Wikipedia:WikiProject AI Cleanup/Noticeboard

Archives

Archive 1

This page has archives. Sections older than 90 days may be auto-archived by Lowercase sigmabot III if there are more than 4.

Cleanup efforts are ongoing and being tracked at /2025-09-17 Thefallguy2025, 206 pages need review.

I am in the process of tagging the edits by Thefallguy2025 as AI generated and I am not even close to being done. Similar as the above: a lot of rewrites from the newcomer edits copyedit/expansion interface, a few smoking guns (ChatGPT parameters left in on some of the ones earlier in the year), extremely AI-esque edit summaries, and enough suspect text elsewhere that I'm pretty sure it's all AI. To the tune of over 1,000 edits. Gnomingstuff (talk) 21:24, 17 September 2025 (UTC)[reply]

Thanks for finding these and please keep it up! We need more LLM policies and being able to demonstrate the magnitude of the problem to the wider community will help move in that direction NicheSports (talk) 03:13, 18 September 2025 (UTC)[reply]

Thanks -- trying not to spam and only mention the really prolific ones, I feel like I'm already taking up a lot of talk page space here. Gnomingstuff (talk) 04:28, 18 September 2025 (UTC)[reply]

Do not worry at all, appreciate all of your work!! ʊnƌer◙swamȹᵗᵅᵜᵏ 13:10, 18 September 2025 (UTC)[reply]

imo there should be an AI version of WP:CCI. Makes it easier to identify who needs to be cleaned up, their edits, and the progress. Hopefully it doesn't become as backlogged as CCI though. ARandomName123 (talk)^{Ping me!} 04:58, 19 September 2025 (UTC)[reply]

That would be helpful but I'm worried about the inevitable "you have no proof why are you putting me on a list" backlash, and I also don't want this to turn into scapegoating or insulting editors. Gnomingstuff (talk) 06:35, 19 September 2025 (UTC)[reply]

According to Wikipedia:Contributor copyright investigations/Instructions, a newly filed CCI has to be evaluated by a clerk or administrator, who would subsequently open a case only if one is warranted. We could implement a similar triage system for LLM cleanup cases, although I think a consensus on the LLM noticeboard (this page) would be sufficient to open an LLM cleanup case (without needing an administrator or designated clerk). — Newslinger talk 15:51, 19 September 2025 (UTC)[reply]

This makes sense to me. It would help with the cleanup process and make it clearer to the wider community how this type of LLM misuse should be handled, likely leading to cases being caught earlier NicheSports (talk) 13:29, 20 September 2025 (UTC)[reply]

No worries, Gnoming! You're on this talk page a lot because you do a lot of good work :) Altoids0 (talk) 22:52, 19 September 2025 (UTC)[reply]

Had a chance to look into this and it is pretty bad too. These four substantive article expansions with obvious LLM-style editorialization were made within 45 minutes: [1], [2], [3], [4]. This article expansion [5] is particularly absurd, and likely involves extensive hallucinations as it is entirely unsourced. This is just a small sample NicheSports (talk) 15:21, 19 September 2025 (UTC)[reply]

yeah, I checked all of their hundreds of edits to make sure at least some of the original text was still in the current version, and I think I only found a handful where either the text had been substantially edited, or where someone had noticed and reverted. Gnomingstuff (talk) 17:54, 19 September 2025 (UTC)[reply]

Ugh.. lots and lots of 'Plot' sections too, which are a nightmare because we rely on the source material for the plots so, short of going to watch the film, independent editors have no way of verifying the content.. I'm assuming that ChatGPT hasn't actually watched the films so must be pulling the summary from somewhere else, or hallucinating it. JeffUK 02:09, 24 September 2025 (UTC)[reply]

This is the problem with Plot sections independent from AI, IMHO. I know that the community thinks that WP:OR is OK when dealing with movies, maybe this will force change in this approach? Викидим (talk) 07:04, 18 November 2025 (UTC)[reply]

Thank you! It would be great if - in mass cases - edit summaries (or hatnotes) identify the diff or at least the offending editor, so that their contribution would be easier to delete. Викидим (talk) 07:01, 18 November 2025 (UTC)[reply]

I've been experimenting with various ways of indicating the diff/editor. The problem is that no matter what I do, people decide I should be doing it some other way, or not doing it at all. Gnomingstuff (talk) 07:14, 19 November 2025 (UTC)[reply]

You are doing an enormous amount of great work, so if it is easier for you to do it in some way, I will not be the one to grumble. Thank you! Викидим (talk) 22:54, 19 November 2025 (UTC)[reply]

• Tracking subpage created, also listed and tracked at the top of this report. Triaged ​all 608 articles, 206 appear to warrant closer review or cleanup.​​​​​ fifteen thousand two hundred twenty four (talk) 14:03, 18 November 2025 (UTC)[reply]

  whoa... did you automate this in any way? If so lmk what tools you are using NicheSports (talk) 14:07, 18 November 2025 (UTC)[reply]

  The list of edited articles was collected and formatted semi-automatically, then the history of each was given a cursory review to filter out low quality cleanup candidates. I'm going to try a slightly different process for article selection for the Bookleo report, will let you know the specifics of how that goes. fifteen thousand two hundred twenty four (talk) 08:09, 19 November 2025 (UTC)[reply]

  Done, contributions were retrieved via CCI's contribution surveyor (uncheck minor edits, set a large negative bytes value like -999999), then formatted using regex and previewed in the editor. Diffs are quickly checked using navigation popups, wikiEdDiff is used for more in-depth checks, Who Wrote That? is used to investigate partial cleanups, and three scripts are used to make diff links more accessible. A final regex pass formats diffs added to notes and marks entries with them as completed.

  Can be done faster if the optional notes and completed entries are left out.​ fifteen thousand two hundred twenty four (talk) 05:10, 20 November 2025 (UTC)[reply]

  Nice, thank you.

  Would it be helpful for me to go through the past few months of tags to identify clusters that could be broken off into subpages (either by me or by somebody else), or would that just create more work? Right now I've only been tagging so as not to spam this page, unless something was really egregious, but absolutely there are cases where it's been dozens/hundreds of articles. (for instance pretty much all Grey's Anatomy articles currently) Gnomingstuff (talk) 14:10, 18 November 2025 (UTC)[reply]

  If you mean clusters of edits by this single user, those would be best kept to the current subpage and could use separate sections each with their own {{AIC article list}}. The formatting isn't strict at all.

  If you mean topical clusters of pages separate from any one user, then making a new report with its own {{AIC status}} and subpage would work. My preference is for reports to concern a single editor whenever reasonable, but if you think making a report about a specific topic could better engage others in a cleanup effort (maybe by inviting relevant WikiProjects?), then give it a try. fifteen thousand two hundred twenty four (talk) 08:09, 19 November 2025 (UTC)[reply]

  Sorry to just now see this. This is about editors who make dozens/hundreds of edits, right now I have been tagging those but not collating them anywhere. Gnomingstuff (talk) 21:08, 26 November 2025 (UTC)[reply]

  Creating a report and subpage about a single user and using multiple sections on the subpage for different topics that user has edited would be best, unless I've misunderstood. fifteen thousand two hundred twenty four (talk) 21:16, 26 November 2025 (UTC)[reply]

Cleanup efforts are ongoing and being tracked at /2025-10-30 Tbound2, 4 pages need review.

I stumbled across an article created by Tbound2 that was nominated for deletion (Adelaide Metro bus route 100), and have discovered a large number of articles created by this user, in implausibly short succession (while obviously this does not inherently signify generative AI editing, all of these were created within 2 hours of each other), and which I believe to be generated by AI large language models due to their gushing prose (and which already have AI tags):

• India Rasheed, created at 10:54 on 10 August ("Rasheed is noted for her clean hands, decision-making, and effective left-foot kicking. She is considered capable of impacting games both at ground level and in aerial contests. Coaches have also noted her endurance and adaptability to different positions.")
• Keeley Kustermann, created at 11:19 on 10 August ("She was known for her “clean by hand” play, strong decision-making, and versatility.")
• Georgia McKee, created at 11:22 on 10 August ("McKee is recognised for her agility, goal sense, and forward craft. She applies strong defensive pressure inside 50 and is regarded as a creative and opportunistic small forward.")
• Brooke Smith (footballer), created at 11:26 on 10 August ("Smith is described as a reliable utility with strong marking ability for her size, a penetrating kick, and the versatility to adapt to multiple positions.")
• Lily Tarlinton, created at 11:32 on 10 August ("Tarlinton is known for her athleticism, marking ability, and adaptability to both forward and ruck roles. Her height and mobility offer Adelaide flexibility in matchups, enabling them to rotate key position players across the ground as needed.")
• Keeley Skepper, created at 11:37 on 10 August ("A midfielder noted for her contested ball-winning ability, composure under pressure, and precise kicking...")
• Poppy Scholz, created at 11:47 on 10 August ("A tall utility renowned for her intercept marking, versatility, and athleticism...")
• The Original Pancake Kitchen, created at 12:14 on 10 August ("aims to deliver the brand’s signature all-day breakfast, diner-style charm, and thick milkshakes in a modern, family-friendly setting.")

There are a number of additional articles created by this user that likely fall under the same category of speedily-generated articles entirely written by a generative AI LLM. I don't really know what to do here. I don't think they're eligible for speedy deletion under G15, given these articles obviously could have plausibly been created a different way, they're not transparently nonsense and they've all been reviewed. But I believe this editing to be highly problematic and I'm not sure what to do about it. What is supposed to be done when an editor is repeatedly generating articles using large language models? LivelyRatification (talk) 00:43, 30 September 2025 (UTC)[reply]

Cases like this are where I've generally been tagging articles. It looks like I've tagged some of this user's contributions but not all, I assume that here, as in similar long-term LLM users, I stuck to the most obvious cases even though it's highly unlikely they stopped using AI. (Because then I would have to deal with "BUT YOU HAVE NO PROOOOOOOF POINT OUT WHAT IS WRONG WITH THIS TEXT" type comments which I do not want to do.) Gnomingstuff (talk) 02:01, 30 September 2025 (UTC)[reply]

@Gnomingstuff: I suppose my question is, what good is tagging articles if the user in question is continually using generative AI to create articles? I could, and indeed might if I have the time, individually fix every article to remove problematic LLM insertions (as was done on Lauren Young (footballer)), but is it not a problem of disruptive editing if a user is repeatedly using generative AI to create articles? LivelyRatification (talk) 02:11, 30 September 2025 (UTC)[reply]

The thing is we have no policy against LLM use currently, so within our existing system, they're not actually doing anything that isn't allowed. If someone isn't responding to attempts to discuss the situation then ANI threads have been opened, but there feels like there's a mismatch in what we enforce and what our policy actually is(n't). Gnomingstuff (talk) 02:18, 30 September 2025 (UTC)[reply]

Hm. That's frustrating, but thank you for your help. I guess I'll just keep an eye out and try and fix what's there. I've also left a message on this user's talk page in regards to this thread, which has not yet been responded to. LivelyRatification (talk) 02:25, 30 September 2025 (UTC)[reply]

I also should note that this user has not responded to any attempts on their talk page to discuss the situation, including from you, myself, and another user. LivelyRatification (talk) 02:31, 30 September 2025 (UTC)[reply]

Looking at their xtools, I see no usage of user talk pages which isn't great. Sarsenet•he/they•(talk) 03:42, 30 September 2025 (UTC)[reply]

They also did this [6] three minutes after you notified them of this discussion on their talk page NicheSports (talk) 04:00, 30 September 2025 (UTC)[reply]

I did see this. It was very strange. LivelyRatification (talk) 04:11, 30 September 2025 (UTC)[reply]

Gnomingstuff is the OG of LLM text identification and tagging (muchas gracias) but I slightly disagree with their perspective on this situation. We don't yet have policies about LLM use in articles but unreviewed use of LLMs is highly likely to lead to violations of WP:V, WP:RS, WP:OR, and WP:NPOV, while high-frequency misuse of LLMs is likely to violate WP:MEATBOT. I would recommend going through some of the user's edits and finding a few examples of the LLM hallucinating a tangible claim or reference - if they have been adding unreviwed LLM-generated content to articles, you will find a bunch. Then we should follow Tbound2's edits and if they continue to make (likely) unreviewed LLM-generated edits without responding to attempts to discuss with them, just notify one of the admins who follows this page, like Newslinger. Thank you for bringing this up! NicheSports (talk) 04:52, 30 September 2025 (UTC)[reply]

Looking at the articles now, Rasheed has 5/11 sources that 404 and when validating the first source there is some information that is hallucinated (the article only mentions her quote saying she is improving her "opposite" foot, but the LLM took that and said "her left foot" in article), but I was unable to tag G15 as there was content written by other editors. I was able to tag G15 for the third article as it both had 404 sources and irrelevant sources, and no other editors added prose to the article.

I've left a warning and I think we can go to ANI if they continue to edit this way. Jumpytoo Talk 04:52, 30 September 2025 (UTC)[reply]

I've also tagged Keeley Skepper for G15 as all of the sources 404ed. Jumpytoo Talk 04:56, 30 September 2025 (UTC)[reply]

Thank you so much! If I have the time I'll also try and take a look at some of these articles, as I said I didn't have time to investigate them deeply beyond their prose. I do hope Tbound2 is able to respond to the criticisms left on their talk page, but I'll keep an eye for any future problematic editing. LivelyRatification (talk) 04:58, 30 September 2025 (UTC)[reply]

Well uh they've "responded" in the form of deleting a thread from this page as "spam"... Gnomingstuff (talk) 12:30, 30 September 2025 (UTC)[reply]

Took a quick look at the references on The Original Pancake Kitchen:

• A reference has a hallucinated title and date
• One reference seems completely hallucinated

Gurkubondinn (talk) 09:05, 30 September 2025 (UTC)[reply]

see subpage
Article Tagged Fixed G15 India Rasheed Y N Y Keeley Kustermann Y N N Georgia McKee Y N Y Brooke Smith (footballer) Y N N Lily Tarlinton Y ~ N Keeley Skepper Y N Y Poppy Scholz Y N N The Original Pancake Kitchen Y Y N

Created above to track cleanup progress, feel free to edit, add, change, remove, or collapse as needed. fifteen thousand two hundred twenty four (talk) 05:27, 30 September 2025 (UTC)[reply]

I've looked through a number of other articles created by Tbound2 and none that I saw seemed to immediately qualify for G15 speedy deletion, as the references seemed to work. I think Mitchell Sariovski is not notable enough to warrant an article so I'm going to AfD that one. I'll also leave a message on Tbound2's talk page again expressing my concerns. --LivelyRatification (talk) 07:32, 30 September 2025 (UTC)[reply]

Whilst I understand everyone's concern, please draftify or userfy, rather than delete these if you feel they aren't suited to publication yet. Whilst this editor's efforts are misguided to some degree, ever since the removal of WP:NAFL and stricter enforcement of WP:GNG, the coverage of female AFLW players has been very poor here (see how few bluelinks are on List of AFL Women's debuts in 2024 and List of AFL Women's debuts in 2025), and this is at least an effort to address the systemic bias that exists here. The-Pope (talk) 06:58, 4 October 2025 (UTC)[reply]

An article which meets G15, the criteria any of the above have been deleted under, is an article which is essentially raw output from an LLM, it will be better to make a new article than to require other editors carefully review and fix the neutrality, formatting, and hallucination issues typical of model output. If an editor wants to fix up raw LLM output, then they can visit any number of chatbots and have it delivered hot and fresh to themselves, without using up the project's time on cleanup as has happened here. fifteen thousand two hundred twenty four (talk) 08:58, 4 October 2025 (UTC)[reply]

I have fixed The Original Pancake Kitchen. Let me know if I missed anything. Kovcszaln6 (talk) 10:08, 19 October 2025 (UTC)[reply]

• Tracking subpage created, 4 articles need review. fifteen thousand two hundred twenty four (talk) 02:27, 28 November 2025 (UTC)[reply]

Wk3v78k23tnsa has made many edits recently primarily to Vince Guaraldi/Charlie Brown-related articles. I noticed some AI tells on one of them while searching, and more jumped out when skimming the other articles. And then I found this. This diff adds a lot of supposed quotes, such as this: Trotter does not just arrange, he amplifies. That bossa nova piece could have been background filler, but instead it feels deeply emotional—like it is telling its own story. Which sounds exactly like ChatGPT coming out of someone's mouth. Annoyingly, the interview it is cited to is a 40-minute video, but it does have a transcript; while I'm pretty sure the transcript is auto-generated, I CTRL-F'd multiple words out of that quotation and not one of them shows up in the transcript, and while machine-learning transcripts can mess up they usually don't mess up that much. So I did the same with some other quotes from the article and still have yet to find a hit. Fucking sheesh.

I don't know how far back this issue goes. Before the Charlie Brown stuff they did a lot of plot summary revisions, which I'm not sure are AI -- for instance, this edit is tonally glib, but it sounds more like human ad copy than LLM slop (this is a gut feeling), and there's at least one grammatical error uncharacteristic of AI. (Also, they seem to have radically changed the way they write edit summaries between now and then.) Gnomingstuff (talk) 00:07, 3 October 2025 (UTC)[reply]

I documented new concerns at Special:Diff/1319145378. I asked for clarification and if I made a mistake I will strike this. But would appreciate if someone here could take a look NicheSports (talk) 01:38, 28 October 2025 (UTC)[reply]

I think you might be right. With the caveat that I haven't checked other recent edits from this user yet. Gurkubondinn (talk) 03:31, 28 October 2025 (UTC)[reply]

Update: There's also this edit with the chatbot response Here is the revised version with all formatting removed while maintaining the academic tone and word limit:, so I guess their rewrites may now be in play too. Gnomingstuff (talk) 00:42, 3 October 2025 (UTC)[reply]

This one is tough because the user has a high level of English fluency (see this exchange on their talk page from 2016). Then you have this, which feels more like a troll than an LLM. But based on the smoking gun you found that included the chatbot response, I think we have to assume that the majority of their plot summary rewrites involved LLMs. I mean on March 2 2024 they rewrote 108 plot summaries, often 1-2 minutes apart, without grammatical errors. There are many examples of this. On Feburary 17 2024 they rewrote these 7 plot summaries in 9 minutes: [7], [8], [9], [10], [11], [12], [13]. And of course they never made any edits like this prior to November 2022. NicheSports (talk) 02:35, 3 October 2025 (UTC)[reply]

Good grief. Sarsenet•he/they•(talk) 04:29, 3 October 2025 (UTC)[reply]

I manually restored the A Boy Named Charlie Brown § Plot section to the revision from before it was rewritten with LLM garbage: Diff/1315012813/1269499672. --Gurkubondinn (talk) 11:19, 4 October 2025 (UTC)[reply]

Just so everyone knows, the supposed Trotter quote listed above is back in the article. Either it was there before the last reversion, or someone put it back in manually. I don't have time to watch the video to correct it myself, but want to flag it. WeirdNAnnoyed (talk) 11:05, 6 October 2025 (UTC)[reply]

I left a note on their talk page about the issue(s), waiting to hear back. Gnomingstuff (talk) 14:51, 6 October 2025 (UTC)[reply]

• Comment - Wk3v78k23tnsa is currently working on resolving the AI issues on the Peanuts articles according to their most recent contributions. sjones23 (talk - contributions) 19:14, 16 November 2025 (UTC)[reply]

  @Wk3v78k23tnsa I see that you've been endeavoring to tidy up the edits yourself (thank you!), do you think you'll be able to handle the totality of the cleanup on your own? If not, could you please disclose the date ranges of when an LLM was used so I can create a list of articles that other editors can help review. fifteen thousand two hundred twenty four (talk) 01:45, 29 November 2025 (UTC)[reply]

This is another big one.

For the past year-ish, Ritwik Deuba has been adding hundreds of instances of text with consistent AI indicators (though nothing unambiguous that I've found), primarily to world politics articles. One other person has pointed this out, but they didn't respond; I left another edit on their talk page.

I'm less sure about this one than some of the others. It looks like they are probably editing the output somewhat -- although not completely enough to be detectable. It also seems like they may be using a newer LLM model and/or different prompts than most do; the older edits seem more problematic than the newer ones. In particular, most of the "reflecting the significance" AI opinion-ese is attributed as other people's opinion rather than just dangling as unattributed editorializing like it usually does, but if it's AI it still may not be an accurate summation of what those people said.

The big challenge here is that almost literally all of these edits involve very contentious geopolitical topics, and some of them have been dragged into edit wars (that don't seem to be about AI). Because of this I haven't touched them with a 1000-foot pole, besides adding the AI generated tag if the indicators were clear enough. The articles are also high-traffic enough that "just reverting" would be difficult to impossible, and might re-ignite the edit warring. So frankly I don't know what to do. Gnomingstuff (talk) 17:41, 3 October 2025 (UTC)[reply]

hmmm. do you have any diffs we can look at? I checked the edits that the other person pointed out and didn't find any hallucinations - all claims are backed by the cited source. LLMs hallucinate at high rates so I think that edit would have been significantly human reviewed. This is just one diff of course so could be missing something NicheSports (talk) 18:17, 3 October 2025 (UTC)[reply]

ok finding problems. I noticed that these three edits [14], [15], [16] were made within 15 minutes (although with significant overlap in content), without any subsequent copy-editing or typo-fixing. This one also has clear copyright violations - to the point it should be reported, which I have never done NicheSports (talk) 18:29, 3 October 2025 (UTC)[reply]

Will take another look tomorrow. I mentioned this on a few of the talk pages but some of the edits are the same paragraph added to multiple articles (which isn't necessarily a problem by itself).

For stuff like this I'm less concerned about blatant hallucinations so much as any NPOV issues/interpretations that might have been introduced by the AI. Gnomingstuff (talk) 05:58, 4 October 2025 (UTC)[reply]

It appears to be someone adding news stories with the assistance of AI. LDW5432 (talk) 15:12, 6 November 2025 (UTC)[reply]

Hintha is a prolific and experienced editor who seems to have started using LLMs at some point. They have made thousands of edits since 2022 so this will be a major cleanup effort. In 2025 alone they have created 40 articles and done hundreds of article expansions. Puffery is a problem but the major issues are widespread WP:V failures and occasional, sometimes significant, copyright violations. LLMs may not be the cause of all of these issues but I am confident it is involved. Their created articles do not qualify for WP:G15 - Hintha seems to be at least ensuring that references exist. They are aware of my concerns as they addressed many of the issues with the Bowkylion article after I documented issues on the talk page. But they have not responded to any of my comments on their user talk page or on any article to which they contributed. They have been inactive for the past two weeks.

• Bowkylion: see this version before some of these issues were fixed. I documented the issues, including a minor copyright violation, on the talk page
• These three articles [17][18][19] cover similar subjects to Bowkylion and were created within 50 minutes of each other
• Private Security Services Law about 20% of this article (300+ words) was copied word for word from [20]. In an LLM tell, some of the copyrighted material was cited to a source other than the one it was copied from. I filed a copyvio report and it has been handled, with revdeletion pending. The editor from CCI who addressed the copyright cleaned up the rest of the article including this WP:V failure [21]
• Fish patty two of the three sources for this article are recipe blogs. The article also has unsourced puffery: bears a striking resemblance to Spanish empanadas and Southeast Asian curry puffs

NicheSports (talk) 13:32, 15 October 2025 (UTC)[reply]

Shouldn't this be in ANI? This falls more into disruptive editing especially as this isn't just an LLM only issue 212.70.110.25 (talk) 14:42, 15 October 2025 (UTC)[reply]

The point of my post is to document the need for cleanup, not to pursue sanctions. Also I think its too early for ANI - this is a long-term editor who is currently (likely temporarily) inactive. They deserve a chance to respond here, hopefully they will be willing to stop using LLMs and help with cleanup. NicheSports (talk) 16:07, 15 October 2025 (UTC)[reply]

Can anyone take a look at this? Would like a second opinion given this is a very experienced editor NicheSports (talk) 20:12, 20 October 2025 (UTC)[reply]

Went through history -- I could be wrong but the LLM use feels fairly recent, nothing prior to 2024-ish jumped out at me. You also probably saw this but they seem to have switched to MediaWiki's content translation tool.

Cleaned up Inya Lake Hotel -- this one honestly wasn't too bad, idk how reliable the sources are but besides a few promotional asides everything was backed up as claimed. Gnomingstuff (talk) 19:46, 22 October 2025 (UTC)[reply]

Did Waziya Cinema, the main problem here seemed to be questionable sourcing (one scoure seemed to be AI-generated) and some close paraphrasing but no major hallucinations. Honestly, I don't really feel like this rises to the level of ANI, the editor seems conscientious enough.

In general, apologies for not doing more actual cleanup -- I don't really consider myself much of a writer. Gnomingstuff (talk) 20:15, 22 October 2025 (UTC)[reply]

Yeah I have no interest in ANI here. Just wanted to get a second perspective on whether LLMs were involved (thank you!). Is it fair to say that most of their substantive edits in 2025 should be checked, but the issues may not go back further than that? NicheSports (talk) 20:18, 22 October 2025 (UTC)[reply]

not sure, apologies - a lot of it is linguistics articles that are way outside my wheelhouse Gnomingstuff (talk) 20:24, 22 October 2025 (UTC)[reply]

A number of editors have noticed that @Gengeros has added realistic but non-existent citations that seem to be generated by ChatGPT, and have asked Gengeros to be more careful. Affected articles have included Charlie Kirk, Ann Coulter, Transgender personnel in the United States military, Social conservatism in the United States, LGBTQ rights in Rwanda, LGBTQ rights in Burundi, and LGBTQ rights in Hungary. Examples of concerns: September 15, 2024 (acknowledged using ChatGPT), March 12, 2025, September 12, 2025 (also acknowledged using ChatGPT), September 30, 2025, and most recently me (October 17, 2025).

I do want to assume good faith; I believe their intent is to make constructive contributions. However, this pattern means that they have made a lot of edits that look decently cited at a glance, but actually need thorough review. I've reviewed some of their recent edits, but more help is needed for review and cleanup. Dreamyshade (talk) 15:55, 19 October 2025 (UTC)[reply]

Given that the user ignores questions about their edits and polite pleas to stop adding LLM generated nonsense, would it be reasonable to take this to WP:ANI? Gurkubondinn (talk) 23:40, 21 October 2025 (UTC)[reply]

Absolutely. You could also consider WP:AE. Kovcszaln6 (talk) 13:46, 22 October 2025 (UTC)[reply]

After reading through User talk:Gengeros § Removal of additional AI content I agree with Kovcszaln6, this should be taken to ANI right away NicheSports (talk) 14:45, 22 October 2025 (UTC)[reply]

Concurring with the others here. Either you or more likely @Dreamyshade should take this to ANI. Athanelar (talk) 16:48, 22 October 2025 (UTC)[reply]

Thanks all, appreciate the validation that this is a serious concern. I'm working on an ANI post. Dreamyshade (talk) 17:07, 22 October 2025 (UTC)[reply]

Posted here: Wikipedia:Administrators' noticeboard/Incidents#Pattern of LLM edits on contentious topics. Dreamyshade (talk) 17:52, 22 October 2025 (UTC)[reply]

Thank you for filing. Btw you might want to move the ANI notice to a new section on their talk page - admins do check to verify that users have been notified and they might miss it at the bottom of the long thread there NicheSports (talk) 18:10, 22 October 2025 (UTC)[reply]

I'm working on cleaning up Transgender rights in the United States, this is a mess. ~ Argenti Aertheri^(Chat?) 21:54, 28 October 2025 (UTC)[reply]

In this teahouse thread the user in question says they've been "using AI on Wikipedia for about 4 something months now [...] hundreds of times" and one reply already notes they reverted a couple of the user's edits due to the results being poor. Might be worth giving some scrutiny to this editor's contributions to see if things need to be reverted/changed. Athanelar (talk) 13:52, 21 October 2025 (UTC)[reply]

I checked one of their most recent significant edits [22] and found multiple verification failures. Agreed this user's entire edit history will need to be checked NicheSports (talk) 15:18, 22 October 2025 (UTC)[reply]

I documented more extensive issues at their talk page. I also notified them about this thread. They are on a wikibreak now, hopefully when they come back they will discuss what happened here NicheSports (talk) 15:46, 22 October 2025 (UTC)[reply]

Maybe it's worth making a {{AINB-notice}} template for informing users like ANI/COIN etc. Athanelar (talk) 15:50, 22 October 2025 (UTC)[reply]

@Athanelar: how does User:EF5/AINB-notice look? ^EF⁵ 16:37, 22 October 2025 (UTC)[reply]

It's stylistically consistent with the ANI and COIN notices which is good. I'd reword to "regarding an AI cleanup investigation which may include activity you were involved in" or something to that effect though because the current verbiage implies that the warned user may be involved in the investigation itself.

Also make sure it's editable with a |thread=threadtitle parameter like the ANI and COIN notices. Athanelar (talk) 16:44, 22 October 2025 (UTC)[reply]

Good idea, but would you and EF5 be fine moving the template discussion to WT:AIC? Once there you should ping Chaotic Enby as they have helped with a lot of the template work here. NicheSports (talk) 16:53, 22 October 2025 (UTC)[reply]

Thanks! I was also thinking of bringing it up here, but hesitated as I had already given them advice at their talk page, good initiative! Chaotic Enby (talk · contribs) 17:13, 22 October 2025 (UTC)[reply]

see also this village pump thread Gnomingstuff (talk) 19:29, 22 October 2025 (UTC)[reply]

I think CostalCal should be indeffed based on editing disruptively and not wanting to correct course. The WikiBreak doesn't do it for me. —Alalch E. 14:48, 25 October 2025 (UTC)[reply]

Note, CostalCal has recently returned from their wikibreak, about ten minutes ago. A block of any length would be justified or unjustified depending on how they edit and act now they've returned, and if there is evidence of continued LLM usage. Hurricane Wind and Fire (talk) 00:50, 26 October 2025 (UTC)[reply]

I hope they are able to turn stuff around, since they are a self-declared young editor and they may not understand how bad LLMs really are for writing encyclopedically/accurately. Sarsenet•he/they•(talk) 01:55, 26 October 2025 (UTC)[reply]

Yet another long series, this time related to singer Jolin Tsai. I'm tagging the articles now but also bringing the issue up here as a permanent note, because based on the user's talk page, I suspect the tags aren't going to be around for very long and that this isn't going to go well. I'm not looking forward to this. Gnomingstuff (talk) 21:44, 22 October 2025 (UTC)[reply]

This editor has authored ~60% of the Jolin Tsai article, oof. --Gurkubondinn (talk) 16:36, 23 October 2025 (UTC)[reply]

They said earlier that they're translations; I haven't had time to confirm but I don't have any reason to disbelieve it. But if they're translations they specifically seem like AI translations, given the text that came out of the translation. Gnomingstuff (talk) 17:27, 23 October 2025 (UTC)[reply]

I went back and read their reply to your questions, and to be fair they just said that they "did not use any AI tools to fabricate or generate text". I want to assume good faith, but the translations could have been produced by an LLM. --Gurkubondinn (talk) 21:17, 25 October 2025 (UTC)[reply]

RobertoBriago (talk · contribs)

Being worked on: Newer user with ~380 edits, many to talk space. Those in draft/article space include article creations and expansions, all of which require review NicheSports (talk) 17:51, 28 October 2025 (UTC)[reply]

Hi! I've been browsing wikipedia articles and I've found a few pages where LLMs were probably used, these pages are the page on bit, the page on revolut, the page on endstream and the page on the mir page. I've cleaned revolut a lot but I think it might benefit from more eyes. 173.206.50.207 (talk) 20:10, 25 October 2025 (UTC)[reply]

there's also the article on HSBC Life that seems LLM generated 173.206.50.207 (talk) 20:31, 25 October 2025 (UTC)[reply]

The same editor that created the bit (payment application) article also created AccessiWay, which other editors have then tried to clean up LLM generated text from, and tagging it as {{AI-generated}}, but the editor removed the hatnote. --Gurkubondinn (talk) 21:05, 25 October 2025 (UTC)[reply]

Asked the editor on User talk:Eliezer1987 § LLM use. --Gurkubondinn (talk) 21:14, 25 October 2025 (UTC)[reply]

i'd also like to point out that, for accessiway, there isn't really ways in which one can automate WCAG standards per https://html5accessibility.com/stuff/2025/03/27/mind-the-wcag-automation-gap/ and https://karlgroves.com/web-accessibility-testing-what-can-be-tested-and-how/ and even W3 points it out at https://www.w3.org/WAI/test-evaluate/tools/selecting/ 173.206.50.207 (talk) 23:47, 25 October 2025 (UTC)[reply]

Hi, first of all, @Gurkubondinn, regarding AccessiWay, I didn't remove the AI-generated tag. I saw the comment and plan to address it later on.

As I mentioned on my talk page, I mainly use AI for translating texts and articles that I first write in Hebrew (which is my primary language).

This is actually the first time I've come across the guidelines about using AI for writing on Wikipedia. I haven't finished reading all the instructions yet, but I'll get to that soon.

Besides that, I guess I'm not the first to say that, in my opinion, artificial intelligence can really help make Wikipedia more reliable and powerful, though of course it also comes with serious risks... Eliezer1987 (talk) 06:22, 26 October 2025 (UTC)[reply]

@Eliezer1987: Do you check the references you (or the AI) cite, and ensure they align with what the article says? ARandomName123 (talk)^{Ping me!} 06:50, 26 October 2025 (UTC)[reply]

When it's content I write from scratch, yes. When it comes to translations I make from Hebrew Wikipedia, unfortunately, I sometimes don't delve deeply into the sources. Eliezer1987 (talk) 07:23, 26 October 2025 (UTC)[reply]

I highly advise you cease using LLMs in any capacity on Wikipedia until you've fully read and understood the LLM essay. While LLM usage is not yet banned on Wikipedia, there is a certain level of understanding and competency about the problems that using AI can create that anyone using an LLM is expected to have. There are more problems than just unreliable references. It's especially risky to use an LLM for machine translation if you're not 100% confident you can verify the accuracy of the output. Athanelar (talk) 07:37, 26 October 2025 (UTC)[reply]

As you can see, I have been an editor long before artificial intelligence changed our lives... Of course I will read the guidelines and follow them. Eliezer1987 (talk) 09:49, 26 October 2025 (UTC)[reply]

@ARandomName123 It doesn't seem that they do, see this page creation where most of the links were never in the internet archive, this other page which contains poor sourcing, links that don't lead to anywhere like one would expect from the text like this one or that one, there are also bizarre translations like in this article's further reading page where the further reading references have been translated to english(!) 173.206.50.207 (talk) 16:11, 26 October 2025 (UTC)[reply]

To say that I translated or wrote all the articles perfectly? Probably not. Unfortunately, these are mostly articles that I wrote in a hurry and did it from my cell phone. In any case, as I wrote above, I will read the instructions thoroughly before repeating the same mistakes. Dear Anonymous, I would be happy if you would also help with correcting the articles, and not just identifying the problems.Eliezer1987 (talk) 19:16, 26 October 2025 (UTC)[reply]

Is the AI really labor-saving if it introduces problems that editors have to spend time identifying and correcting? Einsof (talk) 20:38, 26 October 2025 (UTC)[reply]

@Eliezer1987 Could you explain what you meant by "I would be happy if you would also help with correcting the articles, and not just identifying the problems."? 173.206.50.207 (talk) 19:22, 27 October 2025 (UTC)[reply]

My approach is that Wikipedia is a collaborative project where we aim to improve, not to blame. You have pointed out some issues that were done carelessly (by me, I apologize and will also make sure to fix them). Instead of searching for more mistakes, I would appreciate if you could join in improving them together. Eliezer1987 (talk) 03:37, 28 October 2025 (UTC)[reply]

Thanks for checking, it's unfortunate that they weren't verified beforehand. (also the fixed ping worked) ARandomName123 (talk)^{Ping me!} 02:35, 27 October 2025 (UTC)[reply]

Special:Contributions/Scewing A couple of this user's pages were recently CSDed since one edit included communications for the user. [23] After reviewing their other edits, it seems to me like pretty much everything since June is LLM generated, or at least LLM assisted. Many references are DOA, don't match content, or are straight up references to Wikipedia itself. Everything since June (at least) needs to be reviewed. Lovelyfurball (talk) 21:06, 27 October 2025 (UTC)[reply]

Totally perplexed that someone apparently editing Wikipedia since 2007 is only learning now that Wikipedia articles aren't valid sources for other Wikipedia articles. Einsof (talk) 21:19, 27 October 2025 (UTC)[reply]

It makes me feel like the original owner of the account is no longer in control of it for one reason or another. It just makes no sense to me that such an experienced editor would do this. Lovelyfurball (talk) 22:37, 27 October 2025 (UTC)[reply]

Could also just as likely just be an easy excuse to avoid admitting to uncritical use of LLMs and blindly trusting the output. Gurkubondinn (talk) 22:41, 27 October 2025 (UTC)[reply]

I've removed the LLM generated rewrite of Cooper Township, Clearfield County, Pennsylvania and WP:BLARed North Ridge, Alexandria, Virginia, which had several citations to other Wikipedia articles. A couple articles are still up with CSD notices. Lovelyfurball (talk) 22:36, 27 October 2025 (UTC)[reply]

Markdown syntax: Diff/1319074348. Gurkubondinn (talk) 16:00, 28 October 2025 (UTC)[reply]

Came across the editor on NPP reviewing Magic fallacy, which I'm probably going to redirect somewhere, but most if not all of their major insertions appear to be unedited or minimally edited generative output. I've undid Catallactics and some others have also already been removed, for example at Antisemitic trope, thanks to Cdjp1, however there are still other edits that need to be assessed and cleaned up, such as their edits to Ludwig von Mises, which they have apparently generated 43% of current content and Gabriel Bonnot de Mably, 16%. I think those two are the major ones, but I'm mostly listing it here because I'd probably forget otherwise. Alpha3031 (t • c) 12:36, 28 October 2025 (UTC)[reply]

I noticed that this edit to Ludwig von Mises added a URL that gives a 404 error and was never archived in the Wayback Machine. Seems iffy. Stepwise Continuous Dysfunction (talk) 22:18, 30 October 2025 (UTC)[reply]

Probably going to just remove it all once I have the time to get to it. Have removed the additions to Gabriel Bonnot de Mably, added 50 words myself. I had also removed the addition to Cognitive appraisal, just forgot to note that, so pages that still need to be reviewed are Polylogism (diff, 53.1%), Destructionism, Magic fallacy (I had tagged it but didn't finish looking at it) and I think that should be all. Alpha3031 (t • c) 03:43, 5 November 2025 (UTC)[reply]

Have still been thinking of a redirect target, and I think I'd want to do it to Magical thinking if anywhere given the attestation given in the article appears to be entirely hallucinated. (Indeed, the LLM appears to have resurrected Hayek to publish the book it's attributed to in 2021, after his death in 1992 and the actual year of publication which seems to be 1988, and there are no mentions of the two words together in, AFAICT, either that book or the other two cited, The Road to Serfdom (1944) and The Counter-Revolution of Science (1952), though a few* of one or the other)

Concerningly for cleanup efforts, the IPs which have also edited this made up nonsense, which I'm provisionally calling Special:Contributions/142.245.0.0/16, though I'm not sure if a wider or narrower range could capture things better, has also inserted things into several articles, including mentions of this "magic fallacy" which fails attestation. No doubt those edits also need to be reviewed, ugh. They replied to the IP on Talk:Wall Street crash of 1929 § Unorthodox economic interpretation being presented as NPOV in opening remarks so I'm not sure if they're two people just happening to chance on the same obscure article (on an apparently made up phrase) or a case of MEAT or WP:LOUTSOCK.

* For a list of mentions that I have been able to find,

I think I might just go for an AFD nom actually. Someone else can decide whether they want the redirect or not. Alpha3031 (t • c) 09:06, 5 November 2025 (UTC)[reply]

The line in Polylogism about how ..the writings of Thomas Kuhn and others made relativism a mainstream doctrine is a brutal oversimplification of the philosophy of science. It doesn't reflect the given source, for sure. Furthermore, I suspect the entire "Comparison to Kuhn's incommensurability" section to be OR, whether LLM-enhanced or not.

Moreover, if we look at what the Ludwig von Mises article itself says about "polylogism", we find the following: He rejected the notion that there could be distinct sciences or truths based on race, class, or nationality, such as "Jewish science" or "German science". The given source is a book that does not mention Ludwig von Mises at all. Perhaps unsurprisingly, the text was written first and the citation added later. Stepwise Continuous Dysfunction (talk) 17:53, 6 November 2025 (UTC)[reply]

The book talks about jewish vs german science, and is a citation for such. Follynomics (talk) 18:14, 11 November 2025 (UTC)[reply]

But it doesn't talk about Ludwig von Mises, so it doesn't support the claim to which it was attached, i.e., that Ludwig von Mises rejected the notion that there could be distinct sciences or truths. Stepwise Continuous Dysfunction (talk) 01:14, 15 November 2025 (UTC)[reply]

"Magic fallacy" has been deleted. Stepwise Continuous Dysfunction (talk) 01:16, 15 November 2025 (UTC)[reply]

I don't have much experience with detecting AI, but West Vinod Nagar (edit | talk | history | protect | delete | links | watch | logs | views) looks suspicious to me so bringing it here for more scrutiny. FDW777 (talk) 15:47, 28 October 2025 (UTC)[reply]

At the very least, it's copyvio. I'm seeing copy paste from some of the sources. Likely LLM, too. Lovelyfurball (talk) 16:01, 28 October 2025 (UTC)[reply]

Per this, I'm thinking it's not much of a leap to assume that's an admission it applies to this article too. FDW777 (talk) 16:03, 28 October 2025 (UTC)[reply]

The article in question has now been deleted, the rest of this user's edits need to be reviewed Lovelyfurball (talk) 16:19, 28 October 2025 (UTC)[reply]

It seems like all the major pages they made have been deleted. Lovelyfurball (talk) 16:21, 28 October 2025 (UTC)[reply]

While doing CheckWiki project edits, I stumbled upon two articles, Knowledge process outsourcing and Computers, Freedom and Privacy Conference, which both contained edits that I suspected of being AI-generated. Both users are included in the list of students attending the above course. Unrelated but I randomly clicked on one user's edit and found them to be illegitimately blanking sections. I'm not sure that because this is a Wiki Ed course that the expert would handle this, but I'm not familiar with Wiki Ed. Something to look out for? Ecourter (talk) 08:45, 31 October 2025 (UTC)[reply]

I'm not sure either, but I posted this to WP:ENB so that someone who knows more might weigh in. Einsof (talk) 13:02, 31 October 2025 (UTC)[reply]

Revert their edits and have a stern word with the educator and the Wikipedia ‘expert’. WikiEd has always been a disaster and the Wikipedia editors who help facilitate it are generally more interested in making it happen than even listening to the many many many issues the WikiEd courses cause everyone else. Kingsif (talk) 14:16, 31 October 2025 (UTC)[reply]

It's discouraged in their training materials so definitely let them know. Gnomingstuff (talk) 17:51, 31 October 2025 (UTC)[reply]

Let the instructor know but do be careful with wiki-ed articles, it seems like they are treated differently by some admins. I came across Mattie Jean Adams and G15'd it, an admin rejected it, another editor G15'd it, also denied, then the course instructor G15'd it, it was finally deleted, the student recreated the article in mainspace, and it had to get deleted again. Ugh. I come across WikiEd articles all the time and they are almost always terrible NicheSports (talk) 18:27, 31 October 2025 (UTC)[reply]

Hi all, thanks for jumping in! We at Wiki Education set up a monitoring system using the AI detection software Pangram, which we've found to be relatively accurate in regard to detection of AI generated or assisted content on Wikipedia. We're running student editor diffs through Pangram and creating tickets in our system for our Wikipedia Experts to review. In the case of Computers, Freedom and Privacy Conference, our system caught the same edit that was reverted (see the Pangram report here). We are working through the list of tickets generated (we had a backlog as both our Wiki Expert staff attended WikiConference North America). So thank you to @Ecourter: for reverting before we got to it; we've since followed up with the instructor to ask them to intervene as well. For the Knowledge process outsourcing, Pangram detected no AI used, although no system is 100% accurate. Was there something specific about this that made you think it was AI generated?

Our concern centers particularly around accuracy — we don't want students adding information that fails verification or is otherwise erroneous. Given the challenges AI chatbots have with this, our trainings stress students shouldn't use gen AI for content creation, but may find it helpful during the research process if that's acceptable with their instructor. --LiAnna (Wiki Ed) (talk) 22:39, 31 October 2025 (UTC)[reply]

Regarding Knowledge process outsourcing, I found WP:AITITLECASE and WP:CONCLUSION. I only tagged this article for potential AI usage since I believe it was not beyond a reasonable doubt. This was one of the first times that I encountered potential AI writing so another editor with a better eye could have a more definitive answer. Ecourter (talk) 22:59, 31 October 2025 (UTC)[reply]

Not an unreasonable tag, the added "Challenges" and "Resolution" sections are rough, with prose like: The ever-evolving field of knowledge process outsourcing will continue to improve its current outsourcing practices to develop new process driven and knowledge driven solutions. It's uncommon to see human editors write such fluent ever-evolving knowledge-driven impactful risk-aware innovative... business speak. fifteen thousand two hundred twenty four (talk) 00:18, 1 November 2025 (UTC)[reply]

The first one I'm not sure about, honestly. It has some minor quirks typical of AI, but is grammatically messy in consistent ways that feel less so. I think maybe the outline might have been AI-assisted, and later rewritten/expanded upon. Gnomingstuff (talk) 04:49, 1 November 2025 (UTC)[reply]

That linked outline was copied from the article, and all its sections are word-for-word identical. There may have been some form of expanded outline process, but that linked one isn't part of it at least. fifteen thousand two hundred twenty four (talk) 05:16, 1 November 2025 (UTC)[reply]

Our concern centers particularly around accuracy ... our trainings stress students shouldn't use gen AI for content creation – This is good, but as stated by multiple editors (myself included) the last time this training was discussed, there are concerns outside of content creation and accuracy, like introduced bias via source and "information gap" suggestions. (Though I am glad to see that my concerns about copyediting were addressed!)

our system caught the same edit that was reverted – I see in the report that the citation^[4] numbers^[5] aren't stripped from the text before being processed, but they probably should be. They're uncommon in human-created text, aren't explicitly added by the editor, and may impact the false positive or negative rate. fifteen thousand two hundred twenty four (talk) 01:07, 1 November 2025 (UTC)[reply]

Good point about stripping the citation numbers. I will add that to the processing pipeline. Sage (Wiki Ed) (talk) 17:56, 3 November 2025 (UTC)[reply]

AlphaBetaGamma123 has made hundreds of quick-succession rewrites, almost all of which display the usual signs. Working through tagging them currently.

(At some point the editor was blocked for sockpuppeting reasons.) Gnomingstuff (talk) 06:04, 4 November 2025 (UTC)[reply]

Cleanup note: all edits by this account are eligible to be WP:BANREVERTed, as they were made while violating a block via socking. fifteen thousand two hundred twenty four (talk) 11:58, 16 November 2025 (UTC)[reply]

I've discovered numerous fake references in an article Dance in Thailand since Oct 22, 2025, per Talk:Dance in Thailand#Over 30 fake references were detected which were created by @ShalaylayPumpano. [24][25][26][27][28][29] Despite my efforts, the problem has yet to be resolved even @Xan747 has addressed the issue, ShalaylayPumpano AI-generated edits with fake citations. Additionally, the content of the article is riddled with references generated by AI. Upon verifying the authenticity of some references, I found that they had been replaced with fake ones. Seeking guidance, I consulted with @Paul_012, who suggested reverting the article to a previous version before the fake references were added.[30] However, as I'm not an editor with the necessary authority, I'm hesitant to do so, especially since I have encountered War-edits on this article in the past.

I'm seeking assistance in recovering the article here. Should we attempt to resolve the issue, or should we leave it as is? And, I apologize in advance for not having the courage to handle those issues myself. Quantplinus (talk) 04:39, 8 November 2025 (UTC)[reply]

They made two groups of edits. The most recent group can be undone by restoring the "Ayutthaya Period" section from 1302204829. The older group can be undone by restoring the "Etymology", "Origin", "Nanzhao Period", and "Sukhothai Period" sections from 1296132204 as well as by converting the entire yupho.1973 ref in "Ayutthaya Period" into <ref name="yupho.1973" />.

I've checked the edits following group one [31] and group two [32], and it appears that doing the above restores would have a minimal impact to edits following those by ShalaylayPumpano.​ fifteen thousand two hundred twenty four (talk) 07:46, 8 November 2025 (UTC)[reply]

Thank you for your help! I agree with your decision to revert to 1296132204. However, it would be pleased that you or other editors handle the reversion instead of me. The reason is, in the past, I have been involved in edit wars on this article in 2024, and I do not feel confident in my ability to handle this reversion and to resolve the conflicts. Additionally, this article is often subject to conflicts between Thai and Cambodian editors, and as a Thai editor, I would like to avoid any potential disruptions by requesting that another editor handle the reversion. I'm concerned that my involvement may be seen as disruptive by the other side. Thank you for understanding my concerns and for your assistance in this matter. Quantplinus (talk) 08:54, 8 November 2025 (UTC)[reply]

I agree with your decision to revert to 1296132204 – I made no such decision or recommendation, please read what I wrote carefully. I have no desire to edit-by-proxy, especially when considering that you have opened an ANI report related to the article [33]. fifteen thousand two hundred twenty four (talk) 18:31, 8 November 2025 (UTC)[reply]

I understand the sensitivity given the ANI stuff but I do think this is a good faith and reasonable request to have someone uninvolved in the edit war address the problematic AI content. I'm familiar with the edit warring here from a separate LLM cleanup situation at Traditional Thai clothing so @Quantplinus I will look into the LLM cleanup here NicheSports (talk) 18:51, 8 November 2025 (UTC)[reply]

Thank you for your assistance in helping me clarify my current thoughts and I appreciate that you recognize my positive intentions. I'm eager to resolve these issues but I'm unsure of how to do so. Quantplinus (talk) 02:10, 9 November 2025 (UTC)[reply]

@Fifteen thousand two hundred twenty four I apologize if my writing implies or suggests that you are responsible for decisions that you did not make. I just simply agree with Diff/1296132204 comparison of the diff, even though you did not make the decision. The ANI report is a separate issue from this one, and I'm trying to address each issue individually. Please excuse any confusion in my English communication here as I'm more familiar with the Thai Wikipedia project and I just have good faith intending to resolve the endless conflicts in the article. Quantplinus (talk) 02:05, 9 November 2025 (UTC)[reply]

Be bold and remember to ignore all rules, OP. Not just in wiki but also IRL. ~2025-32267-51 (talk) 08:02, 9 November 2025 (UTC)[reply]

To use "ignore all rules" on Wikipedia the rules must be stopping an improvement to the encyclopedia. In what way are they doing that? Phil Bridger (talk) 11:00, 9 November 2025 (UTC)[reply]

 Cleanup has been requested and is being tracked at /2025-11-09 Marcodicaprio, 33 pages need review.

This person has been making a large amount of fully LLM generated articles replacing redirects. I'll start a dialog with them but I don't have time right now to review each edit. (Though it might be worth mass-reverting depending on just how many there are.) Perryprog (talk) 16:03, 9 November 2025 (UTC)[reply]

Rudolf Sosna is an example of one that is in the middle of a deletion discussion. There are serious source-text integrity problems. -- Reconrabbit 17:35, 12 November 2025 (UTC)[reply]

Tracking subpage created, let me know if I missed anything.​​​​ fifteen thousand two hundred twenty four (talk) 11:32, 16 November 2025 (UTC)[reply]

Not sure why I was notified here, but if that guy had an open talk page I'd let him know that I get what he's doing, when you are a new editor you wonder why certain albums don't have pages for tracks or certain musicians don't have pages, but you can't really make pages for everything if it doesn't have enough notability. I'd advise him to at least gather sources on Google Books until he finds out what topics are notable enough to make pages for and then start writing them with no use of AI. Aradicus77 (talk) 11:42, 16 November 2025 (UTC)[reply]

I notified you since you had performed some cleanup after this user at The Fugs, and one of your edit summaries there was has this person used AI? Sorry for the unclear ping.

There is an open discussion on their talk page at User talk:Marcodicaprio#LLM usage. fifteen thousand two hundred twenty four (talk) 11:52, 16 November 2025 (UTC)[reply]

Yeah I've been cleaning up their Faust IV and Faust (album) at the moment. To me it looks like a user who wanted to make some good-faith contributions by expanding pages for underground bands, but is taking the lazy way out and using AI to do it all. It also seems like they are fans of these bands trying to puff them up and use weasel words like "many rock critics regard them as..." and this is creating issues because the AI will sometimes create claims and link sources that make no mention of it, or just completely conjure up the claims entirely.

I think giving the user a warning about their behavior should be the step forward, but if it persists then it would have to be a block. One thing is I get the concern of niche bands not having as much coverage on the site, but the right way to go about it is to go and look through books, read, research, open up a Google Doc and save your sources, pages, etc. And then try building up a page. It takes way longer, but it's following the rules and how you actually improve an article. Aradicus77 (talk) 12:12, 16 November 2025 (UTC)[reply]

They've not made any more LLM edits since this comment on the 10th, so I'm hopeful there won't be any future issues. I've updated /2025-11-09 Marcodicaprio to reflect the ongoing efforts, thank you. fifteen thousand two hundred twenty four (talk) 13:14, 16 November 2025 (UTC)[reply]

 Cleanup has been requested and is being tracked at /2025-11-12 KLIFE88, 15 pages need review.

KLIFE88 (talk · contribs) has been creating and expanding a lot of articles related to G-Dragon and BigBang (South Korean band). Many of these edits have issues with source-text integrity and have hallmarks of large language model use (such as an edit from Who You? that includes the line Portions of the video were shot by fans themselves and incorporated into the final edit, highlighting the concept of isolation and public observation sourced to an article that just says "G-Dragon and his fans shot the MV together at Ilsan KINTEX. The completed music video shows the fans who took part in the filming, as well as G-Dragon." Several of this editor's articles have been sent to AfD; their response often asks editors to improve articles rather than deleting them. -- Reconrabbit 15:17, 12 November 2025 (UTC)[reply]

As I mentioned earlier, I believe that fixing, improving, or adding clarification notes to articles is far more constructive than deleting them outright. Many of these articles include valuable sources that could significantly enhance the content (that I spend hours and days gathering). After all, Wikipedia's purpose is to share reliable information and promote collaboration in building knowledge, not to remove it unnecessarily. English is not my first language, and while I do my best, working with sources in Korean, Japanese, Chinese, and other Asian languages can be challenging (since the artist I try to improve their info is Korean). Despite that, I always aim to contribute accurately and meaningfully. For this reason, I strongly believe that editing, refining, and improving articles is a better approach than deleting good sections or entire pages that have potential for development. KLIFE88 (talk) 20:06, 12 November 2025 (UTC)[reply]

@KLIFE88 None of that excuses insufficient fact checking to make sure what the AI is producing matches what is being said by sources. When you don't fact check adequately, people are correct to be skeptical of your writing and even go to deletionism if you're doing too much of it. If you made yourself more trustworthy we wouldn't be having this problem. grapesurgeon (talk) 00:37, 13 November 2025 (UTC)[reply]

Tracking subpage created, let me know if I missed anything.​ fifteen thousand two hundred twenty four (talk) 12:31, 15 November 2025 (UTC)[reply]

Cleanup efforts are ongoing and being tracked at /2025-11-13 Bobsanski, 16 pages need review.

First time doing this, apologies if I did anything wrong. Bobsanski (talk · contribs)'s talk page shows numerous warnings since January of LLM-generated content, which definitely checks out for me when looking at diffs. Many articles this user has created have been speedy deleted. Seemingly no response after dozens of warnings. 11wx (talk) 04:37, 13 November 2025 (UTC)[reply]

Investigating... AlphaBetaGamma (Talk/report any mistakes here) 14:30, 14 November 2025 (UTC)[reply]

I'm going to say that I think there are at least a chunk of faulty AI product in every single one of their articles. It's late here, so I'll slam a tag on all of these and deal with them if cleanup hasn't ended by the time I'm free again. AlphaBetaGamma (Talk/report any mistakes here) 14:41, 14 November 2025 (UTC)[reply]

Tracking subpage created, let me know if I missed anything.​ fifteen thousand two hundred twenty four (talk) 06:01, 15 November 2025 (UTC)[reply]

We need an editor who's more familiar with music articles, I'm sure they can do better than me for this one. AlphaBetaGamma (Talk/report any mistakes here) 10:02, 15 November 2025 (UTC)[reply]

Cleanup efforts are ongoing, 2 pages need review.

 To do

• International Day for Mine Awareness and Assistance in Mine Action

• United Nations Information Centres

Y Completed

• Global Initiative for Chronic Obstructive Lung Disease

• United Nations Information Centre Tokyo

N Unnecessary

• Nivia Sports

Raavimohantydelhi (talk · contribs · deleted contribs · logs · filter log · block user · block log)
After deleting one of Raavimohantydelhi's articles that was tagged for G15, I noticed that his talk page was full of similar notices regarding articles being either draftified or speedily deleted, and explicit warnings by other users against this behavior. While he has stopped since being warned, his contributions from last year still show similar issues with AI generation (e.g. International Day for Mine Awareness and Assistance in Mine Action). There are still several articles to go through (both new articles and edits to existing ones), and having more eyes on the matter could be helpful. Chaotic Enby (talk · contribs) 06:40, 13 November 2025 (UTC)[reply]

• United Nations Information Centre Tokyo redirected
• United Nations Information Centre for India and Bhutan tagged for G15
• Kalinga Literary Festival, Anshuman Tiwari Cleanup not needed
• Nivia Sports Seems like their edits introducing bad content were cleaned up fairly quickly. Thanks Fram in 2024. The page looks like it has been tagbombed with ^{[citation needed]} in the lead.
• International Day for Mine Awareness and Assistance in Mine Action needs its duplicate references merged and its tone fixed. No outstanding 404 although text-source integrity is highly questionable.

AlphaBetaGamma (Talk/report any mistakes here) 10:31, 15 November 2025 (UTC)[reply]

Regarding United Nations Information Centre for India and Bhutan, I didn't put it in the list (and G15 got declined by @KylieTastic) as the edits were from 2018, before LLMs were a thing. The usual cutoff I use for these cases is December 2022 (public release of ChatGPT), and I try to be less strict with 2022–2023 edits as LLMs were less experienced back then and results were more obvious. Chaotic Enby (talk · contribs) 18:34, 15 November 2025 (UTC)[reply]

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)[reply]

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)[reply]

**Markdown** added at English afternoon tea [34][35] and Venetian window [36], and removed [37][38][39].

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)[reply]

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)[reply]

Cleanup efforts are ongoing and being tracked at /2025-11-18 Gyða1981, 67 pages need review.

Hello! I am new to this board. I came across the article Svarfdæla saga and saw that it was tagged with containing LLM content. The user who made the suspected edits to that article, Gyða1981 [40] has also made many other edits to other articles, especially ones on lesser-known sagas, that I believe to be LLM-generated. The biggest giveaway, besides no sourcing, is that many of the edits in their edit summary contain communication intended for the LLM user (e.g. "I took the information from the Norse/Icelandic source you provided and listed the events of the saga in chronological order"). I believe this user is using an LLM to "translate" non-English text and pasting it into articles.

I have put an AI notice on all the articles in this genre that I believe Gyða1981 has pasted LLM text into. I can read Old Icelandic and can write better articles for these with correct sourcing, but I don't have very much time at the moment, and probably won't for a few months, so if anyone else can do that, I would greatly appreciate it. A few of them also probably have existing English translations of the texts in question so I can also help locate those. Also, there are a number of topics that this user has touched on that I am not as familiar with, so they should probably be looked over for LLM text as well.

I'm not sure if I should warn the user in question? I am a bit non-confrontational so maybe someone else could do that if it is appropriate. Their talk page seems to be filled with other copyright issues also. But they haven't been active on Wikipedia for a few months, and maybe notifying them will make them come back and make sneakier LLM edits, I don't know.

I'm still learning editor etiquette, so apologies if I made any mistakes. Rorb lalorb (talk) 11:01, 18 November 2025 (UTC)[reply]

Tracking subpage created and populated with articles that warrant further review or cleanup.

Accurate assessment, a warning would be appropriate but Gyða1981 has not edited since 31 August. They certainly are using an LLM to generate plot summaries, as well as to make other edits. I believe this sequence of edits at After I'm Gone is emblematic of their process:

• Special:Diff/1302476101 – Added detailed plot summary based on Goodreads descriptions.
• Special:Diff/1302476261 – Added detailed plot summary based on Amazon descriptions.
• Special:Diff/1302476322 – Added detailed plot summary based on Amazon and Goodreads descriptions.

My favorite diff is this attack of the '''s.​ fifteen thousand two hundred twenty four (talk) 15:52, 18 November 2025 (UTC)[reply]

I noticed that she has also been active on iswiki so I asked her to WP:LLMDISCLOSE there, thinking it might come off more as more gentle in a smaller space: :is:Notandaspjall:Gyða1981 § LLM notkun. But I haven't heard back. --Gurkubondinn (talk) 11:15, 21 November 2025 (UTC)[reply]

I've also brought this up on :is:WP:Potturinn § Gyða1981 og LLM notkun, so this can be addressed on iswiki as well. --Gurkubondinn (talk) 14:58, 21 November 2025 (UTC)[reply]

@Rorb lalorb You can read Old Icelandic? That's a skill I wish I had. Rare and valuable! (I'm not joking.) Cheers. David10244 (talk) 23:12, 26 November 2025 (UTC)[reply]

Cleanup efforts are ongoing and being tracked at /2025-11-19 Noxoug1, 209 pages need review.

Requesting advice/assistance regarding User:Noxoug1. Their May 2025 edits to Arsenate sulfate and Cystinosis#Investigational treatments seem very likely to be AI-generated. I first encountered these in October, notifying them using the {{Uw-ai1}} template in 5 October 2025, which they archived without response. After a month's pause, their pattern of rapid LLM-assisted editing seems to have recommenced in articles like Immunocytokine and Paluratide. Vetting/oversight seems to have improved somewhat, but in my view is still likely to be inadequate given the volume of material being added to the encyclopedia. Can I get a second pair of eyes on this to check whether there's an actual issue here? (If this is actually what's going on, their Xtools edit statistics suggest they only started heavily using LLMs in May, leaving us with 4000 article-space edits to check.) Preimage (talk) 15:37, 19 November 2025 (UTC)[reply]

Confirming that this is a user who's been inserting LLM content, often with too little regard for WP:RS or copyright.

• Special:Diff/1295057234 – rich history sourced to the ever-reliable de.wikipedia.org
• Special:Diff/1298590992 – becoming a beacon of hope and innovation in nurturing young talent and inspiring the next generation of golfers copied from source, and This holistic approach has been recognized for its effectiveness in developing well-rounded young people who excel both on the golf course and in their academic pursuits sourced to tripadvisor.
• Special:Diff/1296972618 – *Imani* also contributed to the growth of Uganda’s film industry, encouraging community-driven storytelling.

etc.

The articles you've singled out structurally look like model creations to me, and some **markdown** made it into Immunocytokine (still live), so they're definitely still using an LLM. It would be helpful if an editor familiar with the topic areas Noxoug is favoring lately could audit their edits for WP:V discrepancies. fifteen thousand two hundred twenty four (talk) 17:57, 19 November 2025 (UTC)[reply]

Immunocytokine is definitely AI created. Please see User talk:Noxoug1#Problems in Immunocytokine for some details. Викидим (talk) 19:30, 22 November 2025 (UTC)[reply]

Draftified and tagged. Extensive issues with citations having an incorrect link, doi, pmc or pmid. Their other creations will likely need to be reviewed as well. fifteen thousand two hundred twenty four (talk) 01:21, 23 November 2025 (UTC)[reply]

Checked multiple article about (mostly unsuccessful) drugs. AI is used almost every time, but mostly results are not bad. Some, with hallucinated sources, were sent to draftspace. Викидим (talk) 02:08, 23 November 2025 (UTC)[reply]

Thank you, if you let me know what articles you've reviewed/how far back you've reviewed I can put the rest into a tracking list. fifteen thousand two hundred twenty four (talk) 04:09, 23 November 2025 (UTC)[reply]

I went to Posdinemab, with two PRODs and one draft. Викидим (talk) 05:58, 23 November 2025 (UTC)[reply]

Tracking subpage created, of 333 articles 209 appear to warrant further review or cleanup, but the editing pattern used has resulted in a less exact triage. Ranges checked are from 01 May 2025 to 22 September 2025, as it seems they likely started using LLMs in May. While reviewing I frequently encountered LLM prose across a wide variety of topics, although less so in medical ones.

Tracking subpage created, of 333 articles 209 appear to warrant further review or cleanup, the editing pattern used has resulted in a less exact triage. Ranges checked are from 01 May 2025 to 22 September 2025, as it seems they likely started using LLMs in May. While reviewing I frequently encountered LLM prose across a wide variety of topics, although less so in medical ones.
I've invited Noxoug1 to join the discussion to disclose more about their LLM use and hopefully assist with cleanup. fifteen thousand two hundred twenty four (talk) 09:16, 23 November 2025 (UTC)[reply]

Y Cleanup has been completed, see /2025-11-20 Bob C. Alexander for reference.

Bob C. Alexander (talk · contribs · deleted contribs · logs · filter log · block user · block log) has been making a ton of high-speed LLM edits, and I don't have the energy to review them myself because I just finished reviewing another large batch of LLM edits by WhiteFactLoom (talk · contribs · deleted contribs · logs · filter log · block user · block log). I've been seeing a lot of LLM slop in draftspace; now I'm turning my focus to mainspace for a bit and MAN do things look bleak. —pythoncoder (talk | contribs) 19:17, 20 November 2025 (UTC)[reply]

Yeah, I noticed this after they edited an article (Organizational culture) I had previously added to my watchlist after it was expanded by a group of undeclared students who were also using LLMs. A lot of the same articles are hit with LLM rewrites over and over again. No bueno NicheSports (talk) 04:24, 21 November 2025 (UTC)[reply]

The corporate/consultant/management/"leadership"-focused articles are doubly hard because so much of the verbiage there sounds like LLMs already. Even before ChatGPT I wouldn't have batted an eye at an article on organizational culture containing a sentence like "sustaining cultural shifts after major events or disruptions requires embedding new expectations into day-to-day routines, leadership behavior, and HR systems, rather than relying on single initiatives or symbolic events". Einsof (talk) 04:30, 21 November 2025 (UTC)[reply]

"MAN do things look bleak" -- yep. this has been my reaction in reverse, collecting draftspace articles to add to the AI dataset after having spent time mostly in mainspace.

The AI text-crunching script I'm working on has some loose categorization of articles based on where they come from. One folder is articles tagged as promotional/advertising prior to mid-2022, to see if it's possible to study how exactly AI promotional slop differs from human promotional slop. Comparing the drafts folder to the promo folder -- not a perfect comparison, but rejected drafts are often more skewed promotional than stuff that survived in mainspace -- the usual AI words still show up at high frequency. The datasets are not perfect matches, and they're too small still for there to be many statistically significant results for phrases, but there are a few:

• been featured in (2079% increase in AI vs. human text)
• serves as the (1416% increase)
• contributions to the (1102% increase)
• various other things that are either different slices of the above ("featured in the"), obvious flukes, or not statistically significant at this dataset size

Take this with a full shaker of salt obviously; but it feels telling that these phrases are showing up more commonly in AI text even when the human text is explicitly restricted to promotional tone and the AI text is a free-for-all. Gnomingstuff (talk) 07:03, 21 November 2025 (UTC)[reply]

• Tracking subpage created. Of 199 articles, 134 appear to warrant further review and cleanup. I'd advocate for indiscriminate mass revert in this instance, there is clear evidence of pervasive unreviewed LLM additions. Also noting that in addition to the promotional tone, the account has added 16 references to a podcast at adammendler.com/blog across multiple articles, probable refspam.

User has been adequately warned on their talk page and any future disruption should also be reported to WP:ANI. Pinging @Tacyarg, as they've reverted many of their additions already. fifteen thousand two hundred twenty four (talk) 06:32, 23 November 2025 (UTC)[reply]

MAN do things look bleak

Been going through the subpage and reverting some of this LLM nonsense and saw Diff/1319446660 with the edit summary "Removed my citation due to 404 error page". --Gurkubondinn (talk) 12:07, 23 November 2025 (UTC)[reply]

Cleanup efforts are ongoing and being tracked at /2025-11-23 HyenaInTheSyntax, 2 pages need review.

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

Many new articles with AI tells (e.g. public-history projects and media essays have highlighted her trajectory in Kateryna Hrushevska § Legacy), as well as one filter hit for Markdown formatting. I checked a citation in Oleksandra Yefymenko:

Surman, Jan (2014). "Gender, Empire and Scholarship: Oleksandra Yefymenko in Context". Austrian Journal of History. 25 (2): 77–102.

That issue (which is actually in German) does not include anything by Jan Surman or with Yefymenko in the title. jlwoodwa (talk) 02:26, 23 November 2025 (UTC)[reply]

Tracking subpage created, please let me know if I've missed anything. fifteen thousand two hundred twenty four (talk) 04:30, 23 November 2025 (UTC)[reply]

@Fifteen thousand two hundred twenty four, @Jlwoodwa Hyena admits to using LLMs "for various purposes during editing" on their Talk page and rather than agreeing to stop, they basically vow to do a better job of verifying the AI slop. David10244 (talk) 23:25, 26 November 2025 (UTC)[reply]

There's currently no policy against general LLM use, only against using them to create new articles from scratch (WP:NEWLLM), or for generating talk page comments (WP:AITALK). So long as Hyena is able to edit in compliance with core policies like WP:V and WP:NPOV, there's nothing to be done about it. fifteen thousand two hundred twenty four (talk) 00:28, 27 November 2025 (UTC)[reply]

Cleanup efforts are ongoing and being tracked at /2025-11-23 Orlando Davis, 17 pages need review.

Orlando Davis (talk · contribs) has almost certainly been using LLMs to create articles while repeatedly denying doing so. I have been watching this user's edits for a while as they have repeated triggered EF 1325.

• Examples of LLM-generated content

• Draft:Aluma Restaurant (Jerusalem): this was full of AISIGNS, and both the History and Concept and Cuisine and Staff sections contained multiple material source-to-text integrity issues. Note: OD extensively rewrote this article after I flagged the issues with it to them (without explaining what caused those issues)
• Draft:Prince Alexander Literary Prize (Belgium): this draft was declined at AfC. It contains a reference with a broken url and this seeming LLM communication intended for the user in the reference section: Multiple sources support the details in this article. It is G15-able but I left it so I can refer to it here
• All of their article creations are written in a similar style, so I assume many more WP:V issues are out there
• They are also now nominating some of their articles for GA review [41]. See the first draft of this article, which was teeming with AISIGNS [42]

• Talk page messages written in the user's voice, which is different from the prose style of their mainspace edits: [43][44][45]
• Denials of using LLMs to generate content: [46] (also see the edit summary when they reverted this message [47]) [48]
• Repeatedly blanking their talk page to remove AfC declines + warnings due LLM use: [49][50][51] (+ many more reversions of warnings about things other than LLM use)
• Recent non-collaborative talk page messages about potential LLM use: [52][53]

NicheSports (talk) 23:11, 23 November 2025 (UTC)[reply]

I am working diligently to bring these articles up to an acceptable standard, and I ask for the opportunity to complete that work. I believe this situation arises from a misunderstanding. I am following policy by working through the Articles for Creation process and revising my drafts based on the feedback provided.

Once I have acknowledged the issues and agreed to improve the material, it is unnecessary—and contrary to the spirit of constructive collaboration—to continue posting repeated messages about matters that are already being addressed. As outlined at [WP:HARASSMENT], continuing to pursue an issue after it is resolved, or repeatedly messaging a user about the same concern, can constitute harassment. The AfC process, its rejections, and the relevant tags are already sufficient to guide improvement. Additional repeated posts go beyond what is needed and are not appropriate. Orlando Davis (talk) 23:18, 23 November 2025 (UTC)[reply]

Additionally, I do not see anything inappropriate about nominating an article for GA review. My intention in doing so is simply to improve the article by inviting structured feedback. The Good Article process exists precisely to help identify areas where an article can be strengthened, and I believe the topic is important enough to warrant broader collaboration. I do not understand why this should be held against me, as seeking constructive review is entirely consistent with Wikipedia’s collaborative principles. Orlando Davis (talk) 23:25, 23 November 2025 (UTC)[reply]

• I had a question last week at this AfC review. OD said they used AI to find sources, which is not inherently a problem. But the formatting of the response they'd given looked to me -- not an expert by any means -- like what I typically get from ChatGPT. FWIW, I've interacted with Orlando Davis at multiple articles so far, and I do think they could develop into a productive editor if they'll just start listening to more experienced editors. Valereee (talk) 17:13, 24 November 2025 (UTC)[reply]
• I am optimistic that OD has heard us as they have carefully rewritten several of their article creations (thanks, OD!). @Fifteen thousand two hundred twenty four sorry for the ping, but after my conversations with OD I think I'm too involved here to constructively contribute to this cleanup. If you have time, can you please take a look at some of OD's earlier article creations? I think they may have started using LLMs more recently, so if they are rewriting their more recent drafts there may be no more cleanup required here. NicheSports (talk) 19:09, 26 November 2025 (UTC)[reply]

  Thank you for letting people in this conversation know that I have rewritten all of the articles that were flagged, except for the MasterCooks article, which I am finishing now.

  Since I have been an editor for a long time, it would be unreasonable to say that I used AI in my early work. I think the confusion comes from the AI detection program noticing that I used ChatGPT to help create infoboxes and citations. I did use it for that, because in my early Wikipedia work I did not know how to do those more advanced technical tasks, and it is now helping me with them.

  I understand that we must be careful with AI because it often writes inaccurately, but I challenge anyone to find any inaccuracies in the work that was flagged as AI-generated. I take pride in ensuring that everything I write matches the sources, but is not plagiarized. I do not add original content.

  I think that Niche does a great job trying to reduce LLM use, but there is a difference between constructive use and misuse. As far as I can tell, the work I have done is solid. I read and reread sources to make sure that I am not plagiarizing or adding original research.

  Thank you, and I appreciate your help in improving my articles. Orlando Davis (talk) 19:37, 26 November 2025 (UTC)[reply]

• Tracking subpage created, only articles edited since December 2024 were considered. Out of all considered, triage indicates 17 may warrant closer review. I personally attempted to review and correct four of the smaller contributions at Frank J. Holmes, Craftsmanship Museum, Argentina at the 2017 World Games and Wayne Wesolowski. The oldest had edits circa January. These edits primarily consisted of adding sources, and verifiability issues were identified in each. Given that Orlando Davis has used LLMs to find sources, this is a somewhat troubling result.

Concerning earlier article creations and possible LLM use: this edit from December 2024 at Samba (book), these edits from January 2025 at Static model aircraft, and this edit from January 2025 at Philip Reed (model ship maker) do not bode well. There is a wide gap in editing activity before December 2024, edits from before that gap are mainly to Guillermo Rojas Bazan, have an entirely different citation style, and do not appear to be assisted.​ fifteen thousand two hundred twenty four (talk) 05:46, 27 November 2025 (UTC)[reply]

Thank you for taking the time to take a look at my work.

Most of my articles were accepted by going through the Articles for Creation process. The exception is Philip Reed, and that article is well-sourced. Wayne Wesolowski, Frank J. Holmes, and Argentina at the 2017 World Games were random articles I found that were not doing well and are now decent articles. For example, with Wesolowski, I changed very little content, but I did add sources to an article that had almost no citations, no image, and no infobox; it essentially had no proper Wikipedia structure. Frank J. Holmes is an American cultural icon, and when an editor attempted to delete the page, I stepped in and helped keep it during an Afd.

As a volunteer receiving nothing in return, I improved Wikipedia significantly. If anything, User:150024 has reminded me that I have made great contributions to Wikipedia that I have not received credit for. Perhaps I could have become a page reviewer by now, but I haven't chosen to, simply because I don't want to dedicate that much time to Wikipedia—I have other hobbies that I care about more. I have been, perhaps, one of the greatest Wikipedia contributors, ranking in the top 1% of editors and creating articles that are far from inconsequential, yet I feel I am not being respected.

Please don't take this personally. I am frustrated by Wikipedia's choice to grant IP-block-exempt to relatively inexperienced editors. I may start a thread at the village pump. My standards for such a role are much higher: perhaps 100 articles created, with at least 20 at GA level, and at least 15 recommendations from fellow Wikipedians. Yet it is fairly easy to become an IP reviewer. During the editing process, User:150024 did some genuinely useful work, but also made an unjustifiable mistake by nearly blanking most of the Argentina at the 2017 World Games article. When I pointed this out, he became defensive and blanked my criticisms on his own page.

Regarding LLM use, I can also assure you that I have never used any AI tools before this past month, and have only used them for technical issues such as finding citations quickly. I welcome any editor to find any information that is not well backed by sources in articles mentioned by 150024. If I have made any errors, it is only because I am mostly tough on my own articles, and I do not care to judge the work of others too much—that would be an arrogance I do not have.

There are so many neutrality issues and so much non-fact-based editing on Wikipedia, yet time is wasted on tearing down the work of an above-average editor. Most of my 15 creations rank as above-average Wikipedia articles. However, I think that Wikipedia's culture is not one that I enjoy. So I will be ending my career as an editor once I take care of the articles I am working on. I am a conscientious person, and feel an obligation to the articles I have written, in making sure that the quality is as high as I can make it.

Thank you for your time and input. Orlando Davis (talk) 10:16, 27 November 2025 (UTC)[reply]

I have used Grammarly, but if that is AI, so is Microsoft Word's spell checker. It is very useful. I have seen many articles badly written by non-native English speakers; their articles might have been better if they had used it. Orlando Davis (talk) 11:35, 27 November 2025 (UTC)[reply]

Curious editors can see my unjustifiable mistake here, and are encouraged to review the subsequent conversation on my talk page here, as well as my [blanking] here. fifteen thousand two hundred twenty four (talk) 18:31, 27 November 2025 (UTC)[reply]

I’ve been thinking about this issue and would like to share my perspective. I was concerned by the way I was asked to provide a “confession,” as it felt as though I was being treated as if I had engaged in serious misconduct. In reality, my use of AI has been limited to citations, spelling, and grammar checks.

There is an important distinction between editors who use AI tools responsibly—as assistants while verifying all information—and editors who use AI to produce fabricated content or invented sources that have no place on Wikipedia. Necessary restrictions on LLM use are understandable, but these restrictions should not become a pretext for challenging good-faith editors who follow policy and use AI only as a tool.

Clearer guidance and training on appropriate conduct in the supervision of LLM use would help avoid future misunderstandings. If an editor suspects that AI was used, the next step should be to review the article and its sources to determine whether there is actually fabricated material. If the content is verifiable, well-sourced, and policy-compliant, the matter should be considered resolved.

It’s also worth remembering that all editors make minor mistakes in good faith. A few imperfect sentences in a long, well-sourced article are not comparable to an article built on fabricated information from start to finish. Orlando Davis (talk) 18:49, 27 November 2025 (UTC)[reply]

One issue that I forgot to discuss is that it doesn't make sense to review my drafts, as I have some articles that I started and decided that I didn't feel like finishing. I believe my draft space should be private, and I should only be judged on live articles, or ones that I have turned in to the Articles of Creation, but only by the page reviewers. Orlando Davis (talk) 01:37, 28 November 2025 (UTC)[reply]

it doesn't make sense to review my drafts – No drafts are listed. fifteen thousand two hundred twenty four (talk) 01:40, 28 November 2025 (UTC)[reply]

That's not true. The articles that were flagged by Niche were the Aluma restaurant draft, the Prince Alexander Literary Prize draft, and the Mastercooks article that had just gotten accepted by the articles of creation. Also, 15224 linked in her response an early draft of the static model aircraft that was way before I turned it in. I did not even use most of what is written in that draft. Orlando Davis (talk) 03:23, 28 November 2025 (UTC)[reply]

While 15224 provided the current version of the Static model aircraft page, she also linked to a very early draft version in her (these edits) link. That draft was only a very rough draft, and was heavily edited before submitting it to Articles for Creation and afterward.

Orlando Davis (talk) 09:13, 28 November 2025 (UTC)[reply]

If you want your drafts to be edited only by you, draft them in user space instead of draft space. Valereee (talk) 01:41, 28 November 2025 (UTC)[reply]

I would like to note that some of the concerns raised about my work seem to be influenced by earlier conversation during this (AfD) discussion. I understand that editors can encounter each other across different pages, but I want to clarify that some of the comments made about me in that AfD felt personal and, in my view, were not grounded in what I had actually written. I also understand that one of the editors (Polygnotus, who I only name for context) involved in that discussion was cautioned by several administrators at this Arbitration/Requests/Enforcement noticeboard regarding possible canvassing.

I am mentioning this only to provide context for why I may feel that certain assumptions about my editing practices are not being made in a neutral light. I am not alleging wrongdoing; I only want to explain why I believe some judgments about my drafts or sources may stem from earlier misunderstandings. Orlando Davis (talk) 09:50, 28 November 2025 (UTC)[reply]

To clarify, I was referring to blanking this page. I also want to clarify that I believe he has every right to blank his talk page. Orlando Davis (talk) 19:00, 27 November 2025 (UTC)[reply]

To quote: blanked my criticisms on his own page, and so I provided a link to that removal as well. (also, not a "he") fifteen thousand two hundred twenty four (talk) 20:04, 27 November 2025 (UTC)[reply]

If you start accusing people of AI use just for using Grammarly or some other constructive tool, you lose support in the fight against LLMs. This is dangerous because it undermines the encyclopedia's reliability and creates false beliefs. Orlando Davis (talk) 02:34, 30 November 2025 (UTC)[reply]

One of your edits had strings like :contentReference[oaicite:3]{index=3} where references should be. Describing that as "constructive use of Grammarly" is insultingly dishonest. jlwoodwa (talk) 03:50, 30 November 2025 (UTC)[reply]

If you start accusing people of AI use just for using Grammarly:

• https://www.grammarly.com – Page title, Grammarly: Free AI Writing Assistance
• https://www.grammarly.com/grammar-check – Write with confidence using Grammarly’s AI-powered grammar checker.
• https://www.grammarly.com/ai – We build AI that transforms how people communicate, making writing faster, clearer, and more impactful—no matter where you work or write.

I have checked the extension pages for chrome, firefox, and safari, as well as the android and apple app stores. Each one has AI in the title. Part of competent editing means understanding the tools you use.

You used an LLM, multiple editors noticed you used an LLM, there were issues found in some of the pages you edited while using an LLM, a report was opened at the LLM Noticeboard, I (an uninvolved editor) was requested to review edits, I closely reviewed four pages where references were added, source-text verifiability issues were found in each, I then compiled a list of other articles that may need further review. This is a logical sequence of events.

Please WP:DROPTHESTICK. fifteen thousand two hundred twenty four (talk) 03:51, 30 November 2025 (UTC)[reply]

Thank you.

What did I write that is not verifiable? Nothing. I'm not worried about it. I know that in the end, the conclusion will be that all of my live articles where accepted at the articles of creation for a reason. I do good work. I apologize if I'm posting too much. it's not a good habit.

Have a good night. Orlando Davis (talk) 04:04, 30 November 2025 (UTC)[reply]

@Orlando Davis, what in the world makes you think this is just about verifiability? This is also about wasting the time of other editors who have to check your work, because we know AI is often wrong. In my own prompts to AI, it often offers me unreliable sources for what it tells me. AI work has to be checked. Which means when Editor A is too lazy to do their own research and writing, Editors B, C, and D have to form a frickin' WikiProject to clean up after them. Using AI wastes the time of other editors.

The discussions you and I had at Talk:Mastercooks of Belgium and at Draft talk:Michael Katz (chef) over which of the dozens of sources for each supported notability represented hours of my work. I thought you were just a slow learner. But now I think you couldn't tell me which of the sources represented significant coverage because you didn't actually create the articles from the sources yourself. So you'd never read them. You were just as unfamiliar as I was with them. So when I asked you which three supported notability, you had no idea. So you kept asking me "What about this one?" And I kept going and reading and saying, "Nope, not sigcov". Over and over and over again. What a waste of my time.

For the record, you are not doing good work. The writing at those two articles is competent, but the sourcing is slipshod. AI is great at writing. It's often pretty crap at research. If you think the writing matters most, you've got it backward. Writing is easy to fix. Checking sources is tedious, thankless work. Valereee (talk) 10:29, 30 November 2025 (UTC)[reply]

Orlando, take a look at this page. This is the work you've caused for other editors because you used AI to create articles. Valereee (talk) 11:06, 30 November 2025 (UTC)[reply]

I understand your frustration.

If I use a Google search or use ChatGPT to search, what's the difference? As long as I read the source. And just to make sure ChatGPT isn't giving fake sources, I cut and paste the headline to verify that it is a source. And are you saying our difference of opinion on notability means I'm stupid and you are smart? In my AVSAB score (Navy), I got a perfect score in reading comprehension, though for privacy reasons I can't prove it.

You know, in the end, there are no rules on Wikipedia. See:

Wikipedia:Ignore all rules

And I was frustrated because I thought you were wasting my time. Because it's a subject that doesn't take long to infer it's notable. The ethical concern is fake articles; my ethics are sound. Everything is true. And to obsess about minor human mistakes, and if there are any, they are human mistakes, is unreasonable. And you didn't have to accept reviewing the article; you could have let a reviewer interested in reading and adding to that article take care of it.

And I was told to drop the stick, which I will, but I think so should you.

Thank you. I do appreciate your help on the articles. Orlando Davis (talk) 11:38, 30 November 2025 (UTC)[reply]

Wasting other editors' time to clean up after you falls clearly under wp:disruptive editing. Doing it intentionally also falls under wp:nothere. Both are blockable offenses. Agree to stop using AI altogether and I'll drop the stick, but you seem completely unrepentant for all the work you've caused other editors. Valereee (talk) 11:51, 30 November 2025 (UTC)[reply]

I understand your frustration.

I am not currently using AI tools and have committed to avoiding them. The articles under review were almost all written before I had heard of ChatGPT or anything similar.

I hope we can continue to work constructively on improving content.

Thank you. Orlando Davis (talk) 12:33, 30 November 2025 (UTC)[reply]

Orlando has since used an LLM to generate this talk page comment at Special:Diff/1324971229. This is the end of the road, they have been given more than enough ROPE but are NOTHERE. Can an admin handle or do I need to take this to ANI? NicheSports (talk) 16:16, 30 November 2025 (UTC)[reply]

What does IPBE have to do with anything, and what's an "IP reviewer"? jlwoodwa (talk) 04:15, 29 November 2025 (UTC)[reply]

• I filed at ANI due to today's continued LLM misuse and apparent retaliatory article tagging, see WP:ANI § Orlando Davis: LLM use, dishonesty, generally NOTHERE NicheSports (talk) 23:04, 30 November 2025 (UTC)[reply]

  FYI, Star Mississippi (talk · contribs · blocks · protections · deletions · page moves · rights · RfA) has blocked OD for one week for disruptive editing. sjones23 (talk - contributions) 00:53, 1 December 2025 (UTC)[reply]

Wayne Wesolowski, one of the pages cited at the cleanup tracker, needs to be inspected again: [54]. Should we just refresh the tracker when the issue is resolved? –LaundryPizza03 (dc̄) 23:16, 30 November 2025 (UTC)[reply]

Y Cleanup has been completed, see /2025-11-23 NatHaddan for reference.

NatHaddan (talk · contribs · deleted contribs · logs · filter log · block user · block log)
I don't know how much cleanup there is to do, but they've made nearly 700 edits which doesn't make me optimistic. They've had lots of warnings and denied AI use so I started an ANI thread here. Kowal2701 (talk) 23:46, 23 November 2025 (UTC)[reply]

not AI generated content please 🙏 NatHaddan (talk) 00:00, 24 November 2025 (UTC)[reply]

NatHaddan, please respond at WP:ANI Kowal2701 (talk) 00:07, 24 November 2025 (UTC)[reply]

Every source in Special:Diff/1320039380 is hallucinated, Special:Diff/1319055226 contains LLM-specific unprintable character sequences (0), Special:Diff/1323328155 hallucinates policy (WP:NODELAY). Looks like an LLM has been used extensively and without care. fifteen thousand two hundred twenty four (talk) 02:33, 24 November 2025 (UTC)[reply]

Tagged articles, one bad addition was already nuked before I did anything. I already lost my cool over Akpanta, Nigeria. AlphaBetaGamma (Talk/report any mistakes here) 06:24, 25 November 2025 (UTC)[reply]

Tracking subpage created, 8 articles need review. fifteen thousand two hundred twenty four (talk) 14:13, 25 November 2025 (UTC)[reply]

Are you using a script to list the diffs? If not, we could request one at WP:US/R? Kowal2701 (talk) 16:13, 26 November 2025 (UTC)[reply]

Somewhat, process and tools used currently are here. In short: contribution surveyor, regex, gadgets and simple userscripts.

No matter what the process will require a lot of manual review, but a userscript to make compiling the initial list more accessible wouldn't be unwelcome. fifteen thousand two hundred twenty four (talk) 19:06, 26 November 2025 (UTC)[reply]

Thanks, I made a request here Kowal2701 (talk) 20:59, 26 November 2025 (UTC)[reply]

Y Cleanup has been completed, see /2025-11-27 RobertFSomething for reference.

RobertFSomething (talk · contribs)
First time I've seen LLMs used for edit summaries. Some of the changes have already been reverted as they mostly seem to have stuck to articles for notable historical figures (Special:Diff/1303472803 for example), but I figured I should post this here anyways in case they start actively editing again. Also, should I create a tracking subpage for a given user before I post to the noticeboard, or should I only do that after there's agreement they've been using AI? Zygmeyer (talk) 14:59, 27 November 2025 (UTC)[reply]

Review and cleanup of flagged articles is completed, see subpage. fifteen thousand two hundred twenty four (talk) 00:27, 28 November 2025 (UTC)[reply]

Just to clarify for any other editors, tracking pages are an entirely optional convenience (but a welcome one).

should I create a tracking subpage ... – If you are certain that an LLM has been used unconstructively, then feel free to create a tracking page ahead of time, otherwise, ask. There's no downside to opening a report and asking for less-certain cases, and it could end up saving yourself and other editor's time if it turns out an LLM wasn't used. fifteen thousand two hundred twenty four (talk) 00:39, 28 November 2025 (UTC)[reply]

Cleanup efforts are ongoing and being tracked at /2025-11-28 ForayHistory, 19 pages need review.

ForayHistory (talk · contribs · deleted contribs · logs · filter log · block user · block log) has made almost 800 edits, mostly back in May and June, many of which seem to be AI-generated. Every article created by this user either is or has been tagged with {{AI-generated}} in part because of references that don't exist or don't support the article content. One of these articles was deleted in an AfD where the user used an LLM to comment. In addition to creating new articles, they've also done major rewrites of several existing articles. Some of the articles edited by this user have already been cleaned up or reverted, but others still need to be cleaned up. EvenTwist41 (talk) 18:09, 27 November 2025 (UTC)[reply]

I've created a subpage (using User:DVRTed/AINB-helper which had a few issues) and noted which edits have already been reverted. I included everything except pages with only minor edits that didn't seem AI-related (e.g. adding existing pages to "See also"), and around 150 edits to Surrealist Ball of 1972 made while it was at User:ForayHistory/sandbox and used to prepare edits to other articles. This was my first time doing this so let me know if I did anything wrong. EvenTwist41 (talk) 16:41, 28 November 2025 (UTC)[reply]

let me know if I did anything wrong – Nothing wrong, but I do have a suggestion, when creating a list consider selecting only entries where LLM involvement is reasonable plausible and potentially impactful. Personally I'd likely exclude edits like Special:Diff/1288074197, Special:Diff/1289823057 or Special:Diff/1287001153/1289718552, but this isn't an exact science. fifteen thousand two hundred twenty four (talk) 23:11, 28 November 2025 (UTC)[reply]

There's currently a village pump discussion related to AI cleanup at Wikipedia:Village pump (idea lab)#Mass-reverting AI serial abusers. Thebiguglyalien (talk) 🛸 04:59, 28 November 2025 (UTC)[reply]

Since I already have this notification here, I'll note that I've opened a different discussion about the public-facing side of Wikipedia's relationship with AI at Wikipedia:Village pump (idea lab)#Wikipedia as a human-written encyclopedia. Thebiguglyalien (talk) 🛸 02:58, 29 November 2025 (UTC)[reply]

Cleanup efforts are ongoing and being tracked at /2025-11-29 AKLKPhilo, 4 pages need review.

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

Many new articles with AI tells, as well as one filter hit for Markdown formatting. M-1 Studios has no fewer than five fictitious references:

• https://patch.com/michigan/troy/troy-based-scottish-society-ferndale-studio-to-produce-scottish-culture-videos
• https://metromodemedia.com/devnews/m1royaloak0108
• https://metromodemedia.com/devnews/m1studiosferndale0283
• https://s3.amazonaws.com/attachments.crainsdetroit.com/article/20120415/Free/304159996/Business_Diary_041512.pdf
• https://oaklandcounty115.com/2012/01/15/m-1-studios-makin-money-makin-movies-2012/

This is after they "verified statements against sources", so their claims about human review shouldn't be taken at face value. jlwoodwa (talk) 04:42, 29 November 2025 (UTC)[reply]

Additional consideration: They are the co-owner of M-1 studios Special:Diff/1307512086, they have been warned about COI editing but have elected to continue regardless, the mainspace M-1 Studios article was created by them after their draft was G11d. Considering this and their unreviewed LLM editing, this might be a WP:NOTHERE issue worth raising at another noticeboard in addition to the report here. fifteen thousand two hundred twenty four (talk) 04:57, 29 November 2025 (UTC)[reply]

This user has been warned for possible AI use many times on their talk page and has neither confirmed or denied it. They have made a lot of edits, which have promotional-tone issues even by AI standards. Gnomingstuff (talk) 17:22, 29 November 2025 (UTC)[reply]

 Cleanup has been requested.

Owais Al Qarni (talk · contribs · global contribs · logs · block log) is an editor, also very active at bn.wikipedia.org, who has been using LLMs on English WP so prolifically since late 2023 that they gained many advanced permissions while doing so, including autopatrolled. They have created hundreds of articles according to XTools. Based on the evidence below, I think we should assume that most of this user's substantive mainspace contributions since late 2023 were LLM-assisted.

While digging into this I also discovered significant copyright violations in their userspace, flagged it for MCE89, who found more in their mainspace contributions, and a CCI request was made and a CCI investigation opened. My intuition is that the pure copyvio is primarily in their userspace or pre-LLM mainspace contributions; mainspace copyvio since late 2023 is (while apparently common) likely LLM-related.

This editor had an intermediate level of English as of early 2023 [55][56][57] - also see [58]. By late 2023 or so they had started contributing via LLM both in and out of mainspace [59]. Some of their earlier LLM article creations were tagged by Gnomingstuff; OAQ then went to G's talk page and denied using LLMs [60]. I then independently found their edits via EF 1325 (hist · log), specifically, their recent creation of Risala Ahlus Sunnat wal Jamaat. That led to this exchange where they repeatedly prevaricated about their LLM usage. They eventually, after the CCI request was opened, admitted to having used LLMs but said they had stopped at some point [61]. I am sure their English has improved in the last few years, but I am highly skeptical of their claim about no longer using LLMs given the following, both of which this user created in the past month.

• Sirat-un-Noman: they created this entire article, but see the Reviews section in particular
• User:Owais_Al_Qarni/35 includes many chatbot responses about creating biographical article summaries, such as If you want, I can **continue translating the rest of his journalistic career** along with his editorial achievements and compile the full biography, covering **all his literary, translation, and journalistic contributions**.. Created at [62]

NicheSports (talk) 03:56, 30 November 2025 (UTC)[reply]

The sandbox is pretty unambiguous. I don't remember whether this was a case where I independently stumbled across multiple of their articles or only tagged the more obvious articles Gnomingstuff (talk) 06:09, 30 November 2025 (UTC)[reply]

This is a new one. I was looking at their current BN user page and noticed they have a Taliban flag there as well as a custom user box that Google translated to "This user supports the Taliban and is proud of it". Unrelated to any LLM stuff but I suppose people should be aware of this if they come back to editing NicheSports (talk) 06:04, 1 December 2025 (UTC)[reply]

Looking at the contributions, on 20 November they moved a mass of articles to "User:Owais Al Qarni/[Article Title]" with the edit summary Rewrite needed due to issues with LLM usage, which seems like a pretty flat admission in those cases I'd say. -- Cdjp1 (talk) 17:27, 1 December 2025 (UTC)[reply]