How to measure your visibility in AI

How to measure your visibility in AI

Measuring your visibility in AI means putting a panel of real questions to each engine (ChatGPT, Perplexity, Claude, Google AI Overviews), recording how often your business is cited, at what rank and in what terms, then repeating that measurement over time with dated screenshots. There is no official counter, the way there is a Search Console for Google: AI visibility is measured by sampling, not through a dashboard supplied by the providers. The good news is that a rigorous method is enough to turn a vague impression ("we are not being seen") into a figure tracked month after month. That is precisely what we do in our AI visibility diagnostic, and it is reproducible in-house with a little discipline.

Why this is harder than measuring SEO

In classic SEO, you have stable tools: a keyword's position on Google is measurable, archivable and relatively reproducible. AI answer engines break that logic on two counts. First, they are non-deterministic: for the same question, the answer varies by session, by the logged-in account, by geographic area and by date. You can be cited one day and absent the next without having changed anything. Second, there is no public, consultable index: no equivalent of Search Console tells you "you were cited 120 times this week".

The methodological consequence is clear. An isolated measurement means nothing. A single test that cites you does not prove you are visible, and a single test that ignores you does not prove the opposite. What matters is the frequency across a panel of questions and its trend over time. You do not measure a position, you measure a probability.

The three metrics that matter

A useful measurement rests on three complementary indicators, not on a single one.

• ·Citation frequency. Across a fixed panel of questions your clients actually ask, in what percentage of answers is your business named? This is the primary indicator. It is calculated engine by engine, because being cited by Perplexity says nothing about your presence in ChatGPT.
• ·Share of voice. When an answer cites players in your sector, do you appear, and alongside whom? Being absent when three competitors are named is a more serious signal than a single isolated non-citation. Share of voice situates your visibility relative to the market.
• ·Position and phrasing. Are you cited first or at the end of the answer? With a link, a brand mention, or just a passing allusion? The way the AI talks about you matters as much as the fact that it names you. An incorrect phrasing is sometimes worth a correction effort in its own right.

The concrete method, step by step

The measurement comes down to five reproducible steps, which you can hold in a simple spreadsheet.

1. Build the question panel. List fifteen to thirty real questions your prospects formulate before deciding, of the type "which solution for X" or "best company for Y in my area". This panel must stay stable from one month to the next, otherwise you are comparing different things.

2. Test under neutral conditions. Query each engine in private browsing, logged out of any account, to prevent your history from personalising the answer. Repeat each question several times, because the variability is real.

3. Capture and date. For each test, save a timestamped screenshot with the model used. Without a dated screenshot, a citation claim is not verifiable, and an unverifiable measurement is worthless.

4. Code the results. For each question and each engine, record whether you are cited, at what rank and in what terms. Aggregate into a citation rate per engine and an overall rate.

5. Repeat every month. Replay the same panel at regular intervals. It is the trend, upward or downward, that drives decisions, not a single day's snapshot.

This discipline turns an intuition into data. If you discover that you are never cited, the next step is to understand the cause, which we detail in why your business does not appear in ChatGPT.

The pitfalls to avoid

The first pitfall is the vanity metric. The traffic sent to your site by AI is interesting, but it massively underestimates your real visibility: an answer that cites your brand without a clickable link makes you visible without generating a measurable click. Measuring only the visits attributed to AI is seeing only the tip of the iceberg.

The second pitfall is the single test run on your own account. Logged into your session, the model's history and memory cite you more readily: you are measuring yourself in a flattering mirror. Private browsing and multiple draws are not details, they are the conditions for the measurement's validity.

The third pitfall is to expect everything from the site. A study published on arXiv in 2025 (reference 2509.08919) measures that between 73% and 92% of AI answer engine citations come from third-party sources, press, directories and communities, and not from the brand's own site. Measuring your visibility without looking at what others write about you amounts to ignoring the main lever. We explain this mechanic in which sources ChatGPT cites.

Spreadsheet, tool or support

For a small organisation, a rigorously kept spreadsheet is enough to start: a stable panel, dated screenshots, a monthly measurement. It is less comfortable than a tool, but it is honest and it costs nothing. AI visibility tracking tools are emerging and automate collection across large query volumes; they save time as soon as the panel becomes large or multilingual. Support is justified when the measurement needs to be tied to an action plan: understanding why a rate stagnates, prioritising the fixes, and tracking the effect over time. In every case, the value comes not from the tool but from regularity and traceability. To know where you stand today, our AI visibility diagnostic establishes a first measurement free of charge. If the subject is new to you, start with what is AEO.

Frequently asked questions

Is there a Search Console for ChatGPT or Perplexity?

No, there is no official equivalent of Search Console for AI answer engines. The providers do not offer a public dashboard showing how many times you were cited. AI visibility is therefore measured by sampling: a fixed panel of questions, tested under neutral conditions and repeated over time, with dated screenshots. Third-party tools automate this collection, but none replaces official data, because none exists.

How often should you measure your AI visibility?

A monthly measurement is a good rhythm for most businesses. Since AI answers evolve with model updates and your external authority, a monthly cadence captures the trends without over-engineering. What matters is not a single day's snapshot, variable by nature, but the trajectory over several months. Keep the same panel of questions from one measurement to the next, otherwise the comparisons lose their meaning.

Why test in private browsing and several times?

Because AI answer engines are non-deterministic and personalised. Logged into your account, the model's history and memory cite you more readily, which skews the measurement in a flattering direction. Private browsing neutralises this personalisation, and multiple draws on the same question capture the real variability. A measurement taken only once, on your own account, has no methodological value.

Is the traffic sent by AI enough to measure my visibility?

No, it is a vanity metric taken in isolation. An AI answer that cites your brand without a clickable link makes you visible at the moment of decision without generating a measurable click in your statistics. Measuring only the visits attributed to AI therefore greatly underestimates your real visibility. Citation frequency across a panel of questions remains the primary indicator; traffic is only a partial complement to it.