How Google AI Mode chooses the brands it names, and how to track it.

It writes its own searches first. Google's guidance for site owners says AI Overviews and AI Mode may use a query fan-out technique, issuing multiple related searches across subtopics and data sources to build one response. Announcing AI Mode in May 2025, Google's Head of Search described it as breaking your question into subtopics and issuing a multitude of queries simultaneously on your behalf. A Google engineering director later put the ordinary case at about a dozen searches; Deep Search in AI Mode can issue hundreds. We collected Google's statements in one question becomes a dozen searches, and the term is defined in the query fan-out entry.

It names the brands those searches surface. Each sub-question, on price, alternatives, who a product suits or what it connects to, returns its own pages. The answer is assembled from them, so a brand that only answers the head term can be left out of every sub-question that decides the shortlist. Most of those pages are lists, reviews and comparisons written by someone else: in the category we sampled for that post, listicles were 44% of what ChatGPT cited for a best-of question.

Eligibility is ordinary Search eligibility. Google says a page must be indexed and eligible to be shown in Google Search with a snippet to be a supporting link, that there are no additional technical requirements, and that you do not need new machine readable files, AI text files or special schema.org structured data.

It runs on Gemini, inside Search. Google says AI Mode and AI Overviews use a custom version of Gemini in Search. That is not the same as the Gemini app, which answers on its own terms; see how Gemini picks the brands it names.

Separately from AI Overviews, and never from one screenshot. Search Console reports AI feature traffic inside the Web search type, so it cannot show which searches AI Mode ran or which brands it named. The only way to see that is to ask AI Mode the questions buyers ask and record what comes back.

1. Write the fan-out yourself. Take each buyer question and list the ten or twelve sub-questions a buyer has to settle under it: price, alternatives, who it suits, integrations, why people leave. That list is your real question set.
2. Ask from a clean context. Signed out, in the market you sell to, with the same wording each time, so the answer reflects a first-time buyer and not your own history.
3. Record four things per answer. Whether you are named, in what position, which competitors are named instead, and which pages are linked.
4. Repeat before you conclude. In the categories we sample, 9% to 27% of the names in an answer change from one day to the next, depending on the engine, and 47% of day-to-day changes reverse a day later (half of daily changes reverse). Read a rate over several days.
5. Check the tool sees the surface. Google loads many AI Overviews after the results page. Our own AI Overview sampler once recorded no overview on 193 of 201 searches because it never waited for them (the tracker that missed 96%). Ask any tool how it captures AI Mode, by name.