English Wikipedia prohibits editors from using AI to write articles

English Wikipedia has updated its editorial rules: editors are now prohibited from creating and rewriting articles using language models. The update appeared in the encyclopedia's official guidance documents last week and represents one of the most concrete responses from a major public platform to the problem of mass AI-generated content. The rationale for the ban is stated directly: texts written by AI systematically violate several of Wikipedia's key principles.

These are requirements for a neutral point of view, verifiability of sources, and the absence of original research. Language models generate convincingly-sounding text that may not reflect the actual state of affairs, mix sources, or add details that were not in the original materials — all without any warnings. However, the new rules do not prohibit AI entirely.

Editors are permitted to use language models in two specific cases. The first is suggesting basic style and grammar edits, provided that the model does not add new content on its own. The second is translating articles from other language sections of Wikipedia into English.

However, even in this case, the editor must verify the result: AI translations cannot be published without verification. This distinction is indicative. Wikipedia is not attempting to completely isolate itself from LLM-based tools, but draws a clear boundary between technical assistance and the generation of meaning.

An editor can ask a model to correct commas or suggest smoother syntax — but not delegate responsibility for content to it. The change currently affects only the English section — the largest, with more than 6.8 million articles.

Other language versions of Wikipedia are managed by independent communities and may make their own decisions. Nevertheless, the English section traditionally sets the tone for the entire ecosystem, and similar rules may well appear in other versions. Over the past two years, some editors have experimented with ChatGPT and similar tools to accelerate work on a large volume of unprocessed topics.

The problem is not malice — but the systemic features of LLMs: models cannot clearly distinguish between what is reliably known and what merely sounds plausible. For an encyclopedia built on verifiable statements, this is a fundamental contradiction. Wikipedia's decision stands out against how most platforms respond to AI-generated content.

Many services either turn a blind eye to the origin of material or limit themselves to vague declarations about responsible use. Here — there is a concrete, operationalized prohibition with clearly defined exceptions. Such an approach is easier to check and enforce in practice.

The question of enforcement remains open. Authorship of text is difficult to verify automatically: AI content detectors have long been recognized as unreliable. The real mechanism is the community of editors, which notices characteristic patterns and brings issues to discussion.

How effectively this will work at the scale of millions of articles remains to be seen. For the broader discussion about AI in publishing, this step is important as a precedent. Wikipedia is one of the few major public resources with transparent governance: rules are adopted by the community, documented openly, and applied without commercial incentives.

If the experience proves successful, it could become a guide for other platforms that so far avoid firm formulations.