I asked Perplexity and ChatGPT the same question. Only one source overlapped.
Perplexity and ChatGPT can cite almost completely different sources for the exact same query. In my 2026-06-25 test for "best AEO tools 2026," Perplexity repeatedly cited a tiny set of aggregator-style "best tools" pages, while ChatGPT 5.5 thinking mode cited a much wider set, mostly tool websites and primary sources. Only one domain appeared in both: hubspot.com. If you want to get cited by AI, the engine you're optimizing for changes what kind of page you need to be.
I run AEO/GEO experiments to understand how AI engines choose sources, cite pages, and recommend tools. This post is part of that open research log: same query, repeated runs, visible citations recorded by hand.
What did I test?
I ran the exact same query, "best AEO tools 2026", in two AI search experiences:
| Engine | Runs | Mode |
|---|---|---|
| Perplexity | 15 | Free basic search |
| ChatGPT 5.5 | 5 | Thinking mode + web search via API |
Each run started in a fresh session. I recorded every visible cited source and grouped the citations by domain.
Research note
This is a small exploratory sample, not a benchmark: Perplexity n=15, ChatGPT n=5, run on 2026-06-25, visible citations only. The sample sizes are different for a practical reason: Perplexity free runs were easy to repeat, while ChatGPT thinking + web search API runs were paid and slower. Treat this as an early signal worth retesting, not a universal law.
Who this matters for
This matters most for small SaaS companies, agencies, consultants, and niche publishers trying to get mentioned in AI-generated tool recommendations. "Get cited by AI" sounds like one game. This test suggests it is already several games, depending on the engine and mode.
What sources did Perplexity cite?
Across 15 Perplexity runs, only 6 distinct sources appeared.
| Source | Runs cited | Type |
|---|---|---|
hubspot.com | ~15/15 | Big publisher / aggregator |
getairefs.com | ~15/15 | Small aggregator |
vismore | ~13/15 | Aggregator |
aiclicks, nicklafferty, inity | 1-2/15 | Noisy tail |
Perplexity's citation set was small and stable. A few domains showed up again and again, then a short tail appeared occasionally.
What sources did ChatGPT cite?
Across 5 ChatGPT 5.5 thinking runs, about 17 distinct sources appeared.
| Source | Runs cited | Type |
|---|---|---|
ahrefs, hubspot, otterly, peec, semrush, tryprofound, scrunch | 5/5 | Mostly tools' own sites |
writesonic | 4/5 | Tool site |
airops, arxiv | 3/5 | Tool site / primary paper |
athenahq, conductor, evertune, yext, brand24, ziptie, nightwatch, higoodie | 1-2/5 | Noisy tail |
ChatGPT thinking mode cast a much wider net. It still had a stable core, but the core looked different: tool pages, vendor pages, and the original arXiv GEO paper rather than mostly listicles.
Do Perplexity and ChatGPT cite the same sources?
Barely. Across both engines, only hubspot.com appeared in both citation sets.
Perplexity's repeated winners (getairefs.com and vismore) did not show up in ChatGPT's visible citations. And ChatGPT's repeated winners (tryprofound, peec, otterly, scrunch, and others) did not show up in Perplexity's.
That was the moment this stopped looking like "AI citations" as one category. It looked more like two separate citation economies sharing the same query.
What kind of pages does each engine seem to reward?
Perplexity favored aggregator and comparison pages: "best AEO tools" listicles, roundups, and pages that already package the answer as a ranked tool set.
ChatGPT thinking mode favored primary or near-primary sources: tools' own websites, official product pages, and the original arXiv GEO paper.
My current hypothesis:
| If you want to win... | Build or earn coverage on... |
|---|---|
| Perplexity | The best aggregator/comparison page for the query |
| ChatGPT thinking mode | A definitive primary page, product page, research page, or source that primary pages point to |
This is not final. It is exactly the kind of pattern I want to keep testing.
How many sources does each engine cite per answer?
Perplexity cited a small, focused set per answer: usually around 3-6 visible sources.
ChatGPT thinking mode cited much more broadly: about 11 visible sources per answer, while still returning to a stable core of roughly 8 repeated domains.
So the difference was not only which domains appeared. It was also the shape of the answer: Perplexity concentrated; ChatGPT explored.
Are AI citations stable across repeated runs?
Within each engine, yes, more than I expected.
The core sources repeated almost every run. The variation lived in the tail: one-off or low-frequency citations that appeared once or twice, then disappeared.
That matters because it suggests AI citation testing should separate two things:
- Core citations: repeated sources that likely define the engine's preferred answer set.
- Tail citations: occasional sources that may be easier to enter, but harder to rely on.
If you only run one prompt once, you cannot tell which is which.
Which engine should you optimize for?
"Get cited by AI" is not one game.
To win Perplexity, become or get included in the strongest aggregator/comparison page for the query.
To win ChatGPT thinking mode, become a primary source: a definitive product page, original research page, documentation page, or canonical guide the model can treat as authoritative.
The practical move is simple: pick the engine your buyers actually use, then build the page type that engine rewards. Optimizing blindly for "AI" risks optimizing for neither.
What this changes about AEO strategy
Traditional SEO taught people to ask, "Can I rank for this keyword?"
AEO needs a more specific question:
"For this exact query, in this exact engine and mode, what type of source gets cited repeatedly?"
For "best AEO tools 2026," Perplexity and ChatGPT answered that question differently. Perplexity seemed to trust packaged comparisons. ChatGPT thinking mode seemed to trust primary pages and authoritative source material.
That is the useful part. Not "Perplexity is better" or "ChatGPT is better." The useful part is knowing that the citation path changes by engine.
I wrote a broader playbook for this in How to Get Cited by ChatGPT and Perplexity, where I break down the earlier citation patterns that led to this follow-up test.
FAQ
Why do Perplexity and ChatGPT cite so differently?
My working explanation is retrieval strategy plus model depth. Perplexity is built around search-first answers, so it tends to pull compact, comparison-ready pages that already look like an answer. ChatGPT thinking mode appears to search more broadly, reason across more sources, and pull in primary or authoritative pages when it can.
Only one source overlapped. What does that mean?
It means citation visibility is engine-specific. Being citable in Perplexity does not automatically mean being citable in ChatGPT, and the reverse is also true. "AI visibility" needs to be measured by engine, query, and mode.
Which engine should I optimize for?
Start with your audience. If your buyers use Perplexity for tool discovery, aggregator coverage may matter more. If they ask ChatGPT for recommendations and comparisons, your own primary page and authoritative mentions may matter more. The wrong move is trying to optimize for a generic idea of "AI" without checking where your buyers actually search.
Is this enough data to make a rule?
No. It is enough to form a testable hypothesis. The next step is repeating the same method across more queries, categories, and engine modes.