What ChatGPT, Gemini, and Perplexity Actually Cite
I checked whether the three engines actually agree on what to cite. They don't, and the pattern in what they do cite instead is the more useful finding.
I expected the three AI engines to agree on what to cite. Out of 566 identical questions, they converged on the same source only 20% of the time.
Everyone in AEO has an opinion about what makes AI engines cite a brand, but very few people are actually watching the same brand across the same questions, month after month, and writing down what happens. So I did.
This analysis draws on our monthly citation tracking across clients in fintech, conversational AI, clean energy, elder care, and consumer nutrition: 325 distinct real-world questions, run against three AI engines, producing 1,829 individual AI responses and nearly 10,000 citation links to work through.
ChatGPT, Gemini, and Perplexity rarely cite the same source for the same question. No single website, however large, accounts for more than a sliver of total citations. And within a brand's own site, it is almost never the homepage that gets cited, it's a specific article that already reads like the answer.
The Three Engines Rarely Cite the Same Source
Only 20.3% of the time do ChatGPT, Gemini, and Perplexity converge on the same brand or competitor page when answering the identical question. The rest of the time, either none of them cite a recognizable brand at all, or each cites someone different.
My first pass at this checked one thing: did each engine mention the client, yes or no, on the same question. By that measure, the three engines matched each other 89.2% of the time, which looked like real convergence.
But that measure has a blind spot. If none of the three engines mentions a brand, that counts as "agreement," even if ChatGPT is quoting one competitor, Gemini is quoting a second, and Perplexity is quoting a third. That's not agreement. That's three different answers that all happen to leave the brand out.
So I checked what actually happens: does the same brand or competitor page show up across multiple engines answering the identical question, or does each engine pull from somewhere different.
| What happened when all three engines answered the same question | Share of cases |
|---|---|
| None of the three cited any brand or competitor page | 48.8% |
| They cited brand or competitor pages, but a different one each | 30.9% |
| They converged on the same brand or competitor page | 20.3% |
Only one in five questions produced real convergence. Even narrowing to just the cases where none of the three mentioned a given client, the pattern holds: just 13% of those show the engines agreeing on a competitor instead. The other 87% are three separate answers that simply didn't include the client.
ChatGPT, Gemini, and Perplexity are not reading from the same shortlist. Getting cited by one of them tells you very little about whether you'll get cited by the other two, which makes a single-engine AEO strategy a bad bet from the start.
No Single Website Owns AI Citations
The ten most-cited domains across every client and question analyzed account for just 16% of all 9,928 citations recorded. AI engines pull from an enormous, thin spread of sources rather than a handful of authority sites.
I pulled every citation link and counted how many unique domains showed up: 2,717. I expected a handful of authority sites to dominate the way they used to dominate Google's top results. Instead, being cited looks less like winning a ranking battle against a few big competitors and more like getting picked out of a very long list.
When I sorted the ten most-cited domains into what they actually are, three kinds of sites showed up:
| Type of source | Share of the top 10 most-cited domains |
|---|---|
| Brand or competitor website | 53.5% |
| Social platform (LinkedIn, Reddit, YouTube) | 39.1% |
| Third-party blog | 7.4% |
Just over half is brand or competitor websites, companies talking about themselves on their own domain. That's expected: a brand's own site and its rivals' sites are always going to get cited when the AI is answering a question about that category. A small slice is third-party blogs with no obvious brand affiliation, content that happened to rank well on the topic. That's the uncontrollable long tail: thousands of other domains behave the same way, each showing up once or twice, impossible to individually target.
The genuinely useful finding is the rest of it: LinkedIn, Reddit, and YouTube. A brand or competitor website is mostly fixed, it already exists and either gets cited or doesn't. A third-party blog is out of your hands entirely. A social platform is different, it's a lever. You can post on LinkedIn, seed a Reddit thread, publish a YouTube explainer, and reasonably expect it to shift how often you show up in an AI answer, in a way that simply doesn't apply to the other two categories. Of everywhere a citation can come from, social presence is the one channel a brand can actually go build.
It's the Article, Not the Homepage
Only 8.7% of citations to a brand or competitor's own domain went to the homepage. 54.8% went to a blog, guide, or resource-style article, and the true content share is likely closer to two-thirds once ambiguous deep pages are accounted for.
Once I saw how scattered the citations were, I wanted to know if there was any pattern to which page within a brand's own site actually got picked. So I pulled every citation that landed on a brand or competitor's own domain, 2,053 of them, and sorted them by what kind of page they actually were.
| Page type cited | Share |
|---|---|
| Blog, guide, or resource article | 54.8% |
| Deep page, hard to auto-classify | 26.4% |
| Product, pricing, or comparison page | 10.1% |
| Homepage | 8.7% |
Homepages barely register. When I looked closer at that unclassified 26%, most of it turned out to be more of the same, articles like "top bank statement analysis software for businesses, explained," just without an obvious "/blog/" in the URL to catch automatically. Whatever brand strength or homepage authority means to these engines, it isn't what's getting quoted. The engines don't seem to reward the biggest company in the category, they reward whoever already wrote the specific article on the topic, which lines up with what Magnent sees across answer engine optimization work generally: a well-structured page beats a well-known brand.
Don't assume getting cited by one AI engine means anything about the others, they mostly pull from different sources even when answering the identical question. Social platforms are the one citation channel a brand can deliberately build, since brand and competitor websites are mostly fixed and third-party blogs are an uncontrollable long tail. And within your own site, stop leaning on the homepage: what gets cited is a specific article written on the topic, not a general company page.
Next time, I zoomed into one specific brand's numbers: how its visibility actually moved month to month, and what that says about how quickly AEO work does or doesn't pay off.
Frequently Asked Questions
Do ChatGPT, Gemini, and Perplexity cite the same sources for the same question? Rarely. Across 566 identical questions answered by all three engines, they converged on the same brand or competitor page only 20.3% of the time. In 48.8% of cases none of the three cited a recognizable brand or competitor domain at all, and in 30.9% each engine cited a different one.
Does getting cited by ChatGPT mean a brand will also get cited by Gemini or Perplexity? Not reliably. The data shows the three engines mostly pull from different sources even when answering the identical question, so getting cited by one tells you very little about whether the other two will follow.
Do AI engines cite company homepages or specific articles? Specific articles, far more often than homepages. Only 8.7% of citations to a brand or competitor's own domain went to the homepage, while 54.8% went to blog, guide, or resource-style content, and the true content share is likely closer to two-thirds once ambiguous deep pages are accounted for.
What kind of website gets cited most often by AI engines? No single kind dominates. Across nearly 10,000 citation links analyzed, the ten most-cited domains combined accounted for just 16% of all citations. Social platforms such as LinkedIn, Reddit, and YouTube made up roughly 39% of those top citations, alongside brand and competitor websites.
How much data does this analysis cover? The analysis covers 325 distinct real-world buyer questions, run against three AI engines monthly, producing 1,829 individual AI responses and 9,928 citation links.