How AI answer engines cite sources (and why it moves)

Marketers keep asking one version of a question: how do AI answer engines decide what to cite, and how do we get cited? The honest reply is that there is no single answer, because ChatGPT, Perplexity and Google AI Overviews do not cite the same kinds of sources — and the mix each one favours keeps moving. We read three of the largest published citation studies and pulled the numbers into one place. Every figure below is quoted from its original study and linked.

The short version

AI answer engines do not cite the same sources. Across 680 million citations analysed by Profound between August 2024 and June 2025, Wikipedia was ChatGPT’s single most-cited source — 47.9% of its top-10 sources — while Reddit dominated Perplexity’s, at 46.7% of its top-10 sources. Google AI Overviews leaned on Reddit too (21.0% of its top-10 sources) but barely touched Wikipedia, which was just 0.6% of all its citations. Otterly’s analysis of over a million citations in January–February 2026 found the brand-versus-community split flips by engine: brand-owned domains were 59.8% of Google AI Overviews citations but only 28.9% of Perplexity’s. And the mix is not stable. Ahrefs found the share of AI Overview citations coming from top-10 organic results fell from 76% in July 2025 to about 38% in early 2026. There is no single “AI SEO”: visibility has to be measured for each engine, and re-measured over time.

Each engine has a different citation backbone

Profound analysed 680 million citations from August 2024 to June 2025 (full study, published June 2025). The clearest finding is that each engine has a distinct backbone — the source type it reaches for first.

Engine	Dominant source (share of its top-10 sources)	That source across all its citations	Also notable
ChatGPT	Wikipedia — 47.9%	Wikipedia 7.8% of all citations	.com = 80.4% of citations, .org = 11.3%
Perplexity	Reddit — 46.7%	Reddit 6.6% of all citations	YouTube 2.0% of all citations
Google AI Overviews	Reddit — 21.0%	Reddit 2.2%, Wikipedia 0.6% of all citations	leans on the existing Google index

A note on how to read this, because it is where secondary write-ups get it wrong. “47.9%” is Wikipedia’s share of ChatGPT’s top ten sources — not of every citation. Across the whole dataset Wikipedia is 7.8% of ChatGPT’s citations. Both numbers are real and they say different things: Wikipedia is disproportionately concentrated at the very top of ChatGPT’s source list, but the long tail is enormous and .com-dominated. The engines diverge sharply at the top: ChatGPT’s anchor is an encyclopedia, Perplexity’s is a forum, and Google AI Overviews sits between them while barely citing Wikipedia at all.

Brand versus community: the split flips by engine

Otterly’s AI Citations Report analysed over 1 million citations across the same three engines in January–February 2026 (report). Overall, community platforms such as Reddit and Quora took 52.5% of citations versus 47.5% for brand-owned domains. But that headline hides a wide spread once you separate the engines.

Engine	Share of citations going to brand-owned domains
Google AI Overviews	59.8%
ChatGPT	44.7%
Perplexity	28.9%

The same query can surface a very different balance of sources depending on which engine answers it. On Perplexity, fewer than three in ten citations went to a brand’s own domain; on Google AI Overviews, nearly six in ten did. If your GEO plan assumes “get cited” means the same thing everywhere, this is the number that breaks the assumption. Otterly also noted AI Overviews appeared for roughly 33% of the queries it tracked — a reminder that on most searches, the AI Overview citation slot isn’t even in play.

The ground is moving: AI Overviews is decoupling from organic rank

The most consequential finding for anyone who “already ranks” is that the old shortcut — rank in the organic top 10 and you’ll be cited — is fading fast. Ahrefs analysed 863,000 keywords and 4 million AI Overview URLs in early 2026 (as reported by ALM Corp) and found the share of AI Overview citations coming from top-10 organic pages had fallen to about 37–38%, down from 76% in a prior Ahrefs study of 1.9 million citations in July 2025.

When	Share of AI Overview citations from top-10 organic pages	Source
July 2025	76%	Ahrefs (1.9M citations; median cited position was 4)
Early 2026	37.1% (organic-only)	Ahrefs (863k keywords, 4M AI Overview URLs)
February 2026	17%	BrightEdge (different methodology, same direction)

In the early-2026 Ahrefs breakdown, 37.1% of organic-listing citations came from positions 1–10, 26.2% from positions 11–100, and 36.7% from beyond position 100. Put plainly: roughly two in three AI Overview citations now come from pages a searcher would never reach on page one. BrightEdge, measuring differently in February 2026, put the top-10 overlap even lower, at 17%. The two studies disagree on the exact figure — that disagreement is itself the point — but they move in the same direction: citation is decoupling from rank, and it is doing so within months, not years.

What this means for GEO

Three consequences follow directly from the data above, and none of them are “write more content.”

1. Measure per engine, not once. A brand-domain citation rate of 59.8% on Google AI Overviews versus 28.9% on Perplexity means the same page can be a citation magnet in one engine and invisible in another. A single “AI visibility score” averages away the only information you can act on.
2. Match the backbone. If Wikipedia anchors ChatGPT and Reddit anchors Perplexity, then presence and accuracy on those platforms is part of the work — being cite-worthy on your own domain is necessary but not sufficient.
3. Re-measure on a schedule. A 76%-to-38% shift in six months means any citation audit has a short shelf life. Treat AI visibility like a metric that drifts, because the studies show it does.

This is the whole premise behind measuring AI citations rather than guessing at them. If you’re starting, our guides on how to rank in ChatGPT, how to get cited by Perplexity and measuring AI share of voice go a level deeper; our State of GEO tools 2026 report covers the tools that do the measuring, and Peec vs Profound vs Scrunch vs Otterly helps you pick between the four that come up most often.

The same volatility shows up outside search, too. Ask an AI assistant to name the best tool in any category — terminal coding agents, say, which our sister index The Agents Index tracks — and the sources behind that recommendation shift by engine and by month, for exactly the reasons above.

Methodology & sources

This report is a synthesis, not original measurement — its value is putting three independently-conducted studies side by side and reading them honestly, including where they disagree. Every figure is quoted from its source study, dated, and linked; nothing is estimated or averaged into a single “AI visibility” number, because the studies show that number would hide the divergence that matters. Where a metric is reported two ways (share of top-10 sources versus share of all citations), we say which is which.

Primary sources: Profound, AI Platform Citation Patterns — 680M citations, Aug 2024–Jun 2025 (link); Otterly, The AI Citations Report 2026 — 1M+ citations, Jan–Feb 2026 (link); Ahrefs and BrightEdge AI Overview citation analyses, early 2026, as compiled by ALM Corp.