Which kinds of sites do AI search engines cite most?

AI search engines cite unevenly. A small group of large, familiar publishers and platforms collects most citations, and a long tail of sites shares the rest.

The clearest number comes from a study of news citations in AI search. It found that the top 20 most frequent news sources accounted for 67.3% of all citations for OpenAI models. That is a measurement of one category (news), on one evaluation platform, over a fixed window.

What it means for you: if your category behaves like news, engines will lean on a handful of names. Your route into the answer often runs through those names, not around them. The rest of this page covers how far the finding stretches, why it happens, and what to check next.

The study is News Source Citing Patterns in AI Search Systems (arXiv:2507.05301, published July 2025). Its data came from AI Search Arena between March and May 2025.

It reported concentration with a Gini coefficient, where 1 means one source takes everything:

Three limits matter before you act on it:

• It is old. As of October 2026 the data window is well over a year behind, and models have changed since.
• It is self-selected. AI Search Arena users chose their own questions, so the prompts are not a random sample of what buyers ask.
• It covers news. It says nothing direct about software, local services or ecommerce.

Read it as strong evidence that citation concentration exists. Do not read it as a current league table for your market.

Big, well-known sources get cited more. That sounds like a ranking factor, but the evidence treats it as an association, not a lever.

Other data points the same way. Semrush's index of 126 million prompts (January to April 2026) found only 36 brands visible across all four surfaces it tracked every month, including YouTube, Google, Reddit, Amazon and Walmart. That list is largely a list of the biggest brands on the web.

The problem for anyone studying this: brand prominence is the dominant confounder in every correlational citation study, and nothing controls for it. Metrics like Moz Domain Authority or Ahrefs DR are built from backlinks and traffic, so they proxy the same confounder. No engine computes them, and Google publishes no site-level authority score.

Two consequences follow:

• Any benchmark that compares your brand with a household name is mostly reporting brand size.
• "Build authority" is not a measurable task. Being named on the sources engines already trust is something you can actually work on.

Also keep in mind that a citation is not the same as influence. OpenAI's API returns the URLs a model consulted separately from the ones it cited, and the consulted set is usually larger. Logging retrieved vs cited sources shows the gap.

Concentration is not one number across AI search. Each engine draws on its own index and retrieval pipeline, so the cited sites differ.

Otterly's 2026 experiment with two AI-generated sites shows how far this goes. After Google deindexed the sites, Microsoft Copilot, which grounds on Bing, kept citing them. Copilot produced 2,718 of the 3,934 citations Otterly tracked for one site (69%), while ChatGPT, Gemini and Google's AI surfaces almost never cited it. It is one small test, but it shows that a site can be heavily cited on one engine and absent on another.

Google's AI features also cite beyond the top organic results. Ahrefs found that 37.9% of URLs cited in AI Overviews also appeared in the first 10 results (March 2026). Most cited URLs came from lower positions.

How engines pick sources also differs. ChatGPT rewrites your prompt into one or more targeted queries before it searches, so the sites it cites answer queries you never see. For specifics, see where ChatGPT citations come from and why Claude and Perplexity cite sources differently.

If you are new, concentration works against you. Whether a brand-new site can get cited depends first on access and indexing, not on size.

General studies tell you concentration exists. They do not tell you which sites matter in your category. You find that by asking.

1. Write 10 to 20 buyer questions. Use the words customers use, such as "best payroll software for a 20-person agency".
2. Run them on each engine separately. Log out and turn memory off so results are closer to what a stranger sees.
3. Record every cited URL. Note the domain, the page type (review site, marketplace, forum, publisher, competitor) and whether your brand is named on it.
4. Group by domain. The domains that repeat across questions are your category's version of the top 20.
5. Act on the gaps. Where a cited page covers your category without naming you, that page is a target for outreach, a listing or a correction. Being named there is associated with appearing in the answer, though no one can promise it.
6. Check your own access too. Make sure retrieval crawlers such as OAI-SearchBot and PerplexityBot are allowed and not blocked by your firewall.

The full method is in finding the third-party pages AI engines cite. If marketplaces dominate your list, read why AI links to resellers instead of your site. For a quick first look, the free checker on this site shows whether engines name you for up to three buyer questions.