ChatGPT and Perplexity Almost Never Cite the Same Source. Here's What That Means If You're Not Wikipedia.

Ten questions, three assistants, and only three businesses everyone could agree on

The Happy Cat, a Manchester AI-visibility agency, ran a small experiment in August 2026 that I keep coming back to. They took ten buyer-style questions, the kind an actual customer would type when they're trying to hire somebody, and ran each one through ChatGPT, Perplexity, and Gemini. Same questions, word for word, three assistants, thirty answers total.

Those thirty answers cited 104 different sources. Only 11 of those sources got cited more than once. And out of everything all three assistants pulled from, only 3 businesses showed up as a recommendation across all three.

Ten questions is a small sample: one agency, illustrative rather than statistical. But it's also a clean, human-sized picture of something the larger studies confirm at scale: three assistants, three different ideas about where a good answer comes from.

The number behind the anecdote

Wellows tracked this properly, 22.7 million citations across multiple engines, and the picture holds up. Only 0.31% of all the sources they saw showed up on all five engines they tracked. Nearly four out of five sources, 79.6%, showed up on exactly one engine and nowhere else. And here's the one that stopped me: for the same question, Perplexity never touched 89.1% of the sites ChatGPT cited. Not "rarely." Never, for the overwhelming majority of them.

Wellows also found the engines are converging a little. Pairwise agreement between engines went from 8.49% in January 2026 to 11.25% by June. That's real movement. But read those numbers again. Eleven percent agreement is still telling you that roughly nine times out of ten, these engines are picking different sources for the same question. A trend line pointing the right direction is not the same thing as the problem being solved. It's still a bad bet to plan around one engine's behavior as if it speaks for all of them.

It's not just different. It's not even stable.

If I were running a business that depended on this, the instability would bother me more than the disagreement. Passionfruit's 2026 research on AI search found that 68% of queries that earned a citation in one month did not earn one the following month. Gone. Not demoted, gone. And only 7% of queries kept a visible citation for four straight months.

So even setting the cross-engine mess aside entirely, even if you only cared about one single assistant, getting cited once tells you almost nothing about getting cited next month. This isn't the same problem as three engines disagreeing with each other. It's the same engine disagreeing with itself over time. Both are true, and both are working against you at once.

I already dug into part of why the source pool skews the way it does in an earlier post, ChatGPT and Perplexity sharing only about 11% of their cited domains overall, with Wikipedia and Reddit dominating the broad top-10 questions. I'm not going to re-run that math here. Go read that one if you want the source-type breakdown. What I want to focus on is what all this means for you specifically, because you are not Wikipedia and you are not trying to be.

You were never actually competing for "how often should you get an AC tune-up"

Here's the reframe that matters. Nobody's local HVAC company was ever going to out-cite Wikipedia on the broad, generic version of a question. That fight was lost before it started, and it's not the fight you need to win.

The question you actually have a shot at is narrower and it's the one only you can answer correctly: your name, your service, your city. "Who does emergency AC repair in your city." That's the question where your own site is a real contender, not a footnote next to an encyclopedia.

Getting picked for that question is still not guaranteed by anything on your end, and I want to be straight with you about where the line actually sits. Our AI Crawler Access module checks whether five specific bots, GPTBot, ClaudeBot, PerplexityBot, OAI-SearchBot, and Bingbot, can actually reach your site through robots.txt. A clean 15 out of 15 there means OpenAI's, Anthropic's, Perplexity's, and Bing's crawlers can get in the door. It says nothing about Google, and nothing about Gemini. Both run on entirely different mechanisms our crawler check doesn't touch at all. And it says nothing about whether any of them actually pick you once they're in. Getting past the bouncer isn't the same as getting asked to dance.

That's why another 68 of our 100 points exist. Structured Data is worth 28, Entity & Trust is worth 20, and Content Structure, worth another 20, is honestly the most relevant module to everything I just described, because it rewards the exact thing an engine is doing when it quotes you: pulling one clean, direct, extractable answer out of your page instead of the whole thing. Question headings answered plainly underneath them. A real FAQ section. Those answers laid out as lists or tables instead of buried in one long wall of text. That's 68 of the 100 points asking whether you're worth quoting once someone arrives. The 15 crawler-access points are only asking whether they can get in the door at all.

Our free Grade Check makes that gap between getting in the door and getting picked concrete, worth knowing about even if you never run it. It's a hard gate. If any one of those five bots is fully blocked in your robots.txt, the scan doesn't bother scoring the rest of your site. It's an automatic F, zero out of a hundred, on the spot. Passing that gate doesn't earn you points elsewhere. It just means you're allowed to compete for them.

Try it on your own name

Before you take my word or anybody's word for any of this, go ask ChatGPT and Perplexity the identical question about your own business. Word for word, both places. See whether either one actually names you and links your site, or whether you get routed to a directory listing instead while the other assistant answers the same question completely differently.

Go see where you actually stand

The free Grade Check on aeocheck.net scans your homepage against the same scoring engine behind our paid audits, the one that gave our own site a 97 when we ran it on ourselves. It hands you a letter grade, a score out of 100, and how that compares to the median for the other sites of your type we've checked. It won't hand you the module breakdown. That level of detail, what's actually costing you points and how to fix it, is what the paid report is for. But the free check will tell you, honestly and in under a minute, whether you've even cleared the gate these engines require before any of them will consider you at all.