AI Visibility Audit

What an AI Visibility Audit Is

An AI visibility audit measures how often, where, and how favorably a brand gets mentioned when AI assistants answer buying questions in its category. It is the AI-search equivalent of an SEO audit: instead of checking search rankings, you check what ChatGPT, Claude, Gemini, Perplexity, and Grok actually say when a prospective buyer asks for a recommendation.

The output a client actually wants is not a raw pile of chat transcripts. It is a mention rate, a leaderboard of who is beating them, and a short list of what to do about it. Everything below is the process for getting from prompts to that output.

The Audit Workflow, Step by Step

1. Define the category. Write down the exact category the client competes in, in the words a buyer would use, not internal jargon.
2. Build the prompt set. Draft prompts across the five buyer-intent types below, using the category and the client's known competitors.
3. Choose engines. Decide which of ChatGPT, Claude, Gemini, Perplexity, and Grok to include; cover all five if the client's buyers could plausibly use any of them.
4. Decide run count. Pick how many times each prompt runs per engine; more than one run per prompt catches the variation between identical questions.
5. Record raw answers. Log every answer in full, verbatim, before you start scoring anything.
6. Extract brand mentions. Go through each answer and mark whether the client's brand appears, and where.
7. Compute the metrics. Turn the marked-up answers into mention rate, average position, share of voice, and sentiment.
8. Identify citation sources. Note which domains the engines cite or draw from when they answer.
9. Turn findings into recommendations. Convert the gaps you found into specific actions.
10. Present to the client. Package the numbers, the leaderboard, and the recommendations into something a non-technical stakeholder can act on.

The Buyer-Intent Prompt Taxonomy

A prompt set built from one question type produces a skewed audit. Use all five types so the audit reflects the different moments a buyer actually asks an AI for help.

Prompt type	Example	What it reveals
Category best-of	What are the best [category] tools in 2026?	Whether the brand makes the default consideration set
Use-case	I need software to [specific job]. What do you recommend?	Whether the brand wins on the jobs buyers actually hire the category for
Comparison	Compare [Brand] vs [Competitor] for [use case].	How the brand fares head-to-head against a named rival
Alternatives-to-competitor	What are good alternatives to [Competitor]?	Whether the brand captures switcher intent away from the market leader
Price or free tier	What's a good free or cheap [category] tool?	Whether budget-conscious buyers hear the brand's name at all

Write two or three prompts per type, substituting the client's real category and named competitors. Ten prompts across the five types is a workable minimum for a single-category audit.

Choosing Engines and Run Count

Cover all five major engines if the budget allows it. Engines disagree with each other more than most clients expect: one brand measured across the same seven category prompts scored 43% mention rate on Gemini, Perplexity, and Grok, 57% on ChatGPT, and 86% on Claude, all on the same day. An audit that only checked ChatGPT would have reported a completely different story than one that only checked Claude.

Run count matters as much as engine count. AI answers are not deterministic, so the same prompt on the same engine can mention the brand on one run and drop it on the next. A single run per prompt is a snapshot; two or more runs per prompt starts to look like a measurement.

Recording Answers and Extracting Mentions

Paste each prompt into a fresh, empty chat for every engine and run, so chat history or memory does not bias the answer toward brands discussed earlier in the session. Save the full text of every answer before scoring anything; you will want to re-check specific answers once you start building the leaderboard.

For each answer, mark: was the client's brand named (yes or no), at what position in the list, and which other brands were named, in order. This is the raw table every downstream metric gets computed from.

Computing the Metrics

From the marked-up answers, compute four numbers per engine and overall:

• Mention rate — the share of prompt runs where the brand was named at all.
• Average position — the average rank of the brand within the answers where it appeared; a brand mentioned last in every answer has a real visibility problem even if its mention rate looks fine.
• Share of voice — the brand's mentions as a percentage of all brand mentions across every answer, which shows how much of the conversation it actually owns versus competitors.
• Sentiment — whether each mention was neutral, favorable, or backhanded, read off the actual sentence the engine wrote.

Build the competitor leaderboard from the same raw table: count every competitor name that appeared across all answers, and rank them by mention count and share of voice.

Identifying Citation Sources

Some engines show their work. Perplexity lists source links under every answer, and ChatGPT's web-search-enabled answers cite pages directly; note every domain that appears. For a broader read on which domains an engine tends to draw from for a category, a third-party AI-mentions dataset can show the sites most often cited or consulted for buyer questions in that space, which is useful context beyond what you can observe from a handful of manual runs.

This list is often the most actionable part of the audit for a client, because it tells them exactly which publications, review sites, or community pages they need a presence on to influence what the engines say.

Turning Findings Into Recommendations

Convert every gap into a specific action, not a general observation. If the brand's mention rate is low on comparison prompts, the fix is a comparison page targeting the competitors that keep beating it. If sentiment is neutral everywhere, the fix is content that states a clear differentiator an engine can quote. If a citation source keeps appearing and the brand has no presence there, the fix is getting listed or reviewed on that specific site. Five to eight recommendations, each tied to a specific metric or gap, is enough for a client to act on without feeling overwhelmed.

Presenting the Audit to a Client

Lead with the headline numbers: overall mention rate, position, and share of voice, then the per-engine breakdown so the client sees where the spread actually is. Follow with the competitor leaderboard, since that is usually what makes the problem feel real. Close with the citation sources and the recommendations, in that order, so the last thing the client reads is what to do next.

Do It by Hand or Automate It

Everything above can be run by hand with five browser tabs and a spreadsheet, and a competent freelancer can deliver a real audit this way. It takes hours: writing the prompts, running them across five engines, logging every answer, and computing the metrics by yourself.

The AI Visibility Snapshot report runs this exact workflow, 10 buyer-intent prompts across ChatGPT, Claude, Gemini, Perplexity, and Grok, 50 answers, in about two minutes, for $19. It is white-label, so you can put a client's name on the "Prepared for" header and hand over a permanent link, and it includes the same mention rate, position, share of voice, sentiment, competitor leaderboard, and citation-source breakdown described in this guide. See what the output looks like on the sample report, check your own numbers first with the free AI visibility checker, and read the full metric definitions in the methodology.