Wikipedia's enforceable AI rule is not the page people cite

The setup

An editor deciding whether to let a large language model touch an article draft has to find the right page first, and the volunteer encyclopedia splits the guidance across two. Wikipedia:Large language models labels itself, in its own page banner, 'an information page' that 'is not a Wikipedia policy or guideline' and 'may reflect varying levels of consensus.' A hatnote on that same page redirects readers to Wikipedia:Writing articles with large language models, which the site's own page metadata categorizes among 'Wikipedia content guidelines' and which states of itself: 'editors should generally follow it, though exceptions may apply.'

What the documents show

The guideline page states its rule in a nutshell: 'don't use large language models such as AI chatbots to generate or rewrite article content,' with two carved-out exceptions. Editors may use an LLM to suggest corrections to their own writing, 'limited to spelling, punctuation, capitalisation, grammar, and other simple mistakes,' after human review, and 'editors must ensure that any text copyedited in this manner is still supported by the sources cited.' They may also use an LLM for translation, following separate linked guidance. Outside those two uses, the page states persistent use 'is considered disruptive and will result in an editor being blocked,' and a pattern of edits linked to undisclosed AI use 'may be presumptively removed.' The companion information page adds that editors 'should' name LLM use in an edit summary 'even if their terms of service do not require it,' framed as a recommendation rather than the guideline's own enforceable language.

The friction

The two-page structure is itself a disclosed source of confusion: the shortcut 'WP:LLMS' resolves to the non-binding information page, which then hatnotes readers over to the actual content guideline, so a reader who stops at the first page can walk away thinking the more detailed hallucination discussion there carries the same weight as the guideline's flat prohibition. The information page's own 'see also' list names a separate item, 'Wikipedia:LLM use disclosure,' explicitly as 'an essay about the idea of requiring LLM use disclosure' — confirming that mandatory disclosure is, as of these pages, a proposal under discussion rather than settled rule.

What changed in the work

For someone drafting or editing with an LLM's help, the enforceable line documented here is narrower than 'be transparent about AI use': it is closer to 'don't generate or rewrite article prose with one, except narrow, human-reviewed copyediting or guided translation.' The richer material on hallucination risk and disclosure norms is real and useful, but it carries the lower authority the page assigns itself.

• Is the page you're reading labeled a guideline, or an information page describing one?
• Does a copyedit suggested by an LLM still match what the cited sources actually say?
• Is a disclosure norm you've heard about an enforceable rule, or a proposal still under discussion?

Two pages with different authority, on the same topic, is itself the finding here: reading only the more detailed one risks mistaking explanation for enforceable rule.