ChatGPT and Google AI Overviews Cite Different Internets

YouTube is the most-cited domain in Google AI Overviews, with 51.9 million citations. In ChatGPT it ranks nineteenth, with 211,000.

Facebook, Instagram and TikTok all sit in Google’s top ten. None of them appear in ChatGPT’s top thirty. Reuters, The Guardian, AP News and The Washington Post appear in ChatGPT’s top thirty and not in Google’s.

One domain is in the top two of both. Reddit.

These are DataForSEO’s numbers from Search Insider: 2026 Edition, built from 137.9 million AI Overview answers and 27.5 million ChatGPT answers collected between mid-2025 and September 2026.

The two top tens, side by side

The two lists are describing different versions of the web. Google’s is social, video and community. ChatGPT’s is encyclopaedias, institutions and reference works.

Don’t compare the raw volumes across the columns. They are different measures from different sized corpora, one counting citations and the other counting answers that cite a domain. The ordering is the finding, not the size.

Why I think they diverge

The report doesn’t explain it, so this is my reading.

Google AI Overviews are grounded in Google’s own index and lean on signals Google already has. That’s a reasonable explanation for YouTube sitting at number one, given who owns it, and for the published finding that organic ranking predicts AI Overview citation reasonably well.

ChatGPT works from training data plus a retrieval layer, and the pattern in its top ten looks like what you would expect from a system that learned the world from text. Wikipedia, Britannica, NIH, Healthline. Stable, structured, heavily represented in written corpora.

The industry data points the same way. Internet and Telecom is third in AI Overview citation volume at 25.2 million, concentrated in just 19 domains. In ChatGPT the same category drops to 559,000. Something about how those sources are surfaced is specific to Google rather than to the topic.

The number that should change how you measure

27.3% of ChatGPT answers cited nothing at all. For Google AI Overviews, 0.7%.

That’s worth reading twice. On more than a quarter of ChatGPT answers there is no citation to win, no source list to appear in, and no link to track. The model answers from what it already holds.

Which means that for ChatGPT, a meaningful share of your visibility has nothing to do with being retrieved and everything to do with what the model has absorbed about you. You can’t optimise a page into that position in the same week. It’s built from what the wider web has said about you over a long period, which is a slower and less glamorous piece of work than publishing content.

It also means any tool reporting your ChatGPT “citation share” is reporting on roughly 73% of the answers your buyers actually see.

Reddit is the only overlap, and that tells you something

Second in AI Overviews at 25.1 million. Second in ChatGPT at 3.25 million. It also appears in roughly one in three of Google’s top ten organic results, at 32.15% in Q3 2026, with a top-ten presence that has moved less than three percentage points across four quarters.

Three different systems, three different mechanisms, same platform near the top of all of them.

I would be careful what conclusion you draw from that. The move most people make is to go and post on Reddit, which is usually transparent, usually badly received, and occasionally gets a brand banned from the subreddit where its buyers actually are.

The useful version is quieter. Find the threads that already rank for your important questions. Read what people recommend and whether you are named. If you are not named at all in the conversations that both Google and ChatGPT are reading, that is the finding, and fixing it is a product and reputation job before it is a marketing one.

What this breaks in how agencies report

A single blended AI visibility score averages two systems that agree on one domain out of ten. It produces a number that goes up and down without telling you which platform moved or why.

Track them apart. We already separate them in client reporting, and the gap between a client’s Google AI Overview presence and their ChatGPT presence is routinely the most useful line in the report, because the fix for each one is different work.

For Google AI Overviews, organic strength and being present where Google already looks. For ChatGPT, reference-grade sources, credible third-party coverage, and consistent facts about your business across the web.

There is more on how the platforms differ in our platform comparison and the AI search platforms hub.

The UK angle nobody is making

The Guardian sits in ChatGPT’s top thirty at 226,000 answers. Reuters at 242,000. AP News at 156,000. None of them are in Google’s top thirty.

For a UK business, that puts a specific value on national press coverage that is easy to miss. A mention in credible national journalism is doing a job for ChatGPT visibility that a YouTube video or a social post will not do. And the reverse holds: if Google AI Overviews are where your buyers are, video and community presence are carrying more weight than your PR budget.

Which of those matters more depends on where your buyers actually start, and that is a question about your market rather than about the data.

The caveats, which are real here

The AI answer corpora are United States English-language. The ChatGPT responses span GPT-4o through GPT-5.6, which is a lot of model change inside one dataset, and citation behaviour is known to shift between model generations.

Other research disagrees on the detail. A DerivateX benchmark found the two engines citing the same sources only 4% of the time, which is an even starker divergence than this, but it rests on 402 citations across 15 queries. Profound’s work puts Wikipedia at 47.9% of citations among ChatGPT’s top ten sources but only 7.8% of its total citations, which shows how much the framing of the question changes the answer.

The direction is consistent across all of them. The precise numbers are not, and I would not build a forecast on any single one.

Where to start

Take ten questions your buyers genuinely ask. Run each one in ChatGPT and in Google, and write down who gets named in each. Two columns, ten rows, one afternoon.

Most people doing this for the first time find they are in one column and absent from the other. The column you are missing from tells you which of the two pieces of work above you have not done.

If you would rather see it measured across both at once, the AI Visibility Score covers it and costs nothing.