The Two Biggest AI Citation Studies Contradict Each Other. Both Are Right.

Two of the largest studies of AI search citations reached conclusions that appear to be irreconcilable.

Yext analyzed 6.8 million citations across ChatGPT, Gemini, and Perplexity and concluded that 86%Yext Research, Jul-Aug 2025 of AI citations come from sources brands already control or manage — websites, listings, and review profiles. The headline was optimistic: visibility in AI search is largely in a brand's own hands.

Featured analyzed 22,881 citations from real buying questions and found that brands' own websites earned 1.7%Featured AI Citation Report, Jun-Aug 2026 of citations — and that 92.6%Featured AI Citation Report, Jun-Aug 2026 of the brands audited never had their own site cited at all in answers about their own category.

86% versus 1.7%. Two serious studies, millions of data points between them, and headlines pointing in opposite directions. One of them must be wrong — or the disagreement itself is the finding.

We think it's the second. Read closely, the two studies aren't contradicting each other. They're measuring different layers of the same system, and the gap between them is the most practically useful insight AI search research has produced this year.

Neither methodology is careless. Yext ran 1.6 million queries per model through a framework that mirrors how a real consumer asks: from a specific location, with a specific intent. Featured audited real buying questions submitted by real users across hundreds of categories. Both published their methods. And both would fail to replicate the other's headline — by design.

Yext's 86% aggregates three categories: first-party websites (2.9M citations), third-party listings a brand can claim — Google Business Profile, MapQuest, TripAdvisor, Zocdoc (2.9M citations) — and reviews/social profiles (545K citations). Featured counted one thing: whether the brand's own domain was cited. Under Featured's definition, most of Yext's 86% wouldn't qualify as brand-owned at all — a claimed TripAdvisor profile is not your website. Under Yext's definition, much of Featured's "editorial and directory" majority would count as manageable. The two headline numbers are answering different questions: "how much of the citation surface can a brand influence?" versus "how often does a brand's own site actually get picked?"

Yext's framework spans objective queries ("what [category] is near me?") through subjective ones ("what's the best...?"). Featured tested buying questions — inherently subjective and comparative. This matters enormously, because both datasets show the same internal pattern: for objective, factual questions, first-party websites dominate — over 40% of citations across all three engines in Yext's data. For subjective questions, the sourcing flips toward third parties: on OpenAI, directories peak at 46.3%Yext Research, Jul-Aug 2025 of citations for branded subjective queries; in Featured's buying questions, editorial content took 64.5% and competitor domains were cited 13× more often than the brand's own site.

Yext injected user location into every query — its core argument is that AI engines answer business questions geographically, so brand-level studies that omit location systematically overweight sources like Wikipedia, Reddit, and national editorial sites. Featured's audits carried no location context. Once location is applied, Yext found Reddit and similar forums drop to roughly 2% of citations. Both things can be true: community content matters for research-mode questions asked from nowhere in particular, and largely disappears when an engine is retrieving results for a specific place.

Yext tested four industries where structured listings are the connective tissue of the entire vertical — retail, finance, healthcare, food service. Healthcare alone drew 52.6% of its citations from directories. Featured spanned 344 industries, most with far thinner listings infrastructure. And the engine mix differed: Yext's totals blend three engines with materially different sourcing behavior (Gemini pulls 52.1%Yext Research, Jul-Aug 2025 of citations from first-party websites; OpenAI leans on listings at 48.7%), while Featured measured Perplexity alone. A year separates the two study windows — an eternity in this market.

Strip away the headlines and the two datasets converge on the same underlying mechanics:

One more note on reading vendor research — ours included. Yext sells listings management; its study's "controllable listings" category is its product. Featured sells earned-media placement; its study spotlights editorial sources most brands have never pitched. That doesn't make either study wrong — the methodologies are published and the data is real — but incentives shape which questions get asked. The same skepticism should be applied to anything we publish here, which is why our sources and methods will always be listed at the bottom of the page.

Buried in Yext's report is the most important sentence in either study: a retail chain might report a 47% first-party citation rate nationally, but location-level analysis could reveal rates of 70% in rural markets and 20% in competitive urban markets — for the same brand, at the same moment.

That's a 3.5× swing in the metric that decides whether a business gets recommended, invisible to every national average — and no published research has mapped it systematically. How does citation sourcing vary across a thousand cities? Does the "best roofer" answer in Austin draw from the same source types as the identical question in Boise? Which industries are aggregator-locked in cities but brand-open in small markets? How stable is any of it week to week?

Answering those questions doesn't take another audit. It takes longitudinal observation across every combination of question, place, industry, and engine — at a scale no one has attempted yet. Until someone maps it, the geographic blind spot in AI search research stays exactly that: a blind spot every national average quietly inherits.