What ChatGPT, Perplexity, Gemini & Claude actually cite

I asked four AI engines the same two questions and logged every source each one cited. They recommended the same handful of brands - and then backed those recommendations with almost entirely different evidence.

The setupHow I ran it.

On 30 September 2026 I sent the same two prompts to four engines, each with live web search enabled, via their APIs:

• Q1 (commercial): "What are the best answer engine optimization (AEO) agencies or consultants?"
• Q2 (informational): "How can a company get its brand cited by AI search engines like ChatGPT and Perplexity?"

The models were ChatGPT (gpt-5.6-sol), Perplexity (sonar-pro), Gemini (gemini-3.8-flash) and Claude (claude-sonnet-5). For each answer I captured the full list of cited source URLs, then reduced them to domains. It's a small sample - two questions, one run, one day - so treat it as a snapshot, not a census. But the pattern was stark enough to be worth showing.

Finding 1They agree on the brands.

On the commercial question, the four engines converged hard. Three names showed up in every single answer, and a second tier appeared in two of the four:

If you only looked at the recommendations, you'd conclude these engines basically agree. That conclusion falls apart the moment you look at how they got there.

Finding 2They disagree on the sources - almost completely.

Across both questions the four engines cited 69 unique domains. Of those, 64 (93%) appeared in only one engine's answers. Just five domains were cited by two engines - firstpagesage.com, minuttia.com, nogood.io, reddit.com, yesoptimist.com - and not a single domain was cited by three or more.

On the "how to get cited" question specifically, the overlap was even starker: the four engines shared zero sources. Four answers to the same question, no common ground in the citations at all.

The engines reached the same conclusions from four almost completely separate slices of the web.

Finding 3Each engine has a source personality.

The counts and the character of the sources were very different:

ChatGPT was the most conservative - when asked how to get cited, it went almost entirely to primary sources (OpenAI's own crawler docs, Bing Webmaster Tools, Google's structured-data docs). Perplexity was the opposite: broad, current, and happy to cite newswire and Reddit. Gemini and Claude both leaned on small, niche industry blogs - but with virtually no overlap between their picks.

Finding 4The fan-out is real, and it compounds.

Two of the engines exposed their query fan-out - the multiple sub-searches they run behind a single question. For Q1, ChatGPT fired eight sub-queries (including targeted site: searches against specific agencies); Gemini ran two. Claude's own answer described the mechanism plainly: when a brand shows up across several of those sub-queries, a process it called Reciprocal Rank Fusion compounds that repetition into a citation.

That's the real lever. You're not trying to rank once - you're trying to appear across many related sub-queries so the fusion step keeps surfacing you.

What it meansPractical takeaways.

• Don't optimize for one engine. With 93% of sources unique to a single engine, winning ChatGPT tells you almost nothing about Perplexity or Claude.
• Diversify where you appear. Own-site content, SEO media coverage, PR/newswire, Reddit and niche industry blogs each feed different engines. You need presence across the whole spread.
• Reddit and listicles matter. They were among the very few sources multiple engines shared - and the "best agencies" roundups clearly shaped every engine's recommendations.
• Consensus wins the recommendation. The brands named by all four are the ones repeated across many independent roundups. Being everywhere is what gets you into the answer.
• Aim to appear across sub-queries, not just to rank for the head term - that's what the fan-out rewards.