igeo.ai DocumentationOpen app

Analyze AI visibility

Understanding chats

How to read individual AI responses-the raw material behind visibility, sources, and competitive views.

View Markdown · AI index · Full documentation for AI

Overview

Each chat here means one stored AI response to a prompt your workspace monitors — the atomic unit behind mentions, visibility, sources, and competitive views. Reading a single response carefully explains what the dashboards are aggregating over time.

Use Mentions (and recent-answer cards on the dashboard when shown) to open prompts, scan the response list, then open a row for the full breakdown.

Anatomy of a response

The detail drawer pairs your user prompt with the model's answer. Above the body, a metadata strip summarizes that run:

  • Provider: which surface produced the answer (for example ChatGPT, Gemini, Claude, Perplexity) — shown with the model's icon.
  • Model: the specific model identifier for that run.
  • Sentiment: how the answer talks about your brand when scored for that response.
  • Visibility score: a radial chart summarizing visibility for that answer.
  • Sources preview: favicons for URLs associated with the run (more load in the body or via load-more when the full list is longer than the first page).

Regions

Prompts run in the regions you configure for your workspace; each stored response carries a region when present. Geography matters for comparability — compare runs that use the same prompt and region when you judge wording or sourcing differences.

Main answer and brands

The body renders the model's text with markdown-style structure (headings, bullets, paragraphs, inline links). When structured brand data exists for the run, you also see ranked companies with short reasons — useful for seeing how the model ordered players in that specific answer, which feeds ideas about position at the aggregate level.

Sources and links in the text

A response may include URLs inline in the narrative and a broader set of URLs attached to the run. Inline links reflect what readers see in the prose; the full source list can be longer or paginated — use Load more sources when available so you review everything the run captured, not only what appears in the excerpt.

Model behavior varies by provider: some runs surface many sidebar sources but fewer explicit in-text citations. Track trends across weeks rather than fixating on a single volatile day.