Over two years English Wikipedia has built an apparatus for removing AI slop — WikiProject AI Cleanup, the G15 speedy-deletion criterion, and a 2026 content guideline that bars LLM-written article text.
2026-10-27 · by Moe Ameen
Over the past two years English Wikipedia has assembled a working apparatus for cleaning up AI-generated content. At its center is WikiProject AI Cleanup, an open group of volunteer editors whose founding members signed up on December 4, 2023. The project finds suspected AI-written text, images, and edits and checks them against Wikipedia's content policies — pulling unsourced or likely inaccurate claims first, and tracking contributors who appear not to understand the limits of the tools they are using.
The sharpest instrument is G15, a speedy-deletion criterion added to Wikipedia's deletion policy on August 4, 2025, for pages that are unambiguously LLM-generated and unreviewed. G15 lets an administrator remove such a page immediately, bypassing the usual seven-day deletion discussion. The signs it keys on are deliberately narrow and objective: leftover chatbot phrasing aimed at the user (a page that opens "Here is your Wikipedia article on…"), and fabricated or implausible citations pointing to sources that do not support — or do not exist to support — the text.
In March 2026 the community went further. A request for comment that closed on March 20, 2026 approved WP:LLM, a content guideline barring the use of LLMs to generate or rewrite article content, by a wide margin. Narrow exceptions survive: basic copyediting of an editor's own writing as long as the tool adds nothing new, and help translating non-English articles when the editor can actually verify the result. The guideline governs English Wikipedia only; each language edition sets its own rules.
Notably, Wikipedia's approach is cautious about detection itself. Editors are warned not to treat writing style — or a score from an AI-detection tool — as sufficient grounds for removal, because some people naturally write in ways that read as machine-generated and detectors misfire. What counts instead is whether the text holds up against sourcing rules and what the editor's track record shows. Text predating ChatGPT's November 30, 2022 release is treated as very unlikely to be AI-generated. Alongside G15, editors use the {{AI-generated}} tag, a "signs of AI writing" guide, and presumptive-removal processes for the many cases that are not clear-cut enough for speedy deletion.
Read Wikipedia's deletion criteria as a checklist of what gives AI writing away: leftover chatbot phrasing, generic machine voice, and citations that do not check out. The lesson for creators is not that AI content is banned everywhere — it is that *unreviewed* AI content is what gets caught. [Kompozy](/) is built to produce the opposite. Every piece runs through a [Persona Brief](/glossary/persona-brief) that governs voice, plus banned-word and AI-catchphrase filters that strip the exact ["AI tells"](/glossary/ai-tells) Wikipedia removes on, so a blog post or newsletter reads like you wrote it rather than like a chatbot handed it back. A per-post review pipeline keeps a human in the loop before anything publishes — the single discipline G15 exists to enforce.
Just as important, Kompozy is for the channels you own, not the encyclopedia. It generates substantive [blog articles, newsletters, and social posts](/glossary/output-buckets) and ships them across the eight social platforms plus blog and email on [Autopilot](/glossary/autopilot) — owned surfaces where consistent, reviewed, voice-locked content builds the real-world notability that eventually earns a neutral Wikipedia mention instead of a speedy deletion. Because Wikipedia is one of the sources AI answer engines lean on most, that credibility compounds: the cleaner and more consistent your owned footprint, the better the picture an assistant assembles when someone asks about you. And if detectors are your specific worry, Wikipedia's own stance — a detector score alone proves nothing — is the right one; our guide on [how AI content detection actually works](/guides/ai-content-detection) explains why sourcing and a human editor matter more than any scanner.
On English Wikipedia, using an LLM to generate or rewrite article content is barred by a content guideline (WP:LLM) adopted in a March 2026 request for comment, with narrow exceptions for basic copyediting of your own writing and for help translating non-English articles. Separately, pages that are unambiguously LLM-generated and unreviewed can be speedy-deleted under the G15 criterion added in August 2025. The rules govern English Wikipedia; other language editions set their own.
It is an open group of English Wikipedia volunteers, with founding members signed up on December 4, 2023, that finds and reviews suspected AI-generated text, images, and edits. It checks content against Wikipedia's policies — removing unsourced or likely inaccurate claims — and maintains tools such as the {{AI-generated}} tag and a "signs of AI writing" guide.
Not on their own. Wikipedia's guidance explicitly warns against treating writing style or an AI-detector score as sufficient grounds for action, because some people naturally write in ways that read as machine-generated and detectors misfire. What matters is whether the text holds up against sourcing rules and what the editor's track record shows.
The pattern Wikipedia removes is unreviewed, unsourced AI output with telltale phrasing and fabricated citations — not AI assistance in general. For your own channels, the durable approach is human-reviewed, voice-consistent content. Kompozy governs voice with a Persona Brief, filters out AI-tell phrasing, and keeps a per-post review step before publishing across the eight social platforms plus blog and email.