How AI Chooses Its Sources: 3,543 Citations, Analyzed
AI doesn't read your website. It reads Reddit, YouTube, Wikipedia, and a handful of independent blogs. We have the data: 3,543 citations across two engines and 60 questions.
How AI Chooses Its Sources
When someone asks ChatGPT or Google "what's the best tool for X," the AI doesn't make it up. It reads. It pulls from sources, summarizes them, and hands back an answer with a few links attached.
So which sources? We checked.
We ran 60 buyer-style questions about AI search, marketing, and visibility software through two engines — ChatGPT and Google AI Overviews — and logged every single source they cited. Not the answers. The citations. The actual URLs the models leaned on to decide what to say.
That's 3,543 citations across 717 distinct domains, collected in June 2026. Here's what the data says about how AI picks where to get its answers.
TL;DR
The short version: AI builds its answers out of other people's content. If you're only publishing on your own site, you're not in the conversation.
AI reads a small slice of the internet
There are 717 domains in this dataset. That sounds like a lot of variety. It isn't.
The top 10 domains account for 33% of every citation the two engines made. The long tail of 700+ sites splits the rest. When AI answers a question, it reaches for the same small set of trusted sources over and over.
Here are the publishers and communities AI pulled from most:
| Source | Citations | What it is |
|---|
| reddit.com | 265 | Community discussion |
|---|---|---|
| youtube.com | 235 | Video / reviews |
| en.wikipedia.org | 187 | Reference |
| techradar.com | 131 | Tech editorial |
| linkedin.com | 64 | Professional / social |
| zapier.com | 53 | How-to / blog |
| medium.com | 44 | Independent blogs |
| amplitude.com | 35 | Vendor blog (editorial) |
Notice what's missing: a company telling you how great its own product is. The sources AI trusts most are places where other people talk about products — forums, video reviews, encyclopedias, and independent tech press.
Vendor and tool websites collectively accounted for 833 citations (24% of the total). But they're distributed thinly across dozens of competing sites. No single product page comes close to the pull of Reddit or YouTube.
Blog posts beat product pages, by a lot
We tagged every cited URL by content type. The pattern is blunt:
| Content type | Citations | Share |
|---|
| Blog | 1,417 | 40% |
|---|---|---|
| Other | 1,123 | 32% |
| Product page | 515 | 15% |
| Docs | 208 | 6% |
| Comparison | 100 | 3% |
| Article | 91 | 3% |
| Review | 86 | 2% |
| Forum | 3 | <1% |
Blog and editorial content gets cited nearly three times more often than product pages. The thing companies spend the most effort on — their own product marketing — is one of the least cited content types.
This isn't a glitch. It's how the models are built. They're trained and grounded on text that explains, compares, and reviews. A product page that says "the leading AI-powered platform" gives a model nothing to quote. A blog post that explains how something actually works gives it a paragraph.
If you want to be cited, write the thing AI wants to read. That's not a brochure.
Third-party sources beat owned ones, 1,000 to 1
When we bucket citations by who controls the source, the gap is almost comical:
| Source type | Citations |
|---|
| Third-party (independent) | 2,278 |
|---|---|
| Vendor / competitor sites | 833 |
| Social | 238 |
| News | 180 |
| Reviews | 12 |
| Owned brand pages | 2 |
Two. Out of 3,543, exactly two citations pointed back to a brand's own owned page in the context where it was being recommended.
AI doesn't take your word for it. When it decides which tools to recommend, it cites the people talking about you, not the page where you talk about yourself. This is the uncomfortable part of AI visibility: the channel that decides whether you get recommended is one you don't own and can't directly edit.
Two engines, two appetites
ChatGPT and Google AI Overviews don't source the same way.
| Engine | Citations | Distinct domains |
|---|
| Google AI Overviews | 2,348 | 414 |
|---|---|---|
| ChatGPT | 1,195 | 376 |
Google AI Overviews cited roughly twice as many sources and pulled from a wider pool of domains. ChatGPT is more concentrated — it leans harder on fewer sources per answer.
The takeaway for anyone tracking AI search: you can't optimize for "AI" as one thing. The engines have different sourcing behavior, so being visible in one says nothing about the other. You have to measure each.
So what do you do with this?
The data points one direction. AI assembles answers from independent, third-party, editorial content — Reddit threads, YouTube reviews, Wikipedia entries, tech blogs, and roundups. Not from your homepage.
That changes the job:
The brands AI recommends aren't the ones with the best landing pages. They're the ones the rest of the internet talks about. The work is getting the internet talking.
FAQ
Where does AI get its sources from?
Across 3,543 citations from ChatGPT and Google AI Overviews, the most-cited sources were community and reference sites — Reddit, YouTube, Wikipedia, and tech editorial sites like TechRadar — not vendor product pages. The top 10 domains accounted for 33% of all citations.
Does AI cite a company's own website when recommending it?
Almost never. In our dataset of 3,543 citations, only 2 pointed to a brand's own owned page in the context of a recommendation. AI overwhelmingly cites independent, third-party sources rather than the brands themselves.
What type of content gets cited most by AI?
Blog and editorial content. Blog posts received 1,417 citations (40% of the total), nearly three times more than company product pages (515). Explainers, comparisons, and reviews get cited far more than marketing pages.
Do ChatGPT and Google AI Overviews use the same sources?
No. Google AI Overviews cited about twice as many sources (2,348 vs 1,195) and pulled from more domains (414 vs 376) than ChatGPT. The two engines source differently, so visibility in one doesn't predict visibility in the other.
How do I get my brand cited by AI?
Earn third-party coverage on the sources AI already reads — community threads, reviews, editorial roundups — and publish genuinely useful content worth quoting. Then measure your AI visibility per engine to see what's working.
Method: 60 buyer-intent questions about AI search, marketing, and visibility software, run through ChatGPT and Google AI Overviews. Every cited source URL was logged and classified by domain, content type, and source type. 3,543 citations across 717 domains. Data pulled June 2026.
Related Articles
AI visibility is the new SEO. Here's...
AI visibility is the new SEO. Here's the data to prove it
In 2010, if you weren't on page one of Google you were invisible. AI search is at that same inflection point, and the data from 91 brands proves it.
How we measure AI visibility
How we measure AI visibility: querying 10 LLMs with 100 prompts per category
10 models. 100 prompts per category. Scored for mentions, recommendations, citations, and sentiment. Here's exactly how the data works, including what it can't tell you.
What AI trusts
What AI trusts: a citation analysis across 3,000 AI responses
AI search engines cite sources. We tracked which ones across 3,000 responses. The hierarchy is clear, and most marketing content sits at the bottom of it.
