> ## 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.

# How can an AI assistant analyse sources and sentiment?

> Explore cited domains and pages, export source activity, classify domains, and investigate sentiment with supporting quotations.

The connector can move from a high-level change to the source pages and answer quotations behind it. Ask it to keep the reporting period and filters visible so comparisons remain understandable.

## Explore cited sources

The assistant can use the same core filters as the Sources area to show:

* a source overview;
* cited domains and their activity;
* the pages under one domain;
* answers that cite a selected domain or page;
* supporting prompt runs; and
* a count for the selected filters.

Results can be filtered by available dates, prompts, groups, platforms, regions, source ownership or other supported source filters.

Example:

> For our “Vendor comparison” group, show the most cited domains in the last 30 days. Separate domains we own from independent publishers, then open the evidence for the top independent source.

## Export source activity

The assistant can export filtered source activity as CSV for a period of up to 365 days and up to 10,000 rows. If the result is larger, narrow the period or filters before exporting.

## Classify a source domain

Owners and admins can classify a cited domain, change its source-tracking state, and save a target URL or notes. Ask the assistant to read the current profile first because omitted optional fields return to their defaults.

This is **source-domain tracking**, which classifies a cited domain. To save and measure a specific published article, interview, podcast or PR URL, use [Tracked coverage](/docs/mcp/tracked-coverage).

## Analyse sentiment

The assistant can examine:

* the sentiment overview;
* change over time;
* platform comparisons;
* brand and competitor comparisons;
* recurring aspects or themes; and
* supporting quotations, continued through multiple pages when needed.

Filters can include the reporting period, platform, region, group, competitor, brand type, sentiment and aspect. Ask for the sample size and supporting quotations when drawing a conclusion.

Example:

> In Clean Digital, find the negative sentiment aspects associated with our own brand over the last 30 days. Compare them with active competitors and quote the supporting answer excerpts with their source pages.

## Combine prompts, groups, competitors, sources and sentiment

Useful multi-step investigations include:

* finding prompts with falling visibility, then checking competitor positions and cited sources;
* comparing two groups, then opening the answer evidence behind the sentiment difference;
* discovering a possible competitor, checking its supporting answers, then staging an identity review; and
* finding a frequently cited domain, classifying it, and exporting the related source activity.

The assistant should distinguish observed data from its interpretation and should not claim that a cited page caused a visibility or sentiment change.