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

# Chats

> The raw captured answers behind every metric: full text, brand detections, and cited sources.

Chats are the receipts. Every time Mavel runs a prompt against an AI surface,
it captures the full answer text, the brands detected in it, and the sources
cited. The Chats view lets you read those answers directly.

When a metric moves and you want to know why, chats are the ground truth.
They show you not just that your brand appeared, but exactly what the engine
said: the sentence, the framing, the position relative to competitors.

## What a chat contains

Each captured answer record shows:

* **Surface and country**: which AI engine ran the prompt, and the country
  context it ran in (shown as a flag alongside the surface label; in the UI
  Google AI Overviews and Google AI Mode appear abbreviated as "Google AIO"
  and "Google AIM").
* **Prompt text**: the exact question that was asked.
* **Full AI response**: the complete answer as it was returned, rendered in
  full. No truncation.
* **Cited sources**: the external pages the engine linked or referenced in
  its answer, listed by domain.
* **Brands mentioned**: every tracked brand detected in the answer, with
  your own brand highlighted separately from competitors.

## Navigating chats

From the dashboard, recent chats appear in the Recent Chats block. Click any
card to open the full chat detail page at `/chats/[id]`.

From other analytics pages (Visibility, Sentiment, Sources), contextual links
take you to the chats behind a specific metric so you can inspect the underlying
answers without losing context.

Each chat detail page includes a **View prompt** link that takes you directly
to the prompt configuration.

## What to use chats for

**Sanity-checking brand detections.** Mavel's classifier flags brand mentions
in every answer. If a detection looks wrong, open the underlying chat and read
the full sentence. The classifier operates on the full response text, so edge
cases are visible here.

**Understanding how engines frame your category.** Visibility tells you whether
you appear; chats tell you how. An engine might mention your brand positively in
one sentence and then recommend a competitor in the next. That nuance does not
show in aggregate metrics; it shows here.

**Quoting AI answers in internal reports.** When you need to show a stakeholder
what ChatGPT or Perplexity actually says about your product, copy from the
captured answer. It is a real answer from the real surface on the day it ran.

**Investigating source citations.** The cited sources panel on each chat shows
which pages the engine read. Cross-reference these with the [Sources](/results/sources)
view to understand why certain domains appear, or to spot pages you own that
are already being cited without your knowing.

## Fanout queries

Some AI engines run background sub-queries behind your tracked prompt before
generating their answer. Where available, these are captured and shown on the
chat detail page. Fanouts are currently collected from surfaces that expose
them during a successful answer retrieval, and the Fanouts section of the
dashboard provides a project-level aggregate view of the sub-queries running
behind your prompt set.

<Note>
  AI answers are non-deterministic: the same prompt run twice on the same
  surface on the same day can return different answers. Each chat record is a
  single captured instance. Trends across many chats, not individual answers,
  are the reliable signal. See [Metrics overview](/results/metrics-overview)
  for how daily runs aggregate into the metric surfaces.
</Note>

<Tip>
  Use the surface pill and country flag on each chat to understand whether
  behaviour differs by engine or by market. A brand that appears in US ChatGPT
  answers but not in UK Perplexity answers points to a market-specific content
  or coverage gap.
</Tip>