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

# Sources

> How Chatobserver organizes domains, citations, mentions, and prompt-level source activity.

# Sources

The sources surface is where raw AI answers become something a team can analyze at scale.

## What sources capture

Chatobserver normalizes citations from prompt runs into source records so you can inspect:

* cited domains
* URLs
* mention counts
* citation counts
* prompt-level source activity
* daily rollups across the current window

## Main ways to use the sources view

Teams usually use sources to answer three questions:

1. Which domains are showing up most often?
2. Which prompts are driving those mentions?
3. How is that changing over time?

## Core views

The sources experience is built around a few high-value slices:

* top domains across the current filters
* daily mentions by domain
* overview metrics for the selected window
* prompt-run level source activity

## Filters that matter

The most important filters are:

* prompt
* domain
* lookback window
* metric
* row count or page size

Use filters deliberately. A wide window is good for trend recognition, but a narrow window is better for investigating a specific movement.

## Interpreting the metrics

Two teams can look at the same sources table and ask different questions:

* a marketing team may care most about share of mentions and top cited domains
* a content or communications team may care most about which URLs and excerpts are appearing

The same source data supports both, but the correct filter window is different.

## Exporting source data

Sources can be exported as CSV for offline analysis, stakeholder delivery, or downstream processing. Export is most useful once you have already narrowed the slice you care about.

## Practical advice

Do not treat one high-citation domain as the whole story. Use sources together with prompts and runs so you understand:

* which questions produced the citations
* whether the results are stable or temporary
* whether the same domains appear across multiple prompts or only one narrow case