Fan-out searches

An assistant rarely answers a buying question from memory. It writes its own searches, reads what comes back and answers from that. Those searches are captured per run and listed under AI searches.

Why they are worth more than a keyword list

They are what the machine actually asked, on the way to the answer that mentions you or does not. A keyword list is a guess about demand; a fan-out list is evidence about the retrieval step that decides your candidacy.

They also look nothing like keywords. They are long, oddly specific, sometimes phrased as a question and sometimes as a fragment, and they frequently include a qualifier the buyer never typed because the assistant inferred it.

What to do with them

In a new project this is the first actionable screen, because it needs no trend data.

1. Sort by how often a search appears across your prompts.
2. For the frequent ones, check whether you have a page that answers that exact search.
3. Where you do not, that is a page to write, and you already know its title.
4. Where you do, score that page against that exact query. See retrieval readiness.

Two kinds

Each fanout is labelled discovery, when the model is assembling a list of candidates, or verification, when it is checking a brand or claim it has already surfaced. The two call for different work, and discovery and verification covers which.

Their relationship to prompts

One prompt produces several searches, and the set changes between runs of the same prompt. A search that appears in most runs of a prompt is a stable part of how that question gets answered. One that appears once is incidental.

Sorting by frequency rather than reading a single run is the difference between a pattern and an anecdote.

What they do not tell you

Volume. A fan-out search is not a search anybody typed, so it has no search volume and no competition number, and treating it like a keyword in a planning tool will produce nonsense. Its value is that it names the retrieval target precisely.