> ## Documentation Index
> Fetch the complete documentation index at: https://docs.mentionlab.io/llms.txt
> Use this file to discover all available pages before exploring further.

# Fan-outs

> See the web searches AI platforms run while answering your queries

To answer a question, an AI platform often runs its own web searches first. Those search queries
are **fan-outs**. They show how the model breaks your query into sub-questions, which makes them a
ready-made list of queries worth tracking.

Only some models report their searches. See
[Features by model](https://docs.mentionlab.io/guides/models-availability.md#features-by-model).

Open **Intelligence → Fan-outs**.

## Headline numbers

| Card | What it counts |
| - | - |
| **Total Fan-Outs** | Every fan-out entry in scope. One response can produce several, so this exceeds the response count |
| **Unique Fan-Outs** | Distinct fan-out texts. The same search across ten responses counts once |
| **Avg per Response** | Fan-outs per response, among responses that produced at least one |

> [!WARNING]
>
> Every rate and average on this page divides by **responses that produced at least one fan-out**,
> never by all responses in the period. That's why **Avg per Response** can never drop below 1.0,
> and why adding models that don't emit fan-outs won't move it.

> [!NOTE]
>
> Uniqueness is an exact-text match. `Best CRM 2026` and `best CRM 2026` count as two distinct
> fan-outs.

## Top Fan-Outs

A word cloud of the most frequent fan-out queries, with **Unigrams** / **Bigrams** / **Trigrams**
toggles.

> [!NOTE]
>
> This is a cloud of *terms extracted from* the top 50 fan-out texts, weighted by how often each text
> occurred, not a cloud of the fan-out queries themselves. Stop words and very short tokens are
> dropped. Bigrams and trigrams are usually the readable views.

## Fan-Outs by Query

Which of your queries generate fan-outs, and how many:

| Column | Meaning |
| - | - |
| **Query** | The tracked query |
| **Total** | Total fan-out entries for this query |
| **Unique** | Distinct fan-out texts (deduplicated) |
| **Avg / exec** | Average fan-outs per response for this query |

Only queries that produced at least one fan-out appear. Sort by any numeric column.

Expand a row to see that query's individual fan-outs, each with its occurrence count and rate. The
header tells you the denominator explicitly: "(*n* responses with fan-outs)".

The ⋮ menu exports the current page along with every fan-out belonging to those queries.

## Turning fan-outs into queries

The practical loop:

1. Read the bigram/trigram cloud for themes you don't currently track.
2. Expand your highest-volume queries and skim the actual fan-out texts.
3. Add the good ones as new queries in **Project settings → Queries**.

A fan-out that appears in most responses is a search the model considers unavoidable in your
category. If you have no visibility on it, that's a gap worth closing.

> [!NOTE]
>
> **View as** has no effect here: fan-outs belong to a response, not to a brand, so the page is
> never re-scoped to another entity.

## Exporting

**Project settings → Export data → Fan-Outs By Query** produces two CSVs: per-query aggregates, and
one row per unique fan-out per query with its occurrence count and rate.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.