AI visibility audit: a 5-step method

In short: an AI visibility audit defines a set of questions, retains the observed responses, and analyses them by model, query, and context. It describes what was measured in that sample; it does not prove a single cause or general visibility beyond that scope.

For a complementary framework, read our GEO audit guide.

Measuring a brand's visibility on conversational AIs is far from obvious. No Google Search Console for ChatGPT. No standardized metrics. No consensus on what to look at.

This absence of an established framework is both a problem and an opportunity. A problem because you need to build your methodology from scratch. An opportunity because actors who master AI visibility auditing today have a rare competitive advantage.

This guide sets out a practical framework: steps, indicators, interpretation limits, and elements that can be standardised.

Why audit AI visibility

Before discussing methodology, let's clarify the exercise's utility.

Establish a baseline. An initial measurement lets you measure observed changes in later runs with a comparable scope. It provides a starting point, not proof that your optimization work caused those changes.

Identify gaps against competitors. Does a competitor keep appearing in answers to "best [product or service]"? That is useful evidence. The audit shows which brands are recommended for the queries you want to be visible on.

Prioritize actions. You're visible on Perplexity but absent from ChatGPT? The audit helps identify platforms where your brand has low visibility.

Demonstrate value. For agencies and consultants, the audit is a sales and proof tool. Showing a prospect their (non-)visibility creates immediate awareness.

Indicators to track

An AI visibility audit can track four main groups of metrics.

IndicatorHow to read it in the observed scope
Mention rateShare of observed units in which the brand is explicitly mentioned
Citation positionRank in the displayed list of cited sources, when one is shown
AI Share of VoiceShare of retained brand mentions captured by the brand
Mention sentimentPositive, neutral, or negative in context

Mention rate

The mention rate is the number of observed units in which the brand appears, divided by the total number of tested units. Define a unit as a query, model, response mode, and repetition where applicable. Record that denominator before comparing two collections.

A 0% rate means that no mention was observed in the tested scope. Interpret it alongside the query selection, models, and collection period.

Citation position

Citation position is the rank of a source in a displayed list of sources. It is not the same as a brand’s recommendation rank in the response text. Record both separately when they are available.

AI Share of Voice

Share of voice relates mentions of your brand to all brand mentions retained within the same scope. With repeated runs, the denominator remains the total observed mentions, not merely the number of queries. State the competitor list and counting rule before making comparisons.

Mention sentiment

A favorable, neutral, or negative mention is useful to an audit because each says something different about visibility and reputation. "X is often criticized for its customer service" is a negative mention: retain and qualify it rather than discarding it.

Practical tip: start with the mention rate, the simplest indicator to collect. Add others as your methodology becomes structured.

Two distinctions to document

Brand, source, and recommendation are not the same observation

A brand mention answers one question: does the brand appear in the response? A URL citation records an explicitly displayed source. Recommendation rank records the order in which a brand is proposed in the text. Store these observations separately: they do not share the same denominator or interpretation.

Response mode belongs in the scope

A response generated from a model’s native knowledge and a response generated with web search should not be combined without a label. AI Labs Audit offers a dual native and web-search mode from the Consultant plan upwards. Keeping these conditions separate allows comparison without automatically attributing a change to work done on the site.

A documented reference shows why scope matters: between September 2025 and June 2026, our barometer analysed 6,820 responses from 44 models across 104 complete audits for 65 brands, in native and web-search modes. These aggregated figures do not define a universal threshold; they show why measurement conditions must be retained.

The 5-step methodology

  1. Define the scope

    An audit without a precise framework produces unusable data. Before launching anything, answer these questions:

    • Which platforms? ChatGPT alone? ChatGPT + Claude + Perplexity + Gemini? The choice depends on your target audience.
    • Which queries? Brand queries ("what do you think of [brand]")? Industry queries ("best [product] for [use]")? Informational queries?
    • What depth? 20 queries for a quick audit? 100 for a complete analysis?
    • Which competitors? Identify 3 to 5 competitors to monitor in parallel.
  2. Build the query list

    The audit quality depends on the relevance of tested queries. Three categories to cover:

    • Brand queries: "What do you know about [brand]", "Reviews of [brand]", "[Brand] vs [competitor]"
    • Industry queries: "Best [product] in France", "Recommend a [service] for [use case]", "[Category] comparison 2025"
    • Problem queries: "How to solve [problem you address]", "Solution for [client pain point]"

    Thirty queries can be an explicit starting choice for an exploratory protocol when categories are balanced. That choice alone does not make the findings generalizable.

  3. Collect the data

    Collection can be manual or automated. Manually, you submit each query in an interface and record the observed responses. Automated collection queries an API. These modes do not measure the same experience: an API response is not a direct measurement of what a user sees in ChatGPT’s public interface. Keep them separate in analysis and reporting.

    Collection points of attention:

    • Use fresh sessions (no conversation history)
    • Record the date, time, interface or API, model, and settings used for every collection.
    • Keep collections with and without web search separate; they describe different response conditions.
    • Keep complete responses, not just presence/absence of mention
  4. Analyze the results

    Once data is collected, structure the analysis around several axes:

    • By platform: Compared performance on ChatGPT, Claude, Gemini, Perplexity
    • By query type: Gaps between brand queries and industry queries
    • By competitor: Who dominates on which queries?
    • Qualitative analysis: How are you described when mentioned?

    Cross dimensions to identify patterns: "Visible on ChatGPT for brand queries, absent on industry queries" is an actionable diagnosis.

  5. Formulate recommendations

    An audit only has value if it leads to concrete actions. For each finding, propose a recommendation:

    • Absence on industry queries : create AEO-optimized content for these queries
    • Dominant competitor : analyze their citation sources and replicate the strategy
    • Negative mentions : identify the source and work on reputation
    • Low Bing visibility is an observation to assess within its own scope. By itself, it does not demonstrate an effect on ChatGPT or Copilot.

Common mistakes to avoid

Mistake #1: Testing a scope that is too narrow for the decision at hand. Ten queries can support an exploratory observation, but their limits must be stated. Results vary with wording, context, and timing, so interpret them within the scope actually tested.

Mistake #2: Ignoring response variability. AI systems do not always give the same response to an identical query. Three repeats can be a protocol choice for exploring variability; keep every response rather than reducing it immediately to an average.

Mistake #3: Focusing only on ChatGPT. Depending on your audience, Claude or Perplexity may be more relevant. A single-platform audit gives a partial view.

Mistake #4: Neglecting qualitative analysis. Knowing you're mentioned isn't enough. How are you described? In what context? A mention in a list of "solutions to avoid" isn't a victory.

Mistake #5: Auditing once and forgetting. Responses can change with models, sources, and query wording. Choose a cadence tied to your decisions and repeat the same scope when comparison is useful.

Automating the audit

Manual collection quickly becomes time-consuming. Several approaches to automate:

Model provider APIs. Where access terms and available interfaces permit, you can script collection. An API response does not directly measure the response in the public ChatGPT, Claude, or Perplexity interface. Keep these observations separate and retain the configuration, complete responses, and any collection errors.

Specialized tools. A dedicated platform can centralize scenarios, responses, collection settings, and reports. Check which observations were collected through an API or an interface, and which response modes were actually compared.

Product walkthrough. To apply this methodology hands-on, follow our tutorial: run your first AI visibility audit (client profile, prompt generation, reading results).

Automation advantages:

  • A query volume suited to the scope
  • Audit repeatability
  • Precise temporal comparisons
  • Standardized client deliverables

Audit report template

An effective audit report follows a logical structure:

  • Executive summary : One-page synthesis: overall score, main findings, 3 priority recommendations
  • Methodology : Tested platforms, number of queries, audit period
  • Results by platform: mention rate, average source rank in citation lists, and average brand rank in recommendations in the response text, reported separately, with response examples. For each average, state the number of observations where a rank is available; absence is not a rank of zero.
  • Competitive analysis : Comparison with identified competitors
  • Qualitative analysis : Mention sentiment, context, significant verbatims
  • Recommendations : Priority actions with effort level and expected impact
  • Appendices : Detail of tested queries and complete responses

Adapt the level of detail to your audience. A decision-maker wants the executive summary. A marketing manager wants detailed recommendations. A technical GEO/AEO specialist wants the full appendices.

Frequency and tracking

A one-time audit gives a snapshot. Regular tracking shows trends.

Initial audit: Define a reference protocol suited to the scope: platforms, queries, modes, repetitions, and competitors.

Follow-up audits: Reuse the same parameters when you need to compare collections. An observed change does not by itself establish its cause.

Targeted audits: A smaller set of priority queries can flag a variation to investigate. It does not replace the reference protocol.

Document each audit in a standardized format to facilitate temporal comparisons.

Key takeaways

AI visibility auditing is an emerging discipline, but the methodological fundamentals exist.

Define a clear scope before starting. Build a query list representative of your activity. Collect data systematically. Analyze across multiple dimensions. Transform findings into actionable recommendations.

Automate repeatable tasks to save time and keep collection consistent. Use regular follow-up audits to track your AI visibility rather than relying on a single snapshot.

Companies that master this methodology have an advantage: they see what others ignore, and can act accordingly.

Platform-specific audit guides

Questions and answers

What is an AI visibility audit?

An AI visibility audit is a systematic analysis of a brand's presence in the responses of conversational AI models such as ChatGPT, Claude, Gemini, and Perplexity.

What metrics does an AI visibility audit measure?

The main metrics are: mention rate, citation position, sentiment, and Share of Voice relative to competitors.

How do you conduct an AI visibility audit?

Define your target prompts, query multiple AI models, and analyse the responses to detect mentions of your brand.

How often should you conduct an AI visibility audit?

A monthly audit is recommended. Conduct a more in-depth audit each quarter to refine your strategy.

What tools should you use for an AI visibility audit?

Use a dedicated platform like AI Labs Audit that automatically queries 100+ AI models.

About the author

Davy Abderrahman

Founder of

As a specialist in visibility in AI answers (AEO/GEO/LLMO), I help agencies and consultants measure and analyze their clients’ presence on ChatGPT, Claude, Gemini, Perplexity and other AI answer engines.

AEO GEO LLMO AI visibility AI audits
How do your brands appear in AI answers?

Agencies, consultants and in-house teams: compare the answers generated for your own scenarios, languages and selected models. Our infrastructure, your scenarios, your expertise.

Was this article helpful?

- (0 votes)