Skip to main content
Plans: All plans · In the app: Sidebar → Analytics → Query fan-outs
Query fan-outs answers: what do AI engines search for when they answer my prompts? Before an engine writes an answer with web search, it often runs several searches of its own. Each one is a query fan-out. Fan-outs show the language your category is searched in, and the questions your pages have to answer to be found. Open it from a brand’s Overview section. Query fan-outs is the tab beside Citations and Sentiment.
Not every engine reports the searches it ran. Quotable records fan-outs only from answers that include them. Engines that report no search in your date range are left out of every per-answer figure, so they do not drag the averages toward zero.

Filters

The toolbar is the Overview’s: Date range (the last 30 days by default), Models, Topic, Locations and Prompt type. Every number on the tab covers the answers in those filters. See Overview filters.

Headline figures

How coverage is matched

Engines rarely repeat your prompt word for word. They drop the question word, add the year, or switch to the plural. So a search counts as covered when at least 80% of the meaningful words of the shorter text, the search or your prompt, appear in the other. Filler words such as “the”, “for” and “best” are ignored, and a plural matches its singular.

Searches by prompt

The main table has two views: By prompt and By topic. By prompt is the default. Only prompts that set off at least one search are listed. The subtitle says how many, for example “12 of 40 prompts set off searches”. Rows are ordered by distinct searches, 10 per page. Click a row to see its top five searches, each with how many times it ran. Analyze opens the full breakdown for that prompt. See The fan-out analysis panel. Use Search prompts to filter by prompt text or by one of a prompt’s top searches.

Searches by topic

Click a topic to list its searches. Each one shows a Tracked badge when a prompt covers it, the prompt that set it off most often (with +N when others did too), and how many times it ran. The first eight show; Show all lists the rest. Click a prompt name to open its analysis. Use Search topics to filter by topic name or by any search under it.

Brands in the searches

Your brand and your tracked competitors, when the engines typed them into their own searches. Names match as whole words, and each run of a search counts. Your brand is marked You. This is not the same as being named in the answer. It shows whose name an engine reaches for while it researches, before it writes the answer.

Common phrases

Two- and three-word phrases that recur across searches, counted per run and seen at least twice. Phrases that start or end with a filler word are skipped. When a short phrase never appears outside a longer one, only the longer one is listed. Each phrase shows where it usually sits in a search, Start, Middle or End, with the share of its runs in that position. Filter by position with All, Start, Middle and End. Up to 50 phrases are shown.

The fan-out analysis panel

Fan-out analysis opens on the right, with the table still behind it. Open it in any of three ways:
  • Analyze on a prompt row.
  • The link under an expanded prompt row.
  • A prompt name inside an expanded topic.
The panel uses the tab’s date range, models and locations. The open prompt is part of the page address, so you can share the link. Your browser’s Back button closes the panel. The header shows the prompt and its topic. View answers opens the prompt’s own page.

Search variations

Every search the engines ran for this prompt, most frequent first, 10 per page. Turn on Highlight new words to mark the words in each search that your prompt does not contain. Filter the list with Search variations.

How the prompt was rewritten

Word by word, how the engines turned your prompt into searches:
  • Added: words the engines put into their searches that the prompt never used.
  • Dropped: words from the prompt the engines left out.
  • Kept: words from the prompt the engines carried over.
Each word shows Searches (how many search runs added, dropped or kept it) and Share of searches. A singular and its plural count as one word. Hide filler words is on by default. Added words are the engines’ own vocabulary for your category: a year, “reviews”, a competitor’s name. A page that uses them truthfully matches more of the searches.

Searches per answer over time

A line chart of average searches per answer, per day, or per week when the range holds more than a month of runs. With more than one engine, each one gets its own line. The list beside the chart shows each engine’s searches per answer. Click an engine to hide or show its line.

Put fan-outs to work

1

Find the gaps

In By prompt or By topic, look for rows with many searches and a low Covered share. Those are questions engines research around your prompts that you do not measure yet.
2

Track the searches that matter

On the Prompts page, open Add prompts and choose the Query fan-outs tab. It lists up to 30 searches from the last 30 days that are not already prompts, each with how often it was seen. Select the ones you want and add them. See Add prompts.
3

Write for the searches

Read How the prompt was rewritten for the words engines add. Then cover those questions on your pages. To draft a page, ask Quotable Agent for an article and name the searches it should answer.
Fan-outs also appear in other places: Searches ChatGPT runs to build a product carousel are a separate set. They appear as Shopping searches in Shopping.

Why a prompt has no fan-outs

The tab shows “No fan-outs captured yet” when no answer in your filters reported a search. The panel shows “No searches in this window” for a single prompt. Common reasons:
  • The engine does not report its searches. Some engines never include them in their answers.
  • The engine answered from memory. An answer written without web search has no searches to report.
  • The prompt has not run in the range. New prompts need their first run. Widen the Date range or check your schedule.
  • Your filters exclude it. A model, topic, location or prompt type filter can leave no answers that searched.
A prompt with no searches is still tracked. Its visibility, citations and sentiment are measured as usual.

Next steps

Write better prompts

Turn fan-outs into prompts you track.

Quotable Agent

Ask for an article that answers the searches.

Citations

See which sources the searches lead to.
Last modified on November 18, 2026