Skip to main content
Query fan-out map connecting an illustrative prompt group, tracked prompt, model, and captured searches
Open Analytics → AI search → Query fan-out to see which searches were captured when models answered your tracked prompts. Use repeated queries to identify questions your content should answer, then inspect the saved answer before prioritizing work. Screenshots use synthetic local examples.

​
Read the summary

  • Groups counts represented prompt groups when groups exist. A prompt can appear under more than one group.
  • Prompts counts represented tracked prompts.
  • Models counts models with captured queries.
  • Queries counts query entries across prompt and model branches, not globally unique searches or search volume.
Access depends on your subscription and project permissions. An upgrade prompt is an access restriction, not evidence that no searches occurred. A run without captured queries cannot establish that the model did not search; capture availability differs between runs and engines.

​
Explore the map

Follow a group to its prompt, model, and query branches. Expand or collapse branches, pan, zoom, and reset the view to compare related searches. Ungrouped prompts appear without a group branch. Select a query to inspect the runs that captured it. Load a run’s sources with Inspect run sources. These are sources returned for that run: exact query-to-source attribution is unavailable. Do not assume every source came from the selected query or was cited in the final answer.
Selected query showing its captured runs and run-level sources with an attribution caveat

​
Compare queries in the table

Query table with Fan-out query, Tracked prompt, Model, Runs, and Last observed columns
Each row groups the same query for one tracked prompt and model. Runs counts saved runs containing that query; repeated copies inside one run count once. Last observed is the latest matching run time. Sort the columns and use pagination to find repeated or recently observed searches. Select a row to open its newest matching saved answer. When more than one run matches, use the run navigation above the answer to compare earlier observations. This reads saved results; it does not execute a new prompt.

​
Inspect the answer and evidence

The answer preview separates Citations, Query fan-out, and Other search sources. Citations are sources marked as cited. Other search sources carry a Not cited label and must not be counted as citations. A query card opens a new Google search for that phrase. Today’s search results are not a replay of the model’s original search. Use the saved answer and captured sources as the evidence for that run. Missing search sections mean no corresponding details are available in that preview, not proof that the model used no external information. See Open an answer for the saved-answer screenshot and review steps.

​
Choose the next action

  • Check whether a repeated query exposes a question your page leaves unanswered.
  • Compare several saved runs before treating one answer as a persistent pattern.
  • Review cited pages separately from sources returned but not cited.
  • Add content work only after checking relevance, existing coverage, and the underlying evidence.

​
Where to go next