ChatGPT, Perplexity, Google: why each AI cites different sources

A page that ranks well on Google almost never shows up in Perplexity's answer to the same question. That's not a fluke — it's the rule. A 2026 study of 22.7 million citations found that ChatGPT and Perplexity share only 10.9% of their sources for a given question. A second study, in which Profound ran 100,000 prompts through both engines, lands on almost the same number: 11%. If your GEO strategy assumes one optimization pass covers every AI engine, these numbers say otherwise. Here's what they actually change.

For a small business or an agency working with a limited budget, this isn't an academic question. Should ChatGPT, Perplexity, and Google AI Overviews be treated as one "AI search" bucket, or as three distinct audiences that each need their own work? Most GEO guides so far — including some of our own — have leaned toward the first answer: the fundamentals are nearly the same everywhere, and work done for one engine carries over to the others. Recent data forces a more careful read of that claim, at least at the level of which specific sources get cited.

Why don't ChatGPT, Perplexity, and Google cite the same sites?

Each AI engine builds its answers from a different pool of sources, with its own preferences. According to a 2026 Wellows study covering 22.7 million citations, 1.15 million questions, and 441,946 websites across 27 markets between January and June 2026, 79.6% of sites cited by at least one AI engine show up on that engine alone. Only 0.31% of sites get cited by all five tracked engines (ChatGPT, Perplexity, Gemini, Google AI Overviews, Google AI Mode) for the same question.

The clearest number involves the ChatGPT/Perplexity pair — the two most-used engines outside Google. When ChatGPT cites a site for a question, Perplexity repeats that exact source only 10.9% of the time: 89.1% of ChatGPT's citations are invisible to Perplexity on the same question. Run it the other way and the figure climbs to 90.1%. A separate Profound study, which ran 100,000 identical prompts through both engines, confirms the same order of magnitude: just 11% of cited domains appear in both ChatGPT and Perplexity. A third study, from Passionfruit, measures 12% overlap across three engines. Three different methods, three results that agree with each other.

How much overlap is there really between two AI engines?

Overlap mostly comes down to who's owned by whom. The Wellows study measured agreement between every engine pair across 531,889 questions where all five engines cited at least one source. Google's own two products (AI Overviews and AI Mode) agree 24.5% of the time — by far the highest pairing, which makes sense since they draw on the same index. At the other end, ChatGPT and Gemini agree only 6.1% of the time, the lowest pairing measured. Across engines from different companies, average agreement is just 7.3%.

Why do engines favor such different types of sources?

Each engine has a clear favorite source family, and the gap is wide. Another Profound study, published in 2025, tracked 680 million citations over ten months (August 2024 to June 2025) and found that Wikipedia alone accounts for nearly half (47.9%) of ChatGPT's most-cited sources. Perplexity leans community-driven: Reddit weighs almost as much, at 46.7% of its most-cited sources, and content published within the last 30 days gets cited 3.2x more than older content. Google AI Overviews is more balanced, with Reddit at 21.0% and YouTube at 18.8%.

In practice, a page that gets strong engagement on Reddit stands a good chance of being picked up by Perplexity, and almost no chance of moving the needle with ChatGPT on that basis alone. A well-sourced Wikipedia entry works the other way: it feeds ChatGPT without doing much for Perplexity. Optimizing for a specific engine isn't just about writing good content — it's also about picking the right distribution channel.

The same Profound study surfaces another, less visible split: the type of domain being cited varies too. Across the full 680-million-citation dataset, 80.41% of citations point to a ".com" domain and only 11.29% to a ".org." In other words, institutional and nonprofit sites — often seen as a mark of credibility — carry far less weight than ordinary commercial sites in the overall citation pool. That's not necessarily reassuring for a business: it mostly means the competition for a spot in AI answers is happening mainly against other commercial sites, not against hard-to-dislodge institutional references.

Take a concrete example. A travel agency that publishes a destination guide that gets heavily discussed on Reddit might see that guide picked up by Perplexity when travelers ask related questions, while ChatGPT never notices it. If that same agency lands a well-sourced Wikipedia entry — through a press partnership or a mention in a reference article — the opposite happens: ChatGPT draws on it more readily, Perplexity much less. The content quality can be identical; it's the distribution channel that decides which engine notices it.

Does this gap also affect brand visibility?

Yes, and the spread across engines can be wide for the same brand. A 2026 Slate HQ study, based on more than 300,000 citations generated by six B2B SaaS brands and their competitors over 90 days (ChatGPT, Perplexity, Gemini, Claude, Google AI Overviews, Google AI Mode), found that Claude gives brands the most favorable share of citations — 9.1% of citations point to the brand itself or a direct competitor — followed by Perplexity at 6.8%. ChatGPT came in last on this measure, consistently, for every single one of the six brands studied.

Another finding from the same study: the brand and its known competitors together account for less than a quarter of all citations measured. The remaining three-quarters (76%) go to third-party sources — comparison sites, review platforms, forums, independent press coverage. A company that only watches its own name and its direct competitors' names in AI answers is missing most of the actual playing field.

That number lines up with a separate finding from Passionfruit: brands with profiles on third-party review platforms (G2, Capterra, Trustpilot, Yelp, depending on the category) are roughly 3x more likely to get cited than brands without one. A brand's AI visibility doesn't just play out on its own site — it also depends on whether it shows up in the third-party sources each engine already favors.

Does this mean you need to rethink your GEO strategy?

Yes, on one specific point: stop assuming a single checklist covers every engine. The GEO fundamentals are shared across all AI engines — clear content, machine-readable structured data, unblocked AI crawlers, claims that cite their sources. We cover that groundwork in our guide to getting cited by ChatGPT, Gemini, and Perplexity. But those fundamentals don't guarantee a citation on every engine — they only make your site citable, not cited everywhere.

In practice, that means three things for a small business or an agency:

1. Check each engine separately instead of stopping at ChatGPT because it's the best known. Our breakdown of where AI traffic is actually going in 2026 already shows how much traffic volume varies by engine; the citation data now shows sources vary just as much.
2. Adapt the channel, not just the content. If your market is active on Reddit, that channel matters for Perplexity. If your industry is well covered on Wikipedia or major reference sites, that matters more for ChatGPT.
3. Measure the real effect of a change, engine by engine, rather than trusting a general impression. That's the approach detailed in our piece on how to prove your GEO is working: an audit before, a targeted change, an audit after, on the engine that matters.

Do these studies actually agree with each other?

Yes — three independent methods land on the same order of magnitude, which strengthens confidence in the result. Wellows tracked 22.7 million citations over six months by querying engines directly. Profound ran 100,000 identical prompts through both engines and compared the domains they cited. Passionfruit cross-checked three engines with its own methodology. All three land on ChatGPT/Perplexity overlap between 10.9% and 12%. When independent studies, with different sample sizes and time periods, converge this closely, the signal is solid — far more so than if it rested on a single source.

That said, these studies have limits worth keeping in mind. They measure citations at a point in time, over a specific set of questions: overlap between engines can vary a lot by industry, query language, or country. The Wellows study covers 27 markets but is still 84% US-weighted; results for other markets could differ in the details, even if the underlying pattern — each engine has its own sources — has little reason to reverse. Another limit: these numbers move over time. Wellows already recorded a 33% rise in cross-engine agreement between January and June 2026. A number that looks fixed today could look different a year from now, and there's no way to say how fast that convergence will continue.

How do you find out which AI engines already cite you?

The only reliable way is to ask each engine your target questions directly and see who gets cited. Do it for ChatGPT, Perplexity, and Google AI Mode separately — don't stop at the first engine that cites you and assume the others will follow. For ongoing tracking without spending your days on it, an AI citation-tracking tool queries these engines on a regular schedule and flags changes as they happen.

Before that, the more structural question is whether your pages have what AI engines generally need in order to cite you at all. An AI visibility audit checks all 46 SEO and GEO criteria in one pass. That's the starting point before you even think about channel or content: if your pages block AI crawlers or lack machine-readable structured data, no distribution channel will make up for it — on any engine.

The GEO market has largely operated on the assumption that good optimization lifts every engine at once. The 2026 numbers tell a more nuanced story: the shared fundamentals exist, but at the level of which specific sources actually get cited, each engine remains largely its own ecosystem. Ignore that, and you end up optimizing for one engine while believing you've covered all three.