Features
Fan-out queries
See the web and maps searches ChatGPT and Gemini ran while answering your prompts, what each one was looking for, and how to export them.
4 minute read · Last reviewed: October 24, 2026
When ChatGPT or Gemini answers one of your prompts, it often searches the web first, sometimes several times with different wording. Those searches decide which pages the AI reads before it writes its answer. If your pages don't come up for them, your brand is less likely to be cited or named. Fan-out queries show you exactly what was searched.
What a fan-out query is
A fan-out query is one web search, or one maps search, that an AI model ran while answering a tracked prompt. Your prompt itself is never listed. Only searches the AI provider reported are shown, in the order the model ran them.
Fan-out queries are recorded for ChatGPT and Gemini. Perplexity, Google AI Overview and Google AI Mode do not report the searches they run, so their answers have no fan-out section.
What each chat can show
Not every answer comes with its searches. Each ChatGPT or Gemini chat shows one of these:
| What you see | What it means |
|---|---|
| A numbered list of queries | The model searched, and reported what it searched for. Maps searches carry a Maps tag. |
| No fanout queries recorded. ChatGPT (or Gemini) answered without searching the web. | The model answered from what it already knew. |
| No fanout queries recorded. ChatGPT (or Gemini) did not report its searches for this chat. | The model may have searched, but its reply did not say what for. This is common. |
| No fanout queries recorded. Not recorded for this chat. | The chat ran before we started recording searches. |
A chat from a model that does not report searches shows no fan-out section at all, not an empty one.
Query types
Each query gets a type that says what the search was looking for:
| Type | The search is looking for |
|---|---|
| Discovery | Products, services or providers of a kind, best or top lists |
| Comparison | Named options weighed against each other, "vs", "alternatives to" |
| Brand lookup | One named company or product |
| Pricing | Prices, costs, fees or plans |
| Reviews | Reviews, ratings, complaints or trust |
| Careers | Jobs, hiring, salaries |
| How-to | How to do or choose something |
| Research | Explanations, facts, evidence, statistics |
| Regulation | Laws, licensing, compliance, eligibility rules |
| News | Recent events, launches, acquisitions |
| Local | A maps search, or providers in a named place |
| Site search | A site: search inside one website |
| Other | None of the above fits clearly |
Site search and maps searches are typed from how they are written. The rest are typed by a classifier, and a query it is unsure about shows as Other. A query not typed yet shows – for a moment and fills in on its own.
Where to see them
On the prompt page
Open a prompt from your prompt list. When it has fan-out queries in the selected date range, a Fanout queries view appears beside Overview. The table has these columns:
- Fanout query: the search, as the model last wrote it.
- Type: what the search was looking for.
- Dates: when it was used.
- Times used: how many chats ran it.
- Models: which AI models ran it.
Search the table with Search queries, or filter it with the type list, which starts at All types and shows a count beside each type. Expand a row to see the chats that ran it, and choose View chat to read the answer. The table shows 100 queries at a time; choose Show more for the next 100.
If the date range holds a very large number of chats, the table says it is showing the most recent ones. Pick a shorter date range to see them all.
In a single chat
When you open a chat, from the prompt page or from Recent Chats on your dashboard, the queries for that chat appear in a Fanout queries section with their count. Free reports have no fan-out section.
Export to CSV
With the Fanout queries view open, choose Export to CSV on the prompt page. The file holds every fan-out query in the date range, with your search and type filter not applied. It has one line per query per chat, with these columns: Prompt, Fanout query, Query type, Search (Web search or Maps search), Order, Model, Provider, Date, Time (UTC) and Chat ID.
Read them with the REST API or MCP
With an API key, you can read the same records programmatically:
- REST:
GET /api/v1/brands/{brandId}/prompts/{promptId}/fanout-queries - MCP: the
get_fanout_queriestool
Both accept the same filters: a chat (run_id on REST, runId on MCP), model, type, from and to, plus offset and limit. The window defaults to the last 90 days and can span up to 365. Pages hold 50 records by default and up to 200.
Each record has the prompt, the chat, the provider and model, the query text, its order, whether it was a web or maps search, its type and its time. The response also lists the chats behind the page with their state (captured, none, not reported or not recorded), so you can tell a chat with no searches from a chat whose searches were not reported. See REST API and Connect Claude, Cursor, or ChatGPT via MCP.
Using them
- Look for queries in your category that your pages would not rank for. Those are content gaps the AI is actively searching.
- A Site search aimed at a competitor's site shows the AI checking that competitor directly.
- Compare queries across ChatGPT and Gemini. They often search with different wording.
Fan-out queries show what a model searched, not what it then read or cited. For cited pages, use the Sources tab.
Frequently asked questions
Why does a chat say no fanout queries recorded?
Either the model answered without searching the web, or it did not report its searches for that chat, or the chat ran before recording began. The chat says which of the three applies.
Why is there no Fanout queries view on my prompt?
The view appears only when the prompt has fan-out queries in the selected date range, and only ChatGPT and Gemini report them. Try a wider date range.