The searches inside the answer

An assistant does not search for the question your buyer typed. It turns the question into a handful of smaller searches, runs them all, and combines the results. The Fan-Out Map shows those searches for every tracked question, and which of your pages competes on each one.

What you see

Open any tracked question to find its panel. It lists the sub-searches the model ran, how often each one appears, and, for each sub-search, the pages of yours that come closest to answering it. The sub-searches are recorded from the real answers, not guessed.

Why it changes what you write

Pages compete on the sub-searches, not on the question itself. A page can lose the question your buyer typed while being the best answer to one of its variations. A page that never mentions the question can still win it through a variation it answers well. Writing for the typed question alone means writing for a search no model runs.

Breadth

Results are combined across the variations rather than picked per search, so a page that shows up for 3 of them can beat a page that leads on one. The map counts how many sub-searches each of your pages reaches. That breadth is what the combining rewards, and it is invisible when each search is read on its own.

Without semantic scoringThe map still lists the sub-searches and how often they occur, which is useful on its own. The scoring columns appear when semantic scoring is running on Visibian’s side; there is nothing for you to set up.

Next: The Similarity Gap →

Updated 23 September 2026. Something unclear? Write to [email protected]. A person reads it.