We split 153 tracked brands by whether they have a Wikipedia article
The sample is every answer Proofsource collected from October 4 to 29, 2026, for an English buyer question that doesn't name the brand, on ChatGPT, Perplexity, Claude and Google AI Overviews: 23,046 answers for 153 brands in 110 accounts.
We looked each brand up on English Wikipedia and counted it as having an article when the article's official website matched the brand's domain, or the exact title was a page about that company. 22 brands had one. 131 didn't.
Brands with an article were named in 52.5% of answers, the rest in 4.6%
The 22 were named in 1,423 of 2,712 answers, 52.5% (95% interval 50.6% to 54.3%). The 131 were named in 927 of 20,334, 4.6% (4.3% to 4.9%). The median brand with an article was named in 55% of its answers. The median brand without one, 1.2%.
The gap held on every engine. ChatGPT named the article brands in 56.9% of answers and the others in 6.1%. Claude: 50.7% and 4.4%.
Their own article was cited in 27 of the 1,423 answers that named them
If the article were doing the work, we'd expect the engines to read it. Mostly they didn't. In the answers that named one of the 22, that brand's own Wikipedia page was among the cited sources 27 times (1.9%):
- Claude: 20 of 332.
- Google AI Overviews: 6 of 353.
- ChatGPT: 1 of 427.
- Perplexity: 0 of 311.
Only Claude cited Wikipedia often, and it didn't name these brands more
Across all 247,363 citations, Wikipedia had 1,638. Claude accounted for 1,517 of them, 2.0% of its citations, and cited Wikipedia somewhere in 966 of its 5,685 answers. ChatGPT cited it 45 times in 46,413 citations, AI Overviews 40 times and Perplexity 36.
ChatGPT barely opened Wikipedia and named the article brands most often of the four. These are companies the models already know. An article is one sign of that, not the cause of it.
Wikipedia only admits brands that already have the coverage
Wikipedia's guideline for companies says an organization is presumed notable only with significant coverage in multiple reliable secondary sources that are independent of it, and that no company is notable by default. Its conflict of interest guideline strongly discourages editing an article about your own company, and paid editing must be disclosed.
So a brand can't get an article without first earning the in-depth independent coverage that models also learn from. Wikipedia does carry extra weight in training: OpenAI's GPT-3 paper gave it 3% of the training mix and passed over it 3.4 times, against 0.44 times for filtered Common Crawl. Current labs don't publish their mixes.
What we'd do about Wikipedia
- If you don't have an article, spend the effort on the independent coverage Wikipedia would require: reviews, analyst notes, trade press that covers you in depth. Engines read those directly.
- If you qualify, propose an article or edits on the talk page with your connection disclosed. Don't write it yourself.
- If you already have one, read it as Claude would. It is the engine most likely to quote it, so dated facts there end up in answers.
How Proofsource handles it
Proofsource records every page each engine cited in each answer, next to the brands it named. You can see whether any engine read your Wikipedia article, or your competitors', on the questions your buyers ask.
Where this could be wrong
This is a comparison of brands, not an experiment. The brands with articles are mostly large, well-known companies, and that, not the article, likely explains most of the gap. 3 of the 22 were named in under 30% of their answers, and 5 of the 131 without one were named in over 30%.
We matched articles by name and website, so a few brands with an article may sit in the wrong group. One brand is tracked in two accounts and counted twice.
Training data can't be seen from outside. An article may help a model learn a brand without ever being cited, and we only measured live citations. Our Claude answers come from the API with web search, and its app may cite differently.
The sample is 26 days, English questions asked from the US, in categories our users chose to track. We left out four Shopify review test stores.
