We mapped every source four AI search engines cited across 100 questions. The most-cited domain got 2.4%. Two thirds of cited domains appeared exactly once.
The 2026 story about AI search is that a handful of big sources win everything. It is half true. A few domains do get cited more than the rest. But when we mapped every source four AI search engines cited across 100 questions, the concentration was far weaker than the leaderboards suggest. The most-cited domain in the whole set took 2.4% of citations. The top ten together took 14%. And more than two thirds of the domains AI cited appeared exactly once. AI search does not have a front page. It has a long tail.
In September 2026 we asked four AI search engines 100 questions across everyday categories (health, personal finance, tech and product reviews, shopping) and recorded every source each one cited. After resolving the engines’ redirect links to the real destination domain, that left 2,270 citations pointing at 1,032 distinct domains. The first thing the data kills is the idea of a dominant set.
forbes.com led the whole set at 2.4% of citations. mayoclinic.org, health.harvard.edu and a spread of reference and reviews sites followed, each around 1 to 2%. The top ten together reached about 14%, and it took roughly fifty domains to cover a third of all citations. That is concentration, but mild. Compare it to a page of search results, where the top handful of domains take the overwhelming majority of clicks. AI search spreads its sourcing far wider than that.
The tail is the real story. Of the 1,032 domains AI cited, 711 (69%) were cited exactly once across the whole study. These are not the usual authority names. They are a specific forum thread, a niche review, a single manufacturer page, a local explainer. AI assembled each answer from whatever fit that question, and for most questions that meant sources it would not reach for again.
A ranked list of the "most AI-cited domains" is not wrong, it is describing the thin head of a very long tail and treating it as a map. For almost any specific brand or page, the relevant question is not "am I in the top ten" (almost nothing is) but "for the questions that matter to me, does AI cite me, on which engine, and does it hold." That is a per-domain, per-engine, repeated measurement, not a leaderboard position. It is the same conclusion as our 8% cross-engine overlap and volatility findings, reached from a third angle: AI citation is too spread out and too specific to manage from a single ranking. Prove it on your own domain with the free AI visibility check.
Across 100 questions and four AI search engines (Google/Gemini grounded, Bing/Copilot grounded, ChatGPT search, and Perplexity), the engines returned 2,270 cited links, which resolved to 1,032 distinct domains. We first resolved the engines’ redirect URLs, for example Google’s vertexaisearch.cloud.google.com links, to the real source domain, so the count reflects actual publishers rather than the engines’ own plumbing.
forbes.com, at 2.4% of all citations. No domain came close to dominating. The top ten domains together accounted for about 14% of citations, and it took roughly fifty domains to reach a third. There is no small club of "AI-cited" sites that captures most of AI search.
It qualifies them. Concentration exists: big trusted publishers and reference sites do appear more often. But it is mild. A 2.4% leader and a 69% one-hit tail is a long-tail distribution, not a winner-take-all one. A leaderboard that ranks the "top AI-cited sites" is describing the thin head of a very long tail.
Stop trying to join a top ten that does not really exist, and measure where you actually get cited. Because citations are spread across a thousand domains and differ by engine and by question, the only reliable signal is your own: which engine cited you, for which query, with the answer kept as evidence, tracked over time.
The 8% cross-engine overlap behind the fragmentation.
Read it →Why a single citation score hides this long tail.
Read it →See which engines cite your domain now, with the answer attached.
Read it →Unsourced captures which AI assistants cite you, proves which crawlers really fetched your pages, and re-checks after you act — evidence, not a score.
© Unsourced, the evidence layer for AI search. Our data and findings are free to quote and cite — please attribute to Unsourced and link to unsourced.app.