In 2005, Wikipedia had teachers panicking. Visible errors, anonymous authors, copy-and-paste: the debate sounds familiar. Twenty years on, the arrival of generative AI is provoking the same reactions. What if the Wikipedia precedent helped us set better rules?
2005: the error that crystallised the fear
In May 2005, a false biography was added to the Wikipedia page of the American journalist John Seigenthaler. The text wrongly presented him as linked to the assassinations of John and Robert Kennedy. The error stayed online for several months, before Seigenthaler denounced it publicly in USA Today.
The case is serious and it left a lasting mark. On its own, it sums up the fear of seeing false information circulate with the appearance of truth. It is the same mechanism that fuels today’s concern about answers invented by an AI.
The comparison that complicated the story
In December 2005, the journal Nature published a blind comparison of 42 pairs of scientific articles, Wikipedia against the Encyclopaedia Britannica. The experts consulted found 162 errors, omissions or misleading statements for Wikipedia, against 123 for Britannica: about four errors per article against three.
This result did not make Wikipedia equivalent to an editorial reference. It reminded us of two more useful things: no source is infallible, and reliability has to be measured with a method. The study itself was challenged by Britannica over its protocol; Nature stood by the substance of its conclusions. That controversy is a lesson in method, not a certificate for blind trust.
The teaching problem is not finding an answer. It is knowing what it is worth, where it comes from and how to check it.
2007: not citing is not the same as not consulting
In 2007, the history department at Middlebury College banned citing Wikipedia in coursework and exams. The nuance is essential: it did not ban consulting it. After a group of students repeated the same error about a seventeenth-century Japanese revolt, the college wanted to stop an encyclopaedia from being used as a final source.
Jimmy Wales, co-founder of Wikipedia, then publicly endorsed the principle: you do not cite an encyclopaedia, you trace it back to its sources. On the same campus, some courses even asked students to contribute to Wikipedia. The tool was neither sanctified nor banned: it became a medium for learning to research, write and justify.
What this precedent changes for AI
AI is not Wikipedia. It produces text, can invent references and answers in very convincing language. The level of vigilance must therefore be high. But the teaching question remains comparable: which uses are acceptable, and which skills do we want to develop?
- Distinguish the uses. An AI can help rephrase a question, generate leads or self-assess. It must not become an unchecked source or the hidden author of an assignment.
- Require evidence of the process. Ask for sources, drafts, methodological choices and, where relevant, a declaration of AI use.
- Assess judgement. Have students compare a generated answer with reliable sources, spot an error, and explain why a piece of information is solid or fragile.
- Make the rules explicit. A clear instruction beats a vague ban: what is allowed, what must be flagged and what will be assessed.
Source literacy remains the central skill
Wikipedia eventually found its place: a frequent starting point, never sufficient proof on its own. AI should follow a similar logic, with safeguards suited to its capacity to produce plausible content.
Banning can sometimes meet an immediate need for fairness or academic integrity. But it does not remove the need to teach. Pupils, students and professionals will have access to these tools outside the classroom. Our responsibility is to teach them not to confuse a fluent answer with verified knowledge.
So the debate is not only: “AI allowed or banned?” It is above all: “How do we learn to verify, cross-check and stand behind our reasoning?”
Frequently asked questions about Wikipedia and AI in education
Why was Wikipedia challenged in education?
Wikipedia was challenged because its content can be edited by anyone and its reliability varies from article to article. The right teaching use is to treat it as an entry point, then trace back to reliable, citable sources.
Should generative AI be banned in class?
A ban can be justified in a specific assessment, notably to guarantee fairness. Outside that setting, the task is to spell out the permitted uses, ask for a declaration of use and assess how results have been verified.
How do you verify an answer produced by an AI?
Identify the verifiable claims, find primary or institutional sources, compare several independent references and flag what remains uncertain. An AI answer should not be cited as a source.
- Jim Giles, “Internet encyclopaedias go head to head”, Nature, 14 December 2005.
- Noam Cohen, “A History Department Bans Citing Wikipedia as a Research Source”, The New York Times, 21 February 2007.
- “Fatally Flawed”, Encyclopaedia Britannica, March 2006.
- Gilles Sahut, doctoral thesis (in French), University of Toulouse, 2015.
