What happens to a public knowledge site when machine-written text starts blending into human work? For Wikipedia, the answer is stricter limits, tighter human oversight, and a visible attempt to protect the credibility of an encyclopedia that already lives with the risks of open editing.
The English-language community behind Wikipedia has explicitly barred contributors from using large language models to write or substantially rewrite articles. The change is narrow in wording but broad in significance. Wikipedia is built on citation, verification, and community review, yet it has also long carried a reputation for uneven reliability because anyone can edit it. That tension existed before generative AI arrived. AI tools intensify it by producing polished text that can sound authoritative even when it shifts meaning, blends facts, or introduces unsupported claims. That concern sits at the center of the new rule.
Wikipedia’s policy still allows two limited uses. Editors may use AI to refine prose they already wrote, and they may use it for translation, but both exceptions come with strict human responsibility. In each case, the editor must review the output carefully for accuracy. The guideline warns that LLMs “can go beyond what you ask of them and change the meaning of the text such that it is not supported by the sources cited.” On a platform where a citation is supposed to anchor every contested claim, that kind of drift is more than a style problem. It threatens the basic relationship between source and summary that keeps an encyclopedia usable.
The decision also lands in a wider debate over how much trust online communities are willing to place in AI-generated language. Wikipedia administrator Chaotic Enby described the rule as part of a larger cultural response, writing, “My genuine hope is that this can spark a broader change. Empower communities on other platforms, and see this become a grassroots movement of users deciding whether AI should be welcome in their communities, and to what extent.” The same administrator also called it a “pushback against enshittification and the forceful push of AI by so many companies in these last few years.”
Wikipedia’s caution did not emerge in a vacuum. Academic guidance has long treated Wikipedia as useful for orientation but weak as a final authority because entries can be outdated, manipulated, or wrong before editors catch the problem. Harvard’s library guidance notes that anyone can post material, and that false information can persist long enough to spread elsewhere. AI raises the stakes because it can multiply plausible-looking wording at scale, making weak sourcing harder to spot quickly, not easier.
Enforcement remains the awkward part. Detecting AI-generated prose is unreliable, and Wikipedia acknowledges that moderators can miss violations, especially on pages with less frequent moderation. Some human editors also write in ways that resemble machine output, which creates room for both false accusations and missed problems. The rule is therefore less a technological solution than a cultural signal: article text is expected to come from accountable human judgment, even when software is available.
The policy is also not universal across the whole Wikimedia world. Each language edition sets its own standards, and Spanish Wikipedia has fully banned the use of LLMs, including refinement and translation. That leaves Wikipedia looking less like a single verdict on AI and more like a federation of communities testing how much automation a shared knowledge project can absorb before trust starts to erode.
