What is an AI visibility audit.

The minimum protocol.

A real audit measures more than "did the brand appear." At minimum it should include:

1. Multiple engines. Anthropic Claude, OpenAI ChatGPT Search, Perplexity Sonar-Pro at minimum. Ideally add Google AI Overviews, Microsoft Copilot, and Gemini once they are stable.
2. Multiple samples. AI answers vary between runs. A single sample tells you nothing about the base rate. Three samples per prompt per engine is the practical floor.
3. Multiple intent categories. Category-named queries pull different vendor sets than problem-first queries. Averaging across the two hides the pattern.
4. Two-stage brand matching. Naive keyword matching inflates counts for any brand whose canonical name is a common English word (Segment, Make, Public, Notion, Motion, Front). Adjudicate ambiguous candidates against the surrounding context.
5. Recommendation classification. Not every mention is a recommendation. See cited vs chosen.
6. Domain citation analysis. Which external domains do the engines cite when answering questions in your category. Which of those are your own, which are review sites, which are competitors.
7. Declared methodology. Version, sample size, prompt list, engine settings, matcher version, right of reply. Buyers can compare audits only if the methodology is comparable.

What The Hint Co. produces.

Every Hint Co. audit produces a structured JSON report and a rendered case file. Both include:

• Full leaderboard across the category with appearance count and pick share of total
• Per-bucket breakdown (which brand wins category discovery, which wins problem-first, which wins comparison, and so on)
• Zero-vendor count per bucket
• Three verbatim response excerpts, chosen from the highest-signal bucket
• Top-30 cited domains with classification (brand-owned, third-party editorial, listing site, community)
• Adjudication summary showing ambiguous candidate counts and confirmed/rejected breakdowns
• Recommendation-status breakdown per brand (SOLE, PRIMARY, ALTERNATIVE, CONDITIONAL, NEUTRAL, UNFAVORABLE)
• Methodology version stamp and last-reviewed date

Public audits publish free at the research library. Client audits produce the same structure with the client's category as the subject.

Right of reply.

Any brand named in a published Hint Co. audit can request a free counter-audit of their own category. Findings publish alongside. This exists because a diagnostic that cannot be challenged is a scorecard, not a diagnosis.

What an audit does not do.

An audit is not a monitoring dashboard. It is a point-in-time measurement with a written interpretation. Dashboards are useful for tracking movement over months. Audits are useful for deciding what to change and in what order. Different products, different jobs. The Hint Co. builds audits and translates them into fix lists through the read. Companies that need ongoing monitoring use tools like Profound, Peec AI, Otterly, or Scrunch alongside a Hint Co. audit.

How to evaluate an audit provider.

• Do they publish their methodology? Ask to see the version history and the extraction rules.
• Do they publish audits of brands other than clients? If they do not, you cannot compare quality before buying.
• Do they extend right of reply to named brands? A provider that refuses this is publishing marketing, not measurement.
• Do they classify recommendation status separately from appearance? Or is every mention counted the same way?
• Do they run multiple samples per prompt? Or one? A single-sample audit measures noise.