Does Wikipedia make a company visible in AI?
Published on
Not directly, and not for most companies. Wikipedia carries significant weight in what AI models learn, but a Wikipedia company page is only allowed if the independent press has already written about it, and it guarantees no citations. In our measurements, Wikipedia pages retrieved by ChatGPT's search on artificial intelligence questions were cited in only 1.3% of cases. Wikipedia makes a company visible in AI when it is already known; it does not replace a page that answers your customer's question.
For what type of question? It all depends on the question. When the user mentions Wikipedia, AIs will read Wikipedia: on the question of this page, ChatGPT ended up citing the encyclopedia's rules themselves. On a product or market question, where no one points to Wikipedia, the encyclopedia is rarely the displayed source.
What Wikipedia weighs in AI training
Wikipedia is part of the training data for large language models, and it carries a weight superior to its size. In the article that introduced GPT-3 in 2020, OpenAI details its data mix: Wikipedia represented only 3 billion tokens and 3% of the mix, but the model traversed it 3.4 times during training, compared to 0.44 times for the filtered Common Crawl corpus, which is nevertheless a hundred times larger. Publishers no longer publish this level of detail for recent models.
AI players also pay to access Wikipedia's data. On January 15, 2026, Wikimedia Enterprise announced new partners, Amazon, Meta, Microsoft, Mistral AI, and Perplexity, joining Google among others. Wikipedia is therefore a source of knowledge for models and a source of data for AI search engines. But what the model has learned and what it cites when answering are two separate things.
Wikipedia in citations: figures that do not measure the same thing
The most widely cited figure comes from Profound, which analyzed ChatGPT citations, Google AI responses, and Perplexity between August 2024 and June 2025. Wikipedia represents 7.8% of all ChatGPT citations. The 47.9% figure, often presented as 'half of ChatGPT's citations', is actually Wikipedia's share among the top ten most cited domains only.
Semrush, which tracked ChatGPT's responses from July 14 to October 12, 2025, measures something else: the share of responses that cite Wikipedia at least once. It went from about 55% to less than 20% between August and September 2025. A share of all citations, a share among the top domains, and a share of responses yield numbers that range from one to seven; comparing them without their units leads to false conclusions.
What we measured on Wikipedia
For each domain, we measure the share of pages retrieved by an AI's search that end up being cited. With ChatGPT, on artificial intelligence questions asked mostly in English, the retrieved Wikipedia pages were cited in 1.3% of cases. Combined with Reddit and YouTube, they remain under 1.5% across all our corpuses, while other retrieved pages are cited in 10% to 15% of cases.
On the question of this page, asked on September 15, 2026, Wikipedia represented 23% then 27% of the results retrieved by ChatGPT, and 24% then 21% for Claude. ChatGPT did not cite any Wikipedia page the first time, then cited one in 17% of its sources: the article on Wikipedia's notability guidelines. Claude cited it in 7% of its sources, for company admissibility criteria. Gemini neither retrieved nor cited it. When the question is about Wikipedia, the encyclopedia rises in the citations; it is the question that summons it, not the company's notability.
Can a company have its own Wikipedia page?
The rules are strict. On Wikipedia in French, an article about a company is only admissible if significant, independent, and published works have been produced about it, spaced at least two years apart. Publications resulting from company communication, such as press releases, article reprints, or advertisements, do not count. On Wikipedia in English, no company is considered admissible automatically, regardless of its size.
Having your page written for a fee is regulated: paid contributions must disclose the identity of the client or employer, and promotional contributions remain strictly prohibited. For an SME without independent press coverage spread over at least two years, a Wikipedia page is therefore not an available lever: it is a consequence of notoriety, not a means of obtaining it.
What this changes for a business
Seeking to get onto Wikipedia to be visible in AI takes the problem backwards. What we observe with ChatGPT is that it writes in its query the names of the sources it knows and cites those as a priority. A name is built through mentions on independent sources that AIs consult; these same mentions are precisely what Wikipedia requires to admit a company.
The viable path is therefore to answer, on pages that AIs read and cite, the questions your customers are asking, and to be mentioned where these questions are addressed. A Wikipedia page can come later, if the press follows; it does not precede visibility.
What we do not know
Our measurements focus on the models' API, not on the interface of ChatGPT, Claude, or Gemini. We do not measure the effect, for a company, of having a Wikipedia page on its citations in AI; nor does any published study we have read measure it.
Sources
- Brown et al., OpenAI — "Language Models are Few-Shot Learners" (GPT-3), table 2.2, .
- Wikimedia Enterprise — partners announced for the 25th anniversary of Wikipedia, .
- Profound, Nick Lafferty — "AI platform citation patterns", .
- Semrush, Luke Harsel — domains most cited by AIs, .
- Wikipedia — Notability of companies, corporations and products, .
- Wikipedia — Paid contributions, .
- FaireDuBruit measurements, from September 4 to 16, 2026 — see "How we measured".