Wikipedia:AI noticeboard/Keizers
.svg){kind=icon}
.svg){kind=icon}
.svg){kind=icon}
Discussion: Keizers
[edit]Keizers (talk · contribs · block log · filter log · page creations · central auth · count)
18 October 2026
[edit]Rapid article creation (11 large articles in the last couple days, ), Special:AbuseLog/39325101 (Nov 2024) had a chatgpt param, Siilawy (11 July) had AIREDCAT, AITABLE, reads strongly like AI but few signs otherwise + Pangram positive . They were warned about LLM use on 12 July , to which they responded with an AI-generated comment which hallucinated a policy (WP:AIASSIST). I've warned and notified them Kowal2701 (talk, contribs) 16:27, 18 October 2026 (UTC)
- (copying over from User talk:Keizers#October 2026, courtesy ping @Keizers)
Kowal2701 (talk, contribs) 18:36, 18 October 2026 (UTC)Thanks for the message, fair enough. Three separate things here, so I’ll take them in turn.
1. The 2024 edit with a URL in the <ref> that had a ? showing access via ChatGPT. This was very early on and would have been a source I’d found through ChatGPT and pasted with that tracking parameter still on the URL. Just very early use of LLMs for research.
2. Siilawy was July 2026. I flagged it as AI-assisted in the edit summary myself. I was told not to do that, and apparently I had an LLM help me formulate the reply and it is indeed ironic that it hallucinated a policy. Like a lot of people, I was experimenting with how LLMs could help with everyday tasks, and that was the phase where I made some mistakes. In any case, after the warning, I did change how I worked and no longer cut and pasted new articles from LLM output.
3. For the recent articles in question, I used Claude to find sources, draft text and prepare the wikicode, then checked every claim against its source, then cut, added, and reworded all the resulting content myself. I thought that put me on the right side of the line, i.e. that the issue was above all factual accuracy, as well as style and tone. I’ve now read NOLLM properly and it’s about the drafting itself. My mistake.
Going forward I’ll write article text myself and use LLMs only for research and the basic copyediting NOLLM allows. If there are articles you want looked at, point me to them and I’ll go through them myself.Keizers (talk) 18:33, 18 October 2026 (UTC)
- Keizers, it doesn't matter if you LLM for research purposes, rewording or to find sources, if you used an LLM, you used an LLM, and it is strictly prohibited, that content sounds like an LLM too. Emmanbotan🌹 (talk) 09:24, 19 October 2026 (UTC)
- (editors are permitted to use LLMs to help with research and correcting spelling/grammatical errors, WP:LLMRESP has advice on this) Kowal2701 (talk, contribs) 13:20, 19 October 2026 (UTC)
- My reply acknowledges your point about the use of LLMs, and my reply is in my own words and structure.Keizers (talk) 16:26, 19 October 2026 (UTC)
- @Emmanbotan, not to rehash the point, but WP:EPAI and WP:NOLLM don't ban using AI for research. AI is banned for editing activities such as modifying or writing actual encyclopedic content (outside a few narrow exceptions) and making talk page comments. Minor grammatical corrections of one's own work is one of the two exceptions to the ban on LLM for encyclopedic content. As for research, it may raise a lot of questions if you used a chatbot to find sources for you, but technically using AI for research is not disallowed by wikipedia policy or guideline (and if you use Google, you might even be doing it unknowingly, since Gemini now has at least some input as to which sources are shown first when you search up something).
- @Keizers, just as an FYI, if you use popular browser extensions such as Grammarly or an online translator, many of these may use AI now, so be careful of that. Other than that, I don't really have much to add, since presumably you now know that LLM usage is generally prohibited for editing encyclopedic content. Accessedgrant (Epicgenius mobile alt) (talk) 03:16, 20 October 2026 (UTC)
- Keizers, it doesn't matter if you LLM for research purposes, rewording or to find sources, if you used an LLM, you used an LLM, and it is strictly prohibited, that content sounds like an LLM too. Emmanbotan🌹 (talk) 09:24, 19 October 2026 (UTC)
@Keizers: can you give us a start date for LLM use? It would really help with cleanup efforts. You said the 2024 edit was "very early on"--can you pin it down a bit more? Thanks. M kuhner (talk) 17:18, 26 October 2026 (UTC)
- Sure!
- The 2024 issue is simply a URL that was opened from ChatGPT. That was not use of Chatgpt to write anything, simply using it to help with research.
- The relevant activity is having an LLM do research and generate an initial draft of an article in Wikicode, then reworked multiple times with prompts from me, and which is then manually checked by me for factual correctness, reworked, sentences rewritten etc - again all manually. This applies only to articles from 30 September through 18 October 2026.
- After that date, I stopped having it generate draft text, and I work from a spreadsheet where I have all the facts and sources in order (including quote fields within the <ref> to ensure the claims do indeed match the source), and then write the text entirely manually.
- Keizers (talk) 22:30, 26 October 2026 (UTC)
- Thank you! I ask that question multiple times a day and I hardly ever get an answer, so I really appreciate your cooperation. I will make a cleanup tracker for those dates only and we can check them out. M kuhner (talk) 22:41, 26 October 2026 (UTC)
- What about the July 2026 one mentioned above? M kuhner (talk) 22:45, 26 October 2026 (UTC)
- Oh, yes also that one. That was a one-off use in a very rudimentary way.Keizers (talk) 18:15, 27 October 2026 (UTC)
- @Keizers, normally we delete AI-generated articles (WP:LLMPRV), but we can draftify some if you want to work on them? It's best to either stubbify and review it for WP:V or completely rewrite the article from the sources. It's up to you, if you're interested just let us know which ones, thanks Kowal2701 (talk, contribs) 13:00, 29 October 2026 (UTC)
- OK so just to understand this would only be for newly created articles, and not for the numerous small edits that I also see in the list. In the list, I see the new articles are still published. If ok with your team I can go through them and review and edit them as necessary and mark them edited and reviewed. I'm not sure if that's kosher for you guys and if so how I would use your tracker.Keizers (talk) 22:17, 30 October 2026 (UTC)
- And I should clarify futher: the new articles are all notable and valuable, and things I would wish to re-create according to policy, even if they were deleted. If you check, you will see they are not AI slop, but rather carefully crafted, despite having (wrongly) used AI to generate a first draft of the text.Keizers (talk) Keizers (talk) 22:19, 30 October 2026 (UTC)
Cleanup
[edit]- User: Keizers (talk · contribs · block log · filter log · page creations · central auth · count)
- Start date: 2026-09-29
- End date: 2026-10-19
Tracking list
[edit] | Notes
|
Each individual row below tracks an article's status using a template with three parameters: {{AIC article row|a page title, |a status, |and optional notes }}
Available statuses are: 
to do/requested, 
ongoing, 
tagged, 
unnecessary, and 
completed.
Please update and add entries as needed.
