A Third of AI Citations Point to Low-Authority Domains. That's Not the Interesting Part.

In August, Featured published one of the more useful datasets we've seen on AI search behavior: 22,881 real citations collected across 405 GEO audits, covering 378 brands in 344 industries, with 1,983 real buying questions tested. The headline stat was that 34.5%Featured AI Citation Report, Jun-Aug 2026 of citations pointed to domains with a Moz Domain Authority below 40.

Look past the headline stat and the study gets stranger, and more interesting.

Three of these deserve a closer look.

First: 76%Featured AI Citation Report, Jun-Aug 2026 of cited domains appeared exactly once. The 100 most-cited domains account for under a quarter of all citations; even the top 1,000 account for less than half. There is no shortlist. The idea that AI answers draw from a fixed set of twenty big-name sources, the mental model most marketers carry over from Google's top-10 era, simply doesn't describe the data.

Second: 92.6%Featured AI Citation Report, Jun-Aug 2026 of brands never had their own website cited in answers about their own category. Brand-owned content accounted for 1.7% of classified citations. Competitor domains were cited 13× more often. Businesses are not just underrepresented in AI answers; they are almost entirely absent from the sourcing layer that produces those answers.

Third: authority didn't disappear. DA 80+ domains still captured 31.2%Featured AI Citation Report, Jun-Aug 2026 of citations, and the citation-weighted median DA was 56, well above the domain-weighted median of 37. A small number of large domains get cited over and over. Most cited domains get cited once.

The tempting takeaway is “Domain Authority is dead.” The data doesn't support that. High-authority domains still concentrate a third of citations. The distribution has a heavy head and an extremely long tail.

The right conclusion is less comfortable: we currently cannot predict, with any confidence, why an AI engine selects one source over another. Authority explains some of it. Freshness explains some of it. Content structure explains some of it. But a model that says “a third of citations go to domains most brands have never heard of, three quarters of sources appear exactly once, and brands themselves are almost never in the answer” is not a model anyone was working from eighteen months ago.

For more than twenty years, the industry built infrastructure to understand one system: Google. Backlinks, authority metrics, technical SEO, rank tracking, and thousands of studies. AI search is a different system. A business is no longer trying to rank a blue link. It is trying to become part of an answer, and that answer changes with the question, the location, the model, the sources available at retrieval time, and variables nobody has isolated yet.

22,881 citations is enough to establish that the distribution of AI sourcing is wider and stranger than assumed. It is not enough to answer the questions that actually matter to a business trying to get recommended:

These are longitudinal, cross-engine, geographically distributed questions. Answering them requires observing AI answers the way meteorologists observe weather: continuously, everywhere, across every condition, not auditing a sample and extrapolating.

If 22,881 citations can overturn assumptions the industry relied on for years, the obvious question is what hundreds of millions of observations would reveal across thousands of cities, hundreds of industries, and every major engine, repeated over time so that cause and effect become visible.

That is the research program AI search actually needs, and almost nobody is positioned to run it. Monitoring tools sample. Audits snapshot. What's missing is longitudinal observation at the scale of the system itself.

We don't need more guesses about how AI search works. We need enough data to understand how it actually behaves: locally, per-question, per-engine, and over time. Until that scale of observation exists, single-number headlines will keep contradicting each other, and every brand will keep optimizing for an average that doesn't describe its own market.