely precise and specific and become way overly verbous, and in some cases they start replying to every new point in a discussion repeating their points back. Cleanup + collapsing was badly needed but I don't think you can outright toss all their arguments out elchupacabra (talk) 13:13, 13 February 2026 (UTC)

User:Tat02948

Obvious LLM copypaste here: Special:Diff/1337727860.

Has created two articles through AFC: HAUI, Peter Hinton-Davis ScalarFactor (talk) 03:38, 11 February 2026 (UTC)

I have always used Wikipedia until this morning. Google AI has shut me out. When I finally got in, Wikipedia’s failed to find an author I have always found on wiki. I’m a regular donor. Please help.

— Preceding unsigned comment added by ~2026-93391-3 (talk) 14:50, 11 February 2026 (UTC)

I removed the cleanup request since it probably doesn't apply here. @~2026-93391-3, what author are you looking for? -- Recon rabbit 16:48, 11 February 2026 (UTC)

This is a place for discussing removal of AI-generated slop. If by "author" you mean someone who writes wiki articles then type user:username into the search bar where "username" is the person you're looking for.

Reddit, Quora, or Gemini Community would be better places to ask questions about problems with Google AI. Uhoj (talk) 17:04, 11 February 2026 (UTC)

User:Treepalozaedits

Same old story: pumps out a bunch of obvious AI text via Newcomer Tasks, is asked whether they used AI on their talk page, uses AI to say they didn't use AI, continues to pump out obvious AI text. Gnomingstuff (talk) 22:04, 13 February 2026 (UTC)

Every single edit I checked from their contributions contained nonsensical citations that no reasonable human could possibly think supported the accompanying text. I don't have experience with reporting people like this to ANI but they really need to stop inserting junk content, regardless of what their LLM workflow is. -- LWG ^talk ^(VOPOV) 22:42, 13 February 2026 (UTC)

Even their user page is AI generated... I just finished reading through their contribs and virtually every edit failed verification. I cleaned up all but a couple of edits that were harder to evaluate. They haven't edited since January, so hopefully they won't continue this behavior. -- LWG ^talk ^(VOPOV) 23:38, 13 February 2026 (UTC)

oh thanks for getting to those, was going to do them later but have been going back and forth between like 6 batches Gnomingstuff (talk) 01:45, 14 February 2026 (UTC)

New article at DYK

If anybody would like to visit Template:Did you know nominations/National bank veto regarding national bank veto, I am concerned about a new article with content that I can't square with a plausible authorial process. (Cites different websites' versions of same historical text without rhyme or reason. Cherry-picks a fine point to cite an argumentatively unimportant proposition.) Since I'm (we're?) still figuring this out, I wasn't sure if I should just drop {{AI-generated}} on it, or seek input first. TheFeds 06:51, 15 February 2026 (UTC)

Yeah this is pretty clearly AI -- thanks for flagging it. Did a quick standard spot check of AI-ese, standard issues turned up:

Jackson's message was widely circulated in Democratic newspapers, which emphasized his arguments about monopoly, privilege, and the influence of foreign stockholders. Cited to SparkNotes, first of all, which only says that "[Senator] Benton convened a House investigation that restated many of Jackson's complaints and publicized them in newspapers across the country." Doesn't say what complaints those were, what newspapers those were, or how widely they were circulated ("across the country" could mean anything; our Bank War article seems to suggest the pro-charter side had more newspaper presence but I don't have in-depth knowledge here). Also, it's SparkNotes.
The diversity of interpretations reflects the Bank Veto's enduring significance as one of the most consequential presidential actions of the early republic. Does the citation (available on wikipedia library) say that? No, obviously. The closest it comes is saying that "some critics blame [the veto] for creating decades of chaos" right before arguing "there are strong reasons to suspect otherwise." I guess that counts as "diversity of interpretations" but everything else is AI slop. (Also, this comes from a libertarian think tank journal, so not really the most unbiased source.)
President Jackson opposed the bank... contributing to the broader conflict over the Bank that occurred during his presidency. Not really a WP:V issue just a hilarious manifestation of AI's obsession with pointing out how everything "contributes to the broader" blah blah blah. No shit, the president's opposition the bank contributed to the conflict over the bank during his presidency? Huge if true!

Gnomingstuff (talk) 16:58, 15 February 2026 (UTC)

Ref 68, ostensibly meant to source an assertion about "later discussions about federal involvement in banking", is a four-page report on the composition and procedures of the UN Security Council. 🤡 Einsof (talk) 20:56, 15 February 2026 (UTC)

messaged Kowal2701 (talk, contribs) 21:18, 15 February 2026 (UTC)

Is this edit LLM-generated?

I believe this edit was LLM generated, although the user denies it. Traumnovelle (talk) 19:27, 16 February 2026 (UTC)

I can't conclusively say it was LLM-generated, but that edit does show signs of the problems that LLM edits tend to have (vague weasel-wording, overly promotional tone, gross liberties taken with the sources). It looks like further edits to the article have removed the problematic content so it might be a moot point now. -- LWG ^talk ^(VOPOV) 19:55, 16 February 2026 (UTC)

Can't say for sure but possibly everything but the "trails" slipup. It does the thing where it smuggles promotional content in via attribution ("the company provided [advertisement-like qualities"]), and even in a short space it has a lot of source-to-text integrity issues:

The service used electric vans to provide flexible, app‑based point‑to‑point transport - This is what I mean about the smuggling promotional content. The source only mentions that there's an app and that the vehicles are electric. That they're specifically vans is not mentioned -- the photo in the article appears to be a regular car -- and the corpospeak obviously isn't either.
Local Board members criticised its cost, limited uptake, and perceived competition with existing bus services. - The article mentions nothing about buses specifically, just "public transit networks."

Gnomingstuff (talk) 23:23, 16 February 2026 (UTC)

User:Ralphster7

Was wondering if someone else could take a look at this user's articles and give a second opinion on whether they're AI. No enormous smoking guns yet, but the things that give me pause here are A) the rapidfire speed that these things are being pumped out at, proportionate to the article size, such as this article being added 3 minutes after this one, and B) a pattern of absolutely everything and everyone "playing a key role," "contributing to _____" etc. throughout them, combined with a bunch of participials. Obviously this only applies to the recent stuff. Gnomingstuff (talk) 00:09, 17 February 2026 (UTC)

I can assure you that these are not LLM generated, though I have used to it to help neaten up certain sentences. I often copy and paste from sandbox, which explains the rapid pace of them being pumped out. The repetition is actually just cause I'm so used to writing all of these that those "linking sentances" just come into my head. Ofc I'm happy to explain / clarify anything. Ralphster7 (talk) 09:06, 17 February 2026 (UTC)

Thanks for the clarification on the speed, and this all makes sense -- the text seemed a lot more uniform than most straight up AI-generated articles. Gnomingstuff (talk) 17:21, 17 February 2026 (UTC)

(just to be clear, I feel pretty confident about the review you're doing of these.) Gnomingstuff (talk) 21:55, 17 February 2026 (UTC)

User:Robert Wellock

I recently stumbled upon the, errr, excuse me, rambling posts by that user that seem to be weirdly formatted (with bolding) and often contain incorrect claims (see for example my replies at Wikipedia:Articles for deletion/Mighty Warriors for identified errors). ZeroGPT gives me a score of 0%, so perhaps it's just the distracting overuse of bolding and known human errors, rather then LLM-talking, but stuff like "The nomination overlooks the Bibliographic Gravity of the subject's 25-year career" really look... weird. Bibliographic Gravity? To me, that's close to WP:DUCK. Or consider - they use of emoji 🦋 in text, and phrases like "innate systems-level thinker" and "industrious builder". Or consider their mainspace edits: removing a single pipe with gargantuan edit summary: "Standardising branding: Correcting "Hero-Quest" (hyphenated) to the single-word "HeroQuest" to align with the primary Milton Bradley trademark and original 1989 rulebook. Resolving visual "logo spacing" hallucinations while maintaining internal link integrity.". Third+ opinion would be welcome, as well as thoughts on whether anything should be done (rambling and making factual errors in discussion are not straight-out bannable offense, but there is the question of WP:NOTHERE). _{Piotr Konieczny aka Prokonsul Piotrus| reply here} 01:24, 18 February 2026 (UTC)

At the very least the edit summaries are AI-assisted (the giveaway is the [making trivial change] "to align with" xyz). I don't think these are big enough edits to really be a priority though, and their 2005 writing style doesn't really seem that far off from now. Gnomingstuff (talk) 04:31, 18 February 2026 (UTC)

User:Seungsahn

User:Seungsahn (talk · contribs · deleted contribs · logs · filter log · block user · block log)

New user who has made edits exclusively in the WP:CT/EE space (RfCs etc.) i.e. only those related to a recent wiki controversy. I noticed signs of LLM writing in their edits and sure enough a check at GPTZero (among others) tells me that they are extensively using LLMs for a lot of their edits. A sample of their edits and results from GPTZ: (99%), (100%), (100%), (88%), (85%), (80%), (80%), (100%), (87%). With the rest of their edits I also mostly get LLM/human results at GPTZ. Considering this I told the user to desist and diclose LLM usage (Talk:Kaja Kallas#RfC: Footnote in infobox birthplace, Wikipedia:Village pump (miscellaneous)/Archive 87#RFC: Baltic bios infoboxes question) but they have vehemently denied using LLMs, calling the accusations false. With the LLM usage being very very evident, I ask for others to weigh in here. Gotitbro (talk) 05:37, 19 February 2026 (UTC)

The general gestalt is there, a bit more concerning is what seems to be some WP:PGAME going on with the smaller edits (most of which have loose connections at best to what they claim to be doing, like this "changing the tone") Gnomingstuff (talk) 06:22, 19 February 2026 (UTC)

Chat hack

Should this be mentioned somewhere ? Yesterday, all my dreams... (talk) 22:36, 20 February 2026 (UTC)

Can I get a more transparent hyperlink? Einsof (talk) 01:46, 21 February 2026 (UTC)

What could be more transparent than that? It's a decent article about how easy it is to hack AIs to give bogus responses, like when spam first started appearing in emails because there were no guardrails in place. The points the article makes bear repeating in a guideline somewhere, about how an AI can give you false answers because it is easy to manipulate. ~Anachronist (who / me) (talk) 02:49, 21 February 2026 (UTC)

I meant a hyperlink that gives an idea of what website it resolves to. Einsof (talk) 03:08, 21 February 2026 (UTC)

Oh, I see what you mean, it's a Google redirect. Just click on it, it resolves to BBC News. Here's the actual link: https://www.bbc.com/future/article/20260218-i-hacked-chatgpt-and-googles-ai-and-it-only-took-20-minutes ~Anachronist (who / me) (talk) 05:18, 21 February 2026 (UTC)

Alas it is likely to become worse. Soon a few people will figure out that this is a new form of SEO and will charge for doing it. So if some ice cream shop wants to be recommended by AI it can be done in few hours for a few hundred dollars. Regarding Wiki where do you suggest a policy modification? Thanks Yesterday, all my dreams... (talk) 09:05, 21 February 2026 (UTC)

The only places I can think of adding this information might be in user talk page templates, such as Template:AfC submission/comments, used when declining an AFC draft that was LLM-generated. ~Anachronist (who / me) (talk) 09:53, 21 February 2026 (UTC)

I guarantee people have already figured this out. It isn't exactly a novel concept. Gnomingstuff (talk) 19:04, 21 February 2026 (UTC)

Just another example of why we need to be very heavy-handed against LLM-generated content to prevent it from infecting articles. Thebiguglyalien (talk) 05:37, 27 February 2026 (UTC)

Alas what is happening is zero, zero, zero. Yesterday, all my dreams... (talk) 11:22, 27 February 2026 (UTC)

Herbert A. Parkyn AI enhanced image

User:Erik Baas seems insistent on keeping an AI-“upscaled” image in this article. I would strongly prefer something that didn’t utilize artificial intelligence with all its notorious unreliability, for example this image I cropped from an ad on Commons. Maybe it’s not the greatest or prettiest but IMO AI should just be avoided wherever possible. Thoughts? Dronebogus (talk) 19:29, 24 February 2026 (UTC)

This appears to be a content dispute. Everybody is entitled to their person opinion about which image is better and why, but one option being AI is not a justification for bypassing the consensus process, nor for canvassing editors to the discussion. Thryduulf (talk) 20:04, 24 February 2026 (UTC)

I’m not canvassing; I just think the article is too low-traffic to get a second opinion naturally. Dronebogus (talk) 20:35, 24 February 2026 (UTC)

Then you should have posted a neutral notice at WP:3O, a suitable wikiproject or other venue (part of) whose purpose is for attracting attention to content disputes. Instead you posted a non-neutral notice to a strongly-opinionated board whose purpose is unrelated to content disputes - that is unarguably canvassing. Thryduulf (talk) 21:08, 24 February 2026 (UTC)

@Thryduulf are you aware of MOS:AIUPSCALE? Given that the picture in question seems to be against the MoS I don't understand how @Dronebogus's original post is canvassing (added) or a content dispute NicheSports (talk) 22:00, 24 February 2026 (UTC)

If the image is not compliant with the MOS guideline (which you will note does state "generally") then that is an argument that should made in the discussion and will be a strong factor in determining consensus. It has absolutely no bearing on whether a non-neutral summary of a content dispute at a non-neutral board not related to content disputes is or is not canvassing. Thryduulf (talk) 22:12, 24 February 2026 (UTC)

Also, the only purpose of this noticeboard is content issues. That's literally the entire point. Gnomingstuff (talk) 22:25, 24 February 2026 (UTC)

The purpose of this noticeboard is for reporting issues of AI misuse, not content disputes. Also, despite the name the guide linked in the instructions about the purpose of this noticeboard makes it clear it's related to issues of LLM-generated text, not disputes about whether or not to use an AN image. However, even if it were a relevant place to bring content disputes that doesn't excuse the blatantly non-neutral summary. Thryduulf (talk) 22:36, 24 February 2026 (UTC)

Hey @Thryduulf, a request. Can we get a little more WP:AGF and a little less wikilawyering here? Saying that image-related questions are out of scope for this noticeboard is a bit much. Yes, the original request could have been written more neutrally. But fundamentally, a user identified an MOS-violating AI upscaled image and reported it to the AI noticeboard. This seems reasonable, and even constructive. I think a better course of action than a black-and-white portrayal of this as canvassing/improper would have been to point out the relevant MOS guideline to both involved users and suggest to Dronebogus - maybe on their talk page - how to phrase the post more neutrally NicheSports (talk) 22:53, 24 February 2026 (UTC)

What Dronebogus did here was write a non-neutral summary of a content dispute they are involved in at a venue that is non-neutral with regards the dispute as framed. Whether they intended the post as canvassing or not, that is the textbook definition of canvassing. If using this board, intentionally or otherwise, for canvassing is not pointed out then other editors are more likely to (in good or bad faith) use the board for the purpose in the future which is not in anybody's interest. The issue is not one that requires the attention of this noticeboard, neutrally written or otherwise. Thryduulf (talk) 23:14, 24 February 2026 (UTC)

I am not sure that this is the best characterization of what happened, so let's try a third opinion I guess? @Admins willing to patrol AINB: can someone provide a third opinion on the issues discussed in this thread? Thanks NicheSports (talk) 23:20, 24 February 2026 (UTC)

Responding to the ping. This noticeboard is for assessing compliance with AI-related policies, guidelines, and community norms (including the MOS:AIUPSCALE style guideline), and for responding to content and conduct disputes related to AI use on Wikipedia. This venue is as neutral as the spam noticeboard (under the scope of WikiProject Spam), which is for assessing and responding to reports of edits in relation to the spam guideline and related community norms.

Dronebogus is allowed to start a discussion on this noticeboard indicating that they believe an edit to be in violation of the MOS:AIUPSCALE style guideline. When an editor starts a discussion on any noticeboard, instead of merely posting a notification on a noticeboard about another discussion elsewhere, the editor is allowed to express their point of view in the discussion they start. In short, I don't see the problem with Dronebogus's participation here. — Newslinger talk 06:44, 25 February 2026 (UTC)

Dronebogus is allowed to start a discussion on this noticeboard indicating that they believe an edit to be in violation of the MOS:AIUPSCALE style guideline. Except that isn't what they did. They came here to attract support for their position in a content dispute about which image to use on an article. Thryduulf (talk) 12:17, 25 February 2026 (UTC)

Um, no, I came here to ask for a third opinion because the other user isn’t responding. Dronebogus (talk) 14:25, 25 February 2026 (UTC)

Also responding to the ping. A notice about an issue with MOS:AIUPSCALE does fall under the scope of this noticeboard, although, being a single issue (rather than repeated AI use necessitating large-scale cleanup), it could have just as well been raised on the article's talk page. Content disputes, when the content in question may break existing policies, can still absolutely be in scope for noticeboards (e.g. Wikipedia:Biographies of living persons/Noticeboard or Wikipedia:No original research/Noticeboard). In this case, the applicable policy is:

AI upscaling software should generally not be used to increase the resolution or quality of an old or low-resolution image. Original historical images should always be used in place of AI upscaled versions. If an AI-upscaled image is used in an article, this fact should be noted in its caption.

The policy is quite straightforward, especially in the case of original historical images (such as this one), where the policy specifies they should always be used, so this should be clear-cut enough. Chaotic Enby (talk · contribs) 13:23, 25 February 2026 (UTC)

@Thryduulf: you don’t appear to be a regular contributor to wp:AIC why have you come to this noticeboard seemingly just to start fights about minutiae? Dronebogus (talk) 11:37, 25 February 2026 (UTC)

Firstly, I have this page watchlisted so that I can keep abreast of the AI issues that people are seeing, I didn't come here specifically for anything. Secondly, why does it matter who calls out canvassing as canvassing? Thryduulf (talk) 12:08, 25 February 2026 (UTC)

Are postings at WP:FTN canvassing? Kowal2701 (talk, contribs) 12:23, 25 February 2026 (UTC)

Potentially. I don't follow that board and have no recent direct experience but I do know there are accusations that it has (or at least had, I've not heard much either way recently) an extremely pro-sceptic culture. A non-neutral post asking participants of a board with a clear POV to support the poster in a clear content dispute is almost always going to be canvassing. A neutral notification alerting participants to potential issues with content by someone not involved in a dispute about that content is rarely going to be canvassing. Thryduulf (talk) 12:44, 25 February 2026 (UTC)

You are wasting everyone’s time with this one-against all Wikipedia:IDHT crusade. You are an admin; act like one instead of a Reddit mod. Dronebogus (talk) 14:23, 25 February 2026 (UTC)

Also, despite the name the guide linked in the instructions about the purpose of this noticeboard makes it clear it's related to issues of LLM-generated text

...so, in other words, a content issue. Text is content. Literally what are you even talking about? Gnomingstuff (talk) 05:38, 25 February 2026 (UTC)

@Gnomingstuff if you cannot understand the difference between a content dispute and the general issues with content that get discussed here then it's amazing you've not been the subject of a CIR block. As you haven't been, then I can only presume that your comments here are not the result of incompetence but something else. Quite what that is I don't know, but it's not a good faith reading of basic norms. Thryduulf (talk) 12:16, 25 February 2026 (UTC)

Hi! Please assume good faith, especially since Gnnomingstuff's position, while not unanimous here, is perfectly reasonable, and accusing them of having further undisclosed motivations is not the most charitable reading of the situation (which is that you two have different reads on what the exact scope of the board is). Chaotic Enby (talk · contribs) 13:16, 25 February 2026 (UTC)

Why are you being so hostile? What did I do to you? All I said was that this board is to discuss issues with content, which is something you have also said. Please point me to the policy that magically redefines "content dispute" to be something other than a dispute about content. It's not WP:CONTENTDISPUTE (which outright suggests asking relevant various WikiProjects and noticeboards). Gnomingstuff (talk) 14:57, 25 February 2026 (UTC)

Yeah, I don't get this either -- goodness knows Gnomingstuff, you and I don't always agree, and yeah, @Dronebogus, that wasn't a neutral notification, if you need a third opinion on AI-MOS issues, just use {{Please see}} in future, but the CIR block threat and comments about incompetence are just downright wrong and insulting. I'm very unimpressed, @Thryduulf. GreenLipstickLesbian 💌🧸 19:14, 25 February 2026 (UTC)

I don’t see how it’s not neutral or even why it has to be given AI is generally discouraged by policy. Dronebogus (talk) 19:16, 25 February 2026 (UTC)

Well, given that neutrality is subjective, it's really best to keep things and concise as possible. In general though, presenting arguments for one side (this particular use is against policy, you don't believe AI upscaling should be used, ect), but without giving the other party in the dispute a right to respond is not, from my perspective, a neutral notification, @Dronebogus. Even if you don't agree with that, then think about it a different way - {{Please see}} is the unofficial, widely-used, and reccomended way of notifying people. It is much harder for somebody to accuse you of canvassing when you use it. GreenLipstickLesbian 💌🧸 19:21, 25 February 2026 (UTC)

Erik Baas should’ve received a ping from me mentioning him. In any case I’ll be sure to use the “please see” template in the future Dronebogus (talk) 19:23, 25 February 2026 (UTC)

Thank you. GreenLipstickLesbian 💌🧸 19:24, 25 February 2026 (UTC)

I’ve pinged them; hopefully they’ll actually respond Dronebogus (talk) 19:27, 25 February 2026 (UTC)

Quite what that is I don't know, but it's not a good faith reading of basic norms. This comment is really out of line, especially considering how adamant you normally are that we should assume good faith in AI-related discussions. There's no reason whatsoever to doubt Gnomingstuff's good faith or competence here. -- LWG ^talk ^(VOPOV) 15:50, 25 February 2026 (UTC)

I agree, and it's disheartening to see a fellow admin doubling down on WP:IDHT WP:WIKILAWYERING. And then to bring up WP:CIR? Well, WP:BOOMERANG comes to mind. On the other hand, being here over 20 years (almost that long for me) I can understand that one can occasionally feel jaded, so I can assume good faith and chalk up Thryduulf's unbecoming lapse to that. I've felt that way too, moreso since this AI infection Wikipedia is experiencing.

The initial comment by Dronebogus could have been worded better, but there is no doubt that an AI cleanup issue was being reported, and this noticeboard is an appropriate venue for making such reports. I see an irrelevant mountain being made of a molehill here. The alleged content dispute could have been ignored and the report taken in the context of what this noticeboard is for. It's a pity that didn't happen. ~Anachronist (who / me) (talk) 06:53, 27 February 2026 (UTC)

User:Mikah Wikiana

Mikah Wikiana (talk · contribs · global contribs · logs · block log)

New account prolifically creating pages with LLM tells. See puffery in articles on Stunna Girl, Talia Jackson, Meg DeLoatch, also suspiciously verbose edit summaries. 11wx (talk) 09:16, 25 February 2026 (UTC)

I've gone ahead and added the AI-Generated template to the most obvious examples. Not sure what the process is for making a tracking subpage. 11wx (talk) 09:30, 25 February 2026 (UTC)

I cleaned up the AI slop in the three articles listed above. GPTzero assessed Stunna Girl as "written by a human and polished with AI", so it didn't qualify as a WP:NEWLLM violation that would merit a speedy WP:G15 deletion. ~Anachronist (who / me) (talk) 06:02, 27 February 2026 (UTC)

I have also issued a temporary block from creating articles in mainspace. Please go through this editor's mainspace creations and clean up the AI slop and puffery. ~Anachronist (who / me) (talk) 06:28, 27 February 2026 (UTC)

User:Real96kmaratha

Real96kmaratha (talk · contribs · deleted contribs · logs · filter log · block user · block log)

This user has created a bunch of machine-generated articles and drafts, collecting a litany of various warnings and notices on their talk. Most of it has been draftified or CSD'd already, but would appreciate some help wading through all of it and seeing if anything has been missed or needs to be cleaned up from mainspace. --Gurkubondinn (talk) 12:09, 27 February 2026 (UTC)

This editor has no communication contributions, does not respond on any talk page. Seems to be a WP:NOTHERE case. ~Anachronist (who / me) (talk) 15:25, 27 February 2026 (UTC)

Agreed, but I'm not an admin so I can't do anything about that. --Gurkubondinn (talk) 15:40, 27 February 2026 (UTC)

@Admins willing to patrol AINB: , they appear to be spamming AfC (they caught a block for spam 5ish years ago), seems like a mainspace block at the very least? Kowal2701 (talk, contribs) 23:11, 2 March 2026 (UTC)

A mainspace block wouldn't help with the AfC spam, but edits like Draft talk:Devka Beach (2) are not beating the spamming allegations. Oddly enough, they do seem to know that their talk page exists. Regarding their recent talk page warning, their activity at Manora, Washim is less than constructive but not blatantly vandalism, but something like 2025 India-Pakistan conflict but india wins safely falls in the WP:NOTHERE category. Blocked until a convincing explanation is given. Chaotic Enby (talk · contribs) 08:02, 3 March 2026 (UTC)

Responding to the ping. I endorse the block. Real96kmaratha needs to engage in communication if they wish to continue editing, more so than "This page should not be speedy deleted as unreviewed LLM content, because... (your reason here)" (Special:Permalink/1341447532). — Newslinger talk 08:14, 3 March 2026 (UTC)

Pornography Act (Austria)

Courtesy link: Pornography Act (Austria)

I suspect widespread use of LLM at this article recently approved at WP:Afc. Can someone have a look and let me know what you think? Also, does anyone know if Afc reviewers generally look at the possibility of LLM usage, and if so, is that a factor in disposition of a submitted article? Thanks, Mathglot (talk) 01:19, 28 February 2026 (UTC)

This is a weird one. It looks like the article may have been AI revised and/or translated, but that may have been done in German rather than English. I don't speak German so I'm kind of out of my depth here -- pinging @Wortulo since they know way more about German AI articles.

The user created the original version an hour or so before the same thing on German Wikipedia. Both are mostly in English. It's 2025, so recent enough to be AI, but the text here doesn't read like AI; it seems clearly copied from somewhere but I don't know where.
Then they translate the English to German (for some reason) and add more German text. At least according to Google Translate nothing weird seems to have happened in translation.
This edit revises the German text, and just going by the general shape of the changes, it really, really seems like a German version of the kind of AI copyediting you see in English: summarizing a direct quote or condensing text into the usual AI-isms. For instance I think these are one pair of before/after (note: this is Google Translate):

Old: The SPÖ had a different focus: they were in favor of the law, but considered appropriate sexual education to be a priority. In their opinion, young people should not have to resort to this "dark and dirty literature" if they were properly educated.

New: The SPÖ emphasized the need for sex education to keep young people away from obscene literature.

(the WP:SUPERFICIAL pattern, putting "the need for X" in wikivoice)

The article gets edited, still in German, for a few months, then gets translated to English starting here. Also seems like a straightforward translation.

Gnomingstuff (talk) 02:07, 28 February 2026 (UTC)

(to answer your other question: there's an AfC decline template for AI-generated articles, but a few reviewers spot it way way more often than others.) Gnomingstuff (talk) 02:11, 28 February 2026 (UTC)

Hi, Gnomingstuff, and thanks. Yes, the editor is definitely a German speaker, and imho a user of LLM for articles and Talk pages. I did not want to bias opinions by raising it in advance, but your analysis aligns with (and goes beyond) mine. Earlier talk sections like this one give you a taste of the issue in the past, pre-LLM era. I think they may have moved to LLM/MT partly as a result of that.

Part of my question relates to content and partly to whether to add the {{llm}} banner to the article, which I am inclined to do. If you think this is slam-dunk, feel free, otherwise we can wait for further opinions. (edit conflict) Mathglot (talk) 02:31, 28 February 2026 (UTC)

Thanks for asking me. Your analysis has some facts I also see as possible signs. But Austria sounds sometimes different and Legal Language is extra sophisticated too. The Author User:The Other Karma is not a newbie and I have asked him on his German discussion page. and mentioned this discussion. Wortulo (talk) 07:08, 28 February 2026 (UTC)

Thanks -- I can't comment at all on the accuracy of content in German, and it seems like the bulk of the revising might have happened based on the German text, so it's helpful to have a native speaker take a look. (Not sure that someone being new is indicative one way or the other though, there are longtime editors who now use AI and new editors who don't.)

As far as tagging goes, I think the fact that they've disclosed using DeepL for at lest some things (on that talk page), plus the stuff mentioned above above, is enough to pass the duck test, but up to you. (DeepL didn't use an LLM until mid-2024, which would account for the huge difference in style between Jan 2024 and now.) Gnomingstuff (talk) 07:50, 28 February 2026 (UTC)

The article, got reviewed by me, the Afc Reviewer and the Edit Request Reviewer, me and the Edit Request Reviewer did that rigorously. Like the Edit Request Reviewer already sayed when he reviewed it. The text is properly reworded from its sources, and so close to the source material, that i am still concerned if it might be close paraphrasing. Which you should also focus on and not AI. You can also just ask Wortulo to check, the text, when triple verifications isn’t already enough for you. Or yeet the source into translation software, and check on your own that everyhing is properly reworded from its sources. The Other Karma (talk) 07:57, 28 February 2026 (UTC)

This doesn't answer the core question, which is whether AI was used. Gnomingstuff (talk) 08:10, 28 February 2026 (UTC)

Eh, why does that matter if everything can be verified? The Other Karma (talk) 08:12, 28 February 2026 (UTC)

Interesting question. G15 in English WP requires "stronger" facts, seems more pragmatic and I like it. But you know, that in German WP use of AI is stongly forbidden. This is the situation there. Left in table: the rules, right: the problems I see. We meet us next week in Berlin for a KI-Meeting, to interprete :-) Therefore I do not yet discuss the German article there. Wortulo (talk) 08:24, 28 February 2026 (UTC)

Because it allows editors to know what review may still need to be done, and AI text tends to need much more done than most people think -- checking everything for source-to-text integrity problems, removing synthesis and editorializing, fixing wordiness and promotional tone creeping in, possibly restoring older edits, etc. Plus it just creates an atmosphere of being collaborative and forthcoming with information, rather than adversarial and evasive. Or, in other words, if someone asks you an on-topic yes-or-no question and the answer is "why does it matter," it suggests that you don't want to give your real yes or no answer. Which in turn makes everyone wonder why.

Regarding everything can be verified... was it, though? Using the emphasized the need for sex education to keep young people away from obscene literature example again, there are three potential points of failure here. One, is it accurate to the original text? (the part I can't comment on). Two, did they actually emphasize it (like I did just there), or is it only mentioned in passing? Three, "the need for sex education to keep young people away from obscene literature" is an opinion that many people don't share. (Sex workers, free speech absolutists, and abstinence-only education advocates would probably not think this is a "need," all for different reasons.) The original text correctly (I assume) said this was the SPÖ's opinion, but now "the need" is just stated in wikivoice as a need of the universe. Which is an extremely common AI pattern. Gnomingstuff (talk) 08:53, 28 February 2026 (UTC)

Klaus, yes, thanks. My experience with translating legal terms from French and from Brazilian Portuguese has led me to be very wary of machine translation and LLMs in this field. For one thing, the terms in different legal systems don't always line up (to say the least!) and sometimes you have to know when to translate from L1 to L2, and when to retain the term consistently in L1 in the L2 text and provide an explanation of it the first time when no L2 expression will do. I expect translating legal concepts from German is at least as difficult, if not more so, than from fr or br-pt. (My German is sketchy, and my only exposure to Austrian legal German is from Kelsen's Reine Rechtslehre ('Pure Theory of Law') but that kind of legal philosophical language is different from (and, I expect, easier to follow than) Austrian jurisprudence.)

My preference in situations like this is to find, as much as possible, secondary sources in English; failing that, secondary sources in German. Primary sources (in either language) should be avoided like the plague, because the risk of WP:Original research (de:WP:KTF) is that much greater in a specialized domain like this. So my default attitude is to be *extremely* wary of an article translated from German in a field like this, rather than researched from scratch using English secondary sources (translated, or not) as much as possible.

Which, circling back, was my original motivation for asking the question about LLM use in the article. I'd rather see a quality article at en-wiki 10% of the size of the German one, but researched and written in English, not translated. There will always be time to improve it incrementally; there is no reason to do it all at once. (edit conflict) Mathglot (talk) 09:00, 28 February 2026 (UTC)

So the underlying process of this (not other ones) article creation was, as far as i can remember: Summarised with the help of AI in German, to speedup the process (the original text you provived a link to got deleted in the process) → Reviewed by me (checking for wrongly worded text or inaccuracies) → Reviewed by Ghilt (Wording), and enhanced by me based on feedback from Ghilt → Updated and Enhanced by other editors when the article was on the main page → Translated to English with DeepL to speed up the process → Reviewed by me (checking for wrongly worded text or inaccuracies) → Reviewed by the AFC Reviewer → Updated by me without AI in the German Article, and checked by me → Updated stuff proposed via Edit Request → Reviewed by Edit Request Reviewer → Structure changes by me after unblock, the were pruposed as an Edit-Request.

But why disclose that? Then people just want to delete the article and block you, so no reason to disclose. I would be way more open, when i do then not have to be afraid of blocking or deletion, i dont care if i need to rewritte the article. What is L1 and L2? The article is based on German and English secondary sources? Regarding the example the underlying sentece is Von Seiten der SPÖ hat man einen anderen Schwerpunkt gesetzt. Die SPÖ empfand das Gesetz zwar gut und richtig, wichtiger wäre allerdings eine vernünftige sexuelle Aufklärung, sodass die Jugend es nicht nötig hätte, sich an diese „dunkle und schmutzige Literatur zu wenden.^[1] Tranlated: The SPÖ took a different approach. Although the SPÖ considered the law to be good and right, it felt that sensible sex education was more important, so that young people would not feel the need to turn to this ‘dark and dirty literature’. I have now checked this again and would still aprove the sentence: The SPÖ emphasized the necessity of sex education to keep young people away from obscene literature. as accuratly translated and reworded. I also checked for terms in other systems e.g. for the Term Unzucht there is no equivalent word in english the best approximation is obscenity, and thats the chase for other stuff in the article too. You can even yeet that article into GPTZero or similar, and i will verfiy that the article is human written, which is the chase due to the source author being an human, and my human review. The Other Karma (talk) 10:02, 28 February 2026 (UTC)

As I understand it the German Wikipedia policy may be different, but as of right now, there is no ban on the English Wikipedia using AI for anything but creating new articles and editing/generating images. However, it is very strongly encouraged (WP:LLMDISCLOSE) to disclose it in any related edit summaries, which the edit summaries on this article don't do (and are a bit misleading overall since the one that summarized the content doesn't mention it was summarized, let alone with AI).

Mentioning these things also just tends to go better for people in general. When users get blocked over using AI it's usually because they've been evasive or argumentative about it. Whereas in a lot of cases including some on this noticeboard, if people answer a question about AI with "yes, and here's specifically what I do," people are much more willing to work with them. Gnomingstuff (talk) 19:51, 28 February 2026 (UTC)

Thx, ill take that into consideration. My experience so far was, when the word AI drops, people jump to the conclusion that the text must be made up and non-compliant (even when its not the case like here), it must be deleted and the user indef blocked. So high insentive to be evasive and secretive. And if everything is verifiable, I don't see a problem (this includes accurate wording). The Other Karma (talk) 03:33, 1 March 2026 (UTC)

Since Karma works in two ‘cultures’: In the German-language Wikipedia, articles generated and revised (!) with LLM are prohibited. This was just decided by a two-thirds majority, without having precise recognition criteria and without knowing when articles will actually be deleted. The implementation still needs to be clarified: does a final human review and the absence of AI-typical quality defects (as with your G15) really fall under the ban or is ‘the smell of AI’ (typical wording) sufficient for deletion? Here, he can deal with this differently and more openly than is currently the case in the German-language Wikipedia. But he must expect that what he says here will also be read elsewhere. I have summarised our situation here in a BLOG, which you can translate in your browser if you want to find out about our situation here – admittedly biased. Wortulo (talk) 07:13, 1 March 2026 (UTC)

But why disclose that? Then people just want to delete the article and block you, so no reason to disclose. Being honest with other editors is still the right thing to do even if it means some of your edits are more likely to be reverted. For example I disclose my religious beliefs on my user page, even though doing that doesn't benefit me at all and makes it more likely that some people will revert my edits. I do it because it's a factor that affects how my edits should be examined by others, and I think being transparent about such matters is the best thing for the wiki as a whole. -- LWG ^talk ^(VOPOV) 06:34, 1 March 2026 (UTC)

More information Off-topic: sidebar on "Unzucht [has] no equivalent word in english" — Mathglot (talk) 21:02, 28 February 2026 (UTC) ...

Off-topic: sidebar on "Unzucht [has] no equivalent word in english" — Mathglot (talk) 21:02, 28 February 2026 (UTC)

O/T: going to try to keep this brief, because this isn't the place to start parsing the translations of individual words, but I just couldn't let a claim like "for the Term Unzucht there is no equivalent word in english the best approximation is obscenity" stand. For starters, most words don't have equivalents in other languages, as you and everyone who has translated knows. (Maybe seventy-nine is an exception, but I wouldn't bet on it.) Researching Unzucht, I see that like most words, it has different meanings in different contexts, so the translation would vary accordingly. The common theme is that Unzucht deals with legalistic descriptions of *human sexual behavior*. The English term obscenity, however, is a property of *content*, like a book, magazine, speech, often used in legal descriptions of limits on expression on sexual topics. Legalistic terms in English about criminalized or regulated sexual behavior (like sodomy) or content (like obscenity) often change over time and by legal jurisdiction, and my impression is that that is the case for Unzucht as well. But in any case, translating the latter as obscenity doesn't work in any context, imho. Mathglot (talk) 21:02, 28 February 2026 (UTC)

Since at least four people have already reviewed the article and confirmed that there are no issues, can I assume that this section can be closed? I can resubmit the article to Afc or lett Wortulo read over the article, but i dont see the reason to be more catholic than the pope, peer-review by four people is already more then overkill. The Other Karma (talk) 12:03, 1 March 2026 (UTC)

I pointed out several issues above. Those apply to just one sentence, so may be similar issues with everything else summarized like so. Gnomingstuff (talk) 19:19, 1 March 2026 (UTC)

The problem with LLM is that it is addictive (hard to get used to work without it) and seductive (it makes you feel prone to believe what it says, even when it is wrong). tgeorgescu (talk) 22:02, 2 March 2026 (UTC)

Refs (Act)

[1]
Graupner, Helmut. "Unzucht und Anstößigkeit: Rechtliche Rahmenbedingungen der Pornografie" [Lewdness and Indecency: Legal Framework of Pornography]. Prostitution und Pornografie [Prostitution and Pornography] (PDF) (in German). Austrian Institute for Family Studies. pp. 10–46. Retrieved October 28, 2025.

LLM use by Kristopher9 and associated socks

Rapid large additions by Jesawhite1

Jesawhite1 has been making very large edits to draftspace and mainspace articles in a very short time span, a large majority which are clearly LLM generated. I've draftified a non-notable draft they created, but there's a lot more edits that haven't been reviewed. monkey_smashing^keyboards (talk) 22:03, 2 March 2026 (UTC)

They also have admitted to LLM use. monkey_smashing^keyboards (talk) 22:04, 2 March 2026 (UTC)

the claim: I have personally reviewed, rewritten, and verified all the content for neutrality, citation accuracy, and Wikipedia compliance.

the reality:

His work with producer Kritical Beatz and Fire King Productions (on "Ndokutendai Mukoma") and Btwbc Entertainment reflects a commitment to polished production within the Zimdancehall genre.
Known for her powerful vocals and relatable lyrics, Nisha Ts has become a prominent figure in Zimbabwe's contemporary music scene.
Voltz JT's musical style, termed "jecha trap," combines elements of drill music with the Shona language, creating a unique sound that resonates with Zimbabwean audiences.

and source-to-text integrity issues:

He joined NashTV in 2023 and went on to secure a landmark distribution deal with Virgin Music Group in September 2024, becoming one of the few Zimbabwean artists signed to an international label under the Universal Music Group umbrella. The source does not mention anything about the number of Zimbabwean artists signed anywhere or there only being a few, not even in the quoted press release.
The contract sparked criticism across Iran's political spectrum, with conservative figures viewing it as a liability for President Masoud Pezeshkian's administration. - The source only quotes one person criticizing it for that reason, and does not mention him being conservative.

as well as copyvio:

Fellow artists and fans have labelled him a "desperate attention seeker" and "childish", though some maintain the beefs are marketing gimmicks. - Almost verbatim from the source (and weasel wording and an WP:BLP issue regardless)

that's just a spot check, too Gnomingstuff (talk) 22:34, 2 March 2026 (UTC)

I've messaged them, but they've previously denied LLM-use so a mainspace block seems likely Kowal2701 (talk, contribs) 22:46, 2 March 2026 (UTC)

Mainspace-blocked for six months. ~Anachronist (who / me) (talk) 16:36, 4 March 2026 (UTC)

Upon reviewing their edits and the articles they created, I have been finding repeated examples of content that might be a serious BLP violation sourced to sources that are unreliable like tabloids and gossip blogs. Most of the time, even those unreliable sources don't contain all the claims that were inserted into articles. I am currently combing through all their edits to remove unverified information. -- LWG ^talk ^(VOPOV) 22:28, 4 March 2026 (UTC)

I see that MSK has draftified the articles this user created. I have gone through all edits this user made to articles they did not create and have removed content that failed verification (which was almost all of it). This user's problem content should now all be out of mainspace, although in may be worth considering deleting their drafts, since drafts that remain publicly visible may still end up poisoning search results/chatbot training data. -- LWG ^talk ^(VOPOV) 22:54, 4 March 2026 (UTC)

Draftspace isn't indexed by search engines, see Wikipedia:Controlling search engine indexing. monkey_smashing^keyboards (talk) 23:09, 4 March 2026 (UTC)

AI "translation" that seems to actually be generated text, at Heinrich Zille

I tagged the Heinrich Zille article as containing AI-generated text, for reasons that will be clear if you read it. The tag was silently removed with the rationale "Tag on the talk page indicates article translated from the corresponding German wiki article."

So judge for yourself: These are the edits (Which claim to be translation and unspecified "added context"); this is the German article it is supposedly translated from, at the time of the edits. It seems as if there is nothing in that article that translates to such slop as:

Possessing an extraordinary talent for depicting the harsh realities of urban life with caustic humour and profound humanity. His works illustrated the struggles of society's most marginalised, including disabled beggars, tuberculosis-afflicted prostitutes, and poorly paid labourers, as well as their children. Zille’s art highlighted their resilience and unyielding determination to find moments of joy and dignity amidst hardship. By blending satire with compassion, he brought attention to the grim living conditions of Berlin’s working classes, particularly those residing in overcrowded tenements, offering a poignant critique of the social challenges of his era. (supposedly from the "Secession und der Erfolg als Künstler" section)
Zille documented these realities with a direct, unsentimental approach, portraying them as lived experiences rather than romanticised narratives. His work illuminated the social conditions of the time, highlighting the struggles of Berlin's urban poor. (supposedly from the "Zille und Kinder" section)
What began as an initially euphoric portrayal shifted to a critical and sarcastic humour, reflecting his growing disdain for the imperialist system. The illustrations unflinchingly addressed the suffering and societal transformations brought about by the First World War, capturing the profound upheavals of the time. (supposedly from the "Militärdienst" section)

Gnomingstuff (talk) 22:14, 2 March 2026 (UTC)

It is blatantly obvious that this content is not 'translated' from the German article, if for no other reason than the fact that the English section is over twice as long as the German section. I've reverted the article to before the massive expansion by Nightsturm, which I'm sure is going to result in a re-reversion. Athanelar (talk) 18:15, 3 March 2026 (UTC)

messaged Kowal2701 (talk, contribs) 11:46, 4 March 2026 (UTC)

Nightsturm (talk · contribs), some of their copy edits have AI signs as well Kowal2701 (talk, contribs) 13:03, 4 March 2026 (UTC)

Reporting User:Hazim BS for possible use of AI

Blake Tillery

Blake Tillery (edit | talk | history | protect | delete | links | watch | logs | views)
MtnKrakatoa (talk · contribs · deleted contribs · logs · filter log · block user · block log)

New user admits to their large addition to this (already very badly sourced and formatted) article being generated by Claude, but they are very unhappy that me and others are reverting them. The origin of the article has gone from "generated by Claude" to "just used AI to improve the English" – the normal pattern when people are caught using AI. I'm not expecting my latest revert to stick, so I'd like further eyes on things and for people to prepare to clean up. • a frantic turtle 🐢 18:13, 5 March 2026 (UTC)

Not a new user - have been using Wikipedia for years. I used AI to help write copy, never said it was fully written by Claude. I personally verified all sources. Article was very biased, and is now non-biased, and properly cited. Went from 3 sources to 27. No hallucinations or slop. MtnKrakatoa (talk) 18:16, 5 March 2026 (UTC)

Ate some lunch. Apologis for being short. Thanks for flagging this. I agree the way I introduced the expansion was not the right workflow.

Clarification: I used an LLM as a language/copyediting aid, but the intent was that each factual claim be backed by inline citations. That said, I understand that AI-assisted additions often correlate with sourcing/weight/style problems and deserve scrutiny.

If anyone here can point out specific factual inaccuracies, citation mismatches, neutrality issues, or BLP/due weight problems in the article, I will fix or remove them. MtnKrakatoa (talk) 18:26, 5 March 2026 (UTC)

I did some cleanup of AI slop that remained. I agree the prior version was abysmal. ~Anachronist (who / me) (talk) 20:01, 5 March 2026 (UTC)

Thanks for the transparency, first of all. I think a frantic turtle was overly aggro here especially since you did mention using AI right from the outset.

Going on to the article, the main immediate issue seems to be source-to-text integrity. I assume most of the claims here should be easily sourceable since they are basic biographical details, but as it stands, the sources chosen don't seem to verify much.

Tillery was born and raised in Vidalia, Georgia, in Toombs County. His father, Mike Tillery, grew up in Rome, Georgia, and his mother, Vicki Tillery, is from Cedartown. The second sentence is verified by the source (paywalled but you can get the text off the Network tab), but the first sentence doesn't seem to be; you could argue the citation only applies to the second half.
graduated from Vidalia High School in 2002 and enrolled at the University of Georgia, earning a Bachelor of Arts in International Affairs in 2006. He participated in the L.E.A.D. Athens leadership development program during his undergraduate years. - Probably sourceable, but very little of this is mentioned in the source given. For intsance, Vidalia High School isn't mentioned here or an archived version. It doesn't seem like the website mentions Vidalia High School anywhere at all (I'm guessing that whatever shows up in here is... uh... not part of the website itself) L.E.A.D. Athens is also not mentioned there or seemingly the website as a whole.
He returned to UGA for law school, where he served in the Student Bar Association, and earned his Juris Doctor in 2010. He was admitted to the State Bar of Georgia on November 4, 2010. - The source only mentions that he attended law school at UGA and was in the Student Bar Association.

Gnomingstuff (talk) 20:07, 5 March 2026 (UTC)

Telecommunications in Oman

User:Markus_WikiEditor

Another user pumping out a large amount of AI-generated articles, mostly about music. They have denied using AI but I don't really think it's arguable; pretty much every article has a lot of the usual AISIGNS all over the place, as well as many of their edit summaries (sometimes including Markdown in the summary) but the smoking gun is this revision of Flora the Red Menace (cast recording), which contains what appears to be a chatbot response in ~~Spanish~~ Portuguese: Aqui estão duas versões mais fluídas que incluem a informação de que o relançamento abrangeu outros cast albums:. Gnomingstuff (talk) 06:17, 8 March 2026 (UTC)

@Gnomingstuff Sorry. I did use AI, mainly to translate content. I have rewritten and corrected the errors you pointed out (they are usually not direct, just general), and I hope they are correct now. Thank you for pointing out the issues. Markus WikiEditor (talk) 06:34, 8 March 2026 (UTC)

You should familiarise yourself with our guideline on using AI for translations. Athanelar (talk) 18:24, 8 March 2026 (UTC)

Not to defend the use of AI but I do think that this guideline being stricter than the “you can help translate this article” banner is likely to lead to frustration.

For reference, the banner includes the phrase “If possible, verify the text with references provided in the foreign-language article”, which implies checking the references is optional. Also, if you encounter it while reading Wikipedia, there is no indication you need to speak the language you are translating from. ExtantRotations (talk) 23:46, 9 March 2026 (UTC)

@Gnomingstuff Regarding this Brigitte Bardot article (B.B. (album)), I noticed that all the edits I made were reverted because you said they had been AI-generated. At the time, I understood the notices to mean that the entire article had been created by AI, including the references I added, which is why I disputed it. I corrected the text shortly after your notice. Could you please take a look and let me know if there are still any AI-related issues? Markus WikiEditor (talk) 07:25, 8 March 2026 (UTC)

Hammer retarder and use of AI generated image

This page is using an AI generated image by User:Викидим. The image seems more or less unproblematic but is an AI generated image in an article about a non-AI subject. I just wanted to bring this to people’s attention as a break from standard practice in this area. Dronebogus (talk) 17:53, 10 March 2026 (UTC)

If the image is unproblematic, then what's the issue? If you think a better image exists then feel free to replace it (and follow BRD if anyone disagrees) as you would any other image. Thryduulf (talk) 18:52, 10 March 2026 (UTC)

I agree with Thryduulf that if you personally don't think the image is a problem, the WP:ONEQUESTION solution is to just not worry about it. There's plenty of actual bad AI content out there to focus on cleaning up. If someone genuinely objects to the image, then per WP:AIIMAGES the onus would be on Викидим to make a strong case for inclusion. -- LWG ^talk ^(VOPOV) 19:02, 10 March 2026 (UTC)

I see no problem here. First of all, if there are any photos or other professional-grade images available, any amateurish work (I am very unaccomplished as an artist) should be thrown away, naturally (I have explicitly requested such replacement on the Talk:Hammer retarder from the get-go).

If I draw a sketch on my phone, I would assume that there should be no issue with an illustration created by me, for the time being (which can be years, just look at the Panty line). It really should not be different if instead of an electronic pen I use AI which is clearly a more gifted graphic designer than yours truly, Викидим (talk) 19:55, 10 March 2026 (UTC)

Look if you must, but don't stare. Thebiguglyalien (talk) 20:53, 10 March 2026 (UTC)

Regardless of your personal opinion, it is Wikipedia policy that AI-generated images should almost never be used in mainspace, and specifically that community members have largely rejected making exceptions merely because an image lacks obvious errors, or because no free non-AI-generated images are available. Following policy is not optional. Gnomingstuff (talk) 23:14, 10 March 2026 (UTC)

Note phrases like "Almost never" and "Generally" and that one should never follow policy blindly. If you think an image (any image) is problematic for some reason (whatever that reason is) you should either replace it or remove it (in that preference order), explaining why you think the image is problematic. If someone objects or reverts you, discuss it. If you can't come to an agreement then follow the usual dispute resolution process.

If you don't think an image is problematic, or not problematic enough to take any action over, then just move on, there almost certainly isn't a problem that needs solving. If someone else thinks the image is problematic enough to take action over then they will take action over it. Thryduulf (talk) 00:26, 11 March 2026 (UTC)

The "almost" refers to things like articles about AI and articles about notable AI-generated images. Gnomingstuff (talk) 00:32, 11 March 2026 (UTC)

If you're going to try and wikilawyer about the exact text of a policy it helps if you do that based on what the policy actually says: Most images wholly generated by AI should not be used in mainspace, especially for named people and in technical or scientific subjects such as anatomy and chemistry. This rule should be applied "subject to common sense and with a number of exceptions". Obvious exceptions include articles about AI and articles about notable AI-generated images; other categories of exceptions may arise through further community discussion. Community members have largely rejected making exceptions merely because an image lacks obvious errors, or because no free non-AI-generated images are available. (emphasis mine)

It's explicitly giving examples of categories of exceptions, not prohibiting individual exceptions, which it implicitly allows.

I also invite you to respond to the part of my comment that you conspicuously ignored. Thryduulf (talk) 01:52, 11 March 2026 (UTC)

Stating what a policy says is not wikilawyering. Wikilawyering is trying to mangle the words beyond the spirit of the policy to argue that someone using an AI-generated picture of a gun component because they don't feel like creating a non-AI illustration is such an unusual scenario as to justify an exception. Gnomingstuff (talk) 16:57, 11 March 2026 (UTC)

Once again I'm going to have to ask you to read what I actually wrote rather than what you think I wrote. I have not expressed an opinion on whether this specific picture should or should not be exempt, just that there is no policy prohibiting an exemption. The spirit of the policy not only matches my words, but it makes that absolutely explicit by the plain and simple wording of the policy itself: This rule should be applied "subject to common sense and with a number of exceptions".

It is also worth nothing that this section was opened by an editor who explicitly thought the image was unproblematic. That other editors have subsequently disagreed with that changes nothing about what the correct course of action is for those who find any image problematic and those who find any image unproblematic (note the AIIMAGES section does not contain any special provision for enforcement, so the course of action or inaction is identical to that for any other image on the project).

Finally I also note that I have not commented on any editor's motivations for creating, using or not using an image and I do not endorse your characterisations. Thryduulf (talk) 17:22, 11 March 2026 (UTC)

Such literal interpretation was obviously never intended, as it will preclude taking pretty using any new photos.

Whenever I press the camera button on my Samsung phone, a pretty complicated combination of software and hardware kicks in. It is called ProVisual Engine and ... drum roll ... it is AI. Any other high-end phone maker has similar technology, and bypassing it for photos is nigh-impossible. We should assume that drafters of WP:AIIMAGES were familiar with the current state of technology, and yet they did not mention it on the list of exceptions, so the list is indeed not closed.
If AI, per #1, can be used to enhance the photos that I take, what is the problem with it enhancing my free-hand drawing doodles? For an already-cited example, see File:Panty line on a croquis of a torso.jpg. There are two versions there. I drew the original by hand (no, not on paper, but on a screen of my phone). I hope that no-one would disagree with me avoiding paper. Triggered by today's discussions, I have asked AI to improve the sketch, the result was the new version of the same image. It looks much nicer, but definitely does not show any creativity - it is still my drawing, simply drawn with a hand that does not shake. This IMHO must be OK, too, it is clearly a derivative of a free "work of art" (I am no Leonardo).

If one agrees with #1 and #2 being OK, the retarder picture appears (to me) to be the same, just a bit more realistic (with the panty line I tried to steer as far away from naturalism as possible), and thus should be OK, too. Викидим (talk) 04:13, 11 March 2026 (UTC)

Great news! The WP:AIIMAGE policy already covers image modifications as well. I suggest that you read it. Gnomingstuff (talk) 16:59, 11 March 2026 (UTC)

WP:AIIMAGE begins by defining the scope as images wholly generated by AI, the only mention of modifications is: Marginal cases (such as major AI enhancement [...]) are subject to case-by-case consensus. I'm not sure how that relates to the context here? Thryduulf (talk) 17:27, 11 March 2026 (UTC)

Good. I found my notes on the image creation for the Glock trigger bar and am going to recreate them with depositing intermediate results to Commons as well for everyone to see. Викидим (talk) 18:35, 11 March 2026 (UTC)

Recreating the image in the old way would have been tedious (doodling on the whiteboard is not my area of strength). Instead, I took a shortcut by looking at FOARP's work below and using the image from the 1998 Glock switch patent that has no copyright notices (should have done it before, but was not thinking quick enough). I also decided not to do the pencil sketch and instead used default ("lifelike") coloring scheme. The results, with all the prompts shown, can be found at File:Glock-like trigger bar 3.jpg. Now, in order to move on:

If anyone wants to see how my original method works, let me know, and I will recreate my doodling (not today, I need some sleep). Result will be better that what I produced in "bar 3", as, to the best of my recollection, the shape of linkage in the actual Glock is closer to the challenged picture than to the one from the patent (there should be no pronounced bend on the long side). I think that the ability of the AI to color a 2D line sketch are now clearly illustrated.
If someone thinks that AI used some copyrighted material while producing the "bar 3" result, let's discuss it.
I would like to place the "bar 3" result into the Trigger bar#Glock section to illustrate the "cruciform" part of the design now outlawed in the US (this is why I have created the article in the first place: it does not happen often for a particular shape of a stamped part to be declared illegal).

Викидим (talk) 20:16, 11 March 2026 (UTC)

You should probably update the file description on Commons to specify it is AI-generated.

Also, if you zoom in on the left of the image, you will notice abnormalities (e.g. a break in the edge of the metal) that show the image is not real. SuperPianoMan9167 (talk) 21:15, 11 March 2026 (UTC)

Gemini also appears to have flattened the indentation that runs through the center of the horizontal bar, replacing it with two parallel grooves that have no basis in the actual design. Do these differences matter? Hard to say without deep subject-matter expertise. This is why WP:AIIMAGES includes Community members have largely rejected making exceptions merely because an image lacks obvious errors. If this discussion hadn't blown up this big, I personally would never have noticed those defects and would have allowed the image to stand. -- LWG ^talk ^(VOPOV) 21:50, 11 March 2026 (UTC)

Updated the image. Викидим (talk) 23:54, 11 March 2026 (UTC)

There's still a break in the edge on the left side at the same exact spot. SuperPianoMan9167 (talk) 02:28, 12 March 2026 (UTC)

Updated the image. At the risk of being a little blunt, the issue here isn't that there were errors in the image, the issue here is that you didn't notice the errors until we pointed them out to you, which means that if we hadn't spent a lot of time scrutinizing your AI images, those errors would now be live on Wikipedia. We don't have time to scrutinize every AI image that people want to include, so unless you can find a workflow where you are consistently catching these errors yourself, we'd rather you didn't insert images like this into articles. -- LWG ^talk ^(VOPOV) 04:26, 12 March 2026 (UTC)

We don't have time to scrutinize every AI image that people want to include This is where I have stopped to understand you, as exactly the same logic IMHO applies to any graphic images generated by editors. Let's suppose that, instead of using AI, I would have generated the same image using some 3D graphic package. Will this hypothetical image be OK despite having imperfections worse than this AI-generated one? This makes little sense to me, as Wikipedia is full of images that have discontinuities on a much larger scale (random example is on the right). For the avoidance of doubt, I think that this particular example image is extremely useful and way better in that sense than the photo in the same article (Ballute). Викидим (talk) 09:47, 12 March 2026 (UTC)

If you can get an image of the switch that is not AI-gen (and you can) then just don't use the AI gen image. Doing otherwise just looks WP:POINTy. FOARP (talk) 13:49, 12 March 2026 (UTC)

Will this hypothetical image be OK? I feel like you are applying this principle backwards: in my view, the factual errors in the trigger bar image made it unsuitable for inclusion in an encyclopedia, regardless of the means by which the image was made. AI enters into the discussion only because it appears to me that AI images are the class of images people have the hardest time assessing for factual errors. To be clear, if I saw that you had a consistent pattern of producing and inserting hand-drawn images with similar errors, I would also request that you stop. -- LWG ^talk ^(VOPOV) 16:20, 12 March 2026 (UTC)

The AI seems to have created a commons:Commons:Derivative works of this copyrighted image down to the angle of the 'photograph', while adjusting the design of the hinged element. CMD (talk) 03:03, 11 March 2026 (UTC)

There is a similar situation at Trigger bar. While we're discussing possible concerns with those images, I'd like to also commend Викидим for inviting scrutiny and clearly labeling his AI-generated images in the captions, which is super helpful and reduces the harm if it the images do turn out to have problems. -- LWG ^talk ^(VOPOV) 03:37, 11 March 2026 (UTC)

The two images of the same object would naturally have similar appearance. It does not mean that one of them is a derivative of the other. You can easily see where my hand went off and the shape was mangled (thus the "AKM-like") and AI did not copy the details absent in my sketch ("19" stamp). You can also see where AI got creative and added a screw head (the retarder does not use a screw there). I am quite willing to recreate the process from the free-hand sketch yet again and show the process. It will not be a wasted effort, as I can draw the "fire assembly" with the trigger that will be more educational (I always wanted to do it, but never had an excuse). The other thing I wanted to try was to try building 3D model instead of a doodle. BTW, I fail to understand at what point an elaboration of my own work depicting a physical object can become a derivative of some other image. Викидим (talk) 04:29, 11 March 2026 (UTC)

@Викидим, have you read this disclaimer on AI-generated images on Commons?

Most image-generating AI models were trained using works that are protected by copyright. In some cases, such models can output content with major copyrightable image elements which are identical to or derivative of the original training data, making these outputs derivative works.

SuperPianoMan9167 (talk) 16:55, 11 March 2026 (UTC)

Yes, I have. I don't see any such elements in a particular picture, that's why I have by now re-done it from scratch without ever telling AI what is it working with. See details elsewhere on this thread. Викидим (talk) 19:48, 11 March 2026 (UTC)

Then, again, there are dots on the surface that I did not make in my sketch, so AI could have been inspired by other images, as my final prompts were not tight at all. Will try again with tighter prompt "using only the image provided, ..." Викидим (talk) 06:12, 11 March 2026 (UTC)

Just to note here that free-to-use images of hammer-retarders exist. For example, the images of expired US patent no. 2,459,158 are public domain (they are images in an expired US patent to which no copyright notice is attached). There is no need to use AI-generated images of something that may or may not actually be a hammer-retarder.

More generally, a hammer-retarder is just a kind of rate-reducer, a common feature of all weapons capable of fully-automatic fire. This article should be renamed to Rate reducer (the more inclusive title) and example other than the AK platform included. For example the M-16's rate reducer was called the "buffer" and can be seen on p. 3-23 of this US Department of Defence-published (and thus public domain) manual. FOARP (talk) 12:01, 11 March 2026 (UTC)

UPDATE: I believe the same is true for Trigger bar - this US DoD-produced manual shows the disassembly of a M1911 .45 pistol, I believe page 3-13 shows the trigger bar. I don't think there's ever a good reason to use AI-gen images in this field, since these are all known and technically-described/drawn items, so there is no need to use AI to generate something that may ( or may not be) the article actually being discussed. FOARP (talk) 12:58, 11 March 2026 (UTC)

Note: both images have now been replaced by free, non-AI-generated images that were discovered as a result of this discussion. -- LWG ^talk ^(VOPOV) 18:32, 11 March 2026 (UTC)

Propaganda in Augustan Rome

Signpost Article

Just wanted to draw attention to this recent Signpost article about AI possibly inserting citations to medical papers in inappropriate articles. For example, in this edit to the article on the Great Leap Forward, a user added a citation to a paper about methanol poisoning. And in this edit to the article on Arctic char, another user added a paper about lorazepam to the sentence, “The first migration of Arctic char has been found to occur between 4 and 13 years of age.” BusterTheMighty (talk) 01:10, 12 March 2026 (UTC)

Obviously this is a problem, but is it a different problem than the already-known issue of editors (not all of whom used an LLM) inserting a random source so the content has a source? I.e. is this a new problem that we need to look out for and figure out the appropriate resolution to (which may or may not be something we're already doing) or is it just a specific example of an something we're already dealing with? If the latter, it may be useful as a way of finding instances of this specific problem (how useful will depend on what the signal to noise ratio is, something the article doesn't mention).

We know that LLMs do this because they don't understand why Wikipedia articles have sources, they just know they do and this is almost certainly true of some humans too. I would guess that editors intending to deceive would be more likely to use a source that would at least plausibly verify the content from a superficial look, so using a pubmed source for a non-medical article feels less likely to be bad faith than either a lack of understanding or of simply not checking the sources the LLM found (which may of course be due to that same lack of knowledge). It is also possible (although unlikely) that a medical paper does actually verify some non-medical claim (I can't remember any examples but I know I've cited things to academic papers that on the surface had little relation to the article subject). Thryduulf (talk) 02:24, 12 March 2026 (UTC)

Uhhhh.....

The first example, "a study on rat livers appears in List of Disney television films," was the case in 2022, so it is almost definitely not AI's fault.
Same for the second example.
And the third.

I stopped here because... dude, you can't even do the most basic diligence possible, like what are you even doing here? (edit to clarify: talking about the author of the Signpost article here, not the thread)

I scraped the titles of PMIDs 1–1000, had some code search Wikipedia for articles containing those titles, and then asked Claude to filter out any articles where the PMID could somewhat plausibly be relevant. This is not a sophisticated approach. It is embarrassingly simple.

Oh. Gnomingstuff (talk) 03:01, 12 March 2026 (UTC)

i've heard of folks inserting their own articles into wikipedia articles supposedly to inflate some metric. (I'm not sure which metric and pretty sure no scholarly field cares if an article is used inside wikipedia, but I think MDPI tracks the citations of articles inside wikipedia?) User:Bluethricecreamman (Talk·Contribs) 03:13, 12 March 2026 (UTC)

According to the article comments this is apparently a VisualEditor bug. Gnomingstuff (talk) 03:15, 12 March 2026 (UTC)

oh my. nvm then. User:Bluethricecreamman (Talk·Contribs) 03:16, 12 March 2026 (UTC)

Haziran11

User:Bronchalinox and Pangeanic

User:Pehlivanmeydani

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

This user has very rapidly authored a lot of BLPs in mainspace, so fast that they cannot possibly have written them themselves. I came across this user because one of those BLPs had an OAICITE (so I nominated it for G15). Can someone else take a look as well? --Gurkubondinn (talk) 13:06, 13 March 2026 (UTC)

seems like it may be GPT-5; not the most egregious case, just some mild promotional tone and possibly WP:NCORP issues Gnomingstuff (talk) 13:54, 13 March 2026 (UTC)

And potential source-to-text issues, I haven't invested the time to do source analyses. --Gurkubondinn (talk) 14:19, 13 March 2026 (UTC)

I would like to respond to this concern.

I have been creating wrestling-related articles on Wikipedia for years, especially biographies of wrestlers and pages about wrestling tournaments. This is not a new topic area for me. Over time, I have also created or expanded many pages about recent wrestling competitions, including World Championships, European Championships, Mediterranean Games events, and similar tournaments.

Because I have worked in this area for a long time, I am already familiar with many of the wrestlers and competitions I write about. The fact that I can create these pages quickly does not mean they are unreviewed. It reflects that I already know the subject, the standard structure of wrestling articles, and the source base commonly used for this topic.

If there are specific problems in any article, such as malformed citations, source-to-text issues, or wording that should be improved, I am willing to fix them. One article containing an OAICITE-related formatting problem should not be treated as proof that all of my work is machine-generated or unchecked.

I have created many wrestling biographies and tournament pages over the years, including pages related to recent World and European Championships, and those pages have generally not had this kind of issue. If there are concerns about particular articles, I would prefer that editors point to specific examples so they can be reviewed and improved individually. Pehlivanmeydani (talk) 00:28, 14 March 2026 (UTC)

One article containing an OAICITE-related formatting problem should not be treated as proof that all of my work is machine-generated or unchecked. But it is proof that you are inserting unreviewed AI-generated content at least some of the time. Also, this kind of wording; "If there are any specific problems, tell me; don't talk about my AI usage, though..." is very typical of AI-generated responses to these kinds of accusations.

Put simply: are you or are you not using AI assistance when writing these articles? Athanelar (talk) 01:44, 14 March 2026 (UTC)

Refs (Act)

Related Articles