AI Visibility Audit

Audit how a brand appears in a dated sample of AI answers and identify useful investigations. Connect through the unifapi skill. Read the current schema and price for each operation, set a budget, and reuse existing product context.

Workflow

1. Choose 10–30 prompts covering definitions, comparisons, purchase criteria and concrete use cases. Freeze brands, aliases, citation domains, market and language.
2. Use POST /geo/answers for ChatGPT or Gemini. Record engine, surface, exact prompt, search mode, reported model (or null), observation time, answer, sources, brand observations and request id. ChatGPT natural and forced-search samples belong in separate groups.
3. Use /geo/serp for Google AI Mode and /seo/serp with include_ai_overview for Google Search AI Overviews. Inspect answer references; a top-level target match can also refer to another link. A result's rank is not a brand recommendation rank. Do not attribute either Google response to ChatGPT.
4. Keep corpus discovery separate. /geo/mentions/search, /top-domains and /top-pages identify indexed answers and frequently represented sources. They do not confirm the result of a particular live prompt. /cross-aggregated-metrics returns group counts with potentially overlapping groups, not a computed citation share.
5. Save raw normalized responses and billing. Use the runner in llm-mention-tracking/scripts/monitor.mjs for budgeted ChatGPT/Gemini collection, resumable snapshots and CSV evidence.

Measure coverage and tracked-brand share

Use the definitions in references/geo-methodology.md. Report completion rate, answer rate, brand mention coverage and citation coverage per engine. Coverage measures presence across successful collections, including valid no-answer results. Share measures a brand's fraction of the summed observations for tracked brands; it does not estimate total market share.

Do not merge citation links with unused search results. Deduplicate a brand within each answer. Missing denominators are N/A. If weighting by estimated AI search volume, show the unweighted result and use a denominator summed across every tracked brand. Never use an “any brand appeared” denominator and label the result share.

Investigate misses