New data: AI recommends brands only in their core topic
A study of 1,094 categories shows AI cites brands on almost any topic but names them only where they run deep. Here's how to earn the recommendation.
AI search will cite almost anyone. It recommends far fewer. A new study of 1,094 categories found that a brand can show up as a source on topics far from its expertise, but AI names it as a recommendation mainly where it has built repeat topical authority. If you want to get recommended in AI search, the move is depth in one category, not a thin presence across many.
The gap between a citation and a recommendation
Kevin Indig pulled the numbers from Semrush's AI Visibility Toolkit, covering U.S. ChatGPT answers from January to June 2026: 1,094 categories, five prompt variants each, 283,215 citations and 76,493 named brand mentions. He published the analysis in his Growth Memo on August 3, and Search Engine Land ran it two days later. The split is the story. A citation is a link or a reference the model leans on. A named mention is the model saying your brand out loud as an option. Those are two levels of trust, and the data shows they don't move together.
In categories close to a brand's core expertise, 74% of brands got cited, 44% got named, and 34% got both. Step out to distant categories and citations hold up while recommendations drop hard: 50% still get cited, only 25% get named. Citation-only presence barely moves with topic relevance, 41% in distant categories against 40% in close ones. Read that again. You can be a source almost anywhere. You get recommended where you go deep.
Getting recommended in AI search takes repeat topical authority
The pattern behind the numbers is repetition. A category isn't won on one mention. The brands AI recommended were the ones that showed up across all five prompt variants for a topic, not the ones that appeared once and vanished. That's topical authority doing its job inside the model: consistent, specific coverage the system has seen enough times to trust as an answer instead of a footnote. If you've read our take on building a topic cluster that actually ranks, the same logic carries over. Depth on one subject beats a shallow page on ten.
Breadth still buys citations, and that's worth something
The effect isn't uniform across fields. Finance and real estate brands expanded into adjacent categories on the strength of citations alone, so breadth paid off for them. Legal and healthcare were stricter: AI held back recommendations until a brand earned repeat mentions, because the cost of a wrong answer runs higher. So the read depends on your field. In lower-risk categories, citations can carry you onto new ground. In high-stakes ones, you earn the recommendation slowly or you don't earn it at all.
What to do about it this week
Three moves. First, change what you measure. If your AI visibility report counts citations and stops there, you're grading yourself on the easy metric. Track named mentions on their own, because that's the number tied to being recommended. Second, pick one category and go deep before you spread out. A single topic covered from every angle earns the repeat mentions that turn into recommendations. Third, keep the three jobs separate: ranking in Google won't get you cited in ChatGPT on its own, and getting cited won't get you recommended. This data shows the last job is the hardest to fake.
If you want help turning this into a plan, picking the one category to own and building the depth that gets you named, that's the work we do.
