How to measure your AI visibility

In the news: Since June 2026, Google Search Console has offered a dedicated Generative AI report, the first official view of how often your pages appear inside AI Overviews and AI Mode, though for now it counts impressions only and shows no click data (Google Search Central, June 2026).

If you do not know whether an AI assistant names your brand, you are optimising blind. And you cannot fix a number you have never looked at. Before rewriting a single page or chasing a single citation, there is a step almost everyone skips: measuring where you stand today, engine by engine. The good news is that a serious first measurement takes a spreadsheet and an afternoon, not a big budget. This guide walks through exactly how to do it, and how to read the numbers once you have them.

What does measuring AI visibility actually mean?

Measuring your AI visibility means quantifying how often, and how favourably, AI assistants name your brand or cite your content when people ask questions in your market. You track mentions, position, sentiment and the sources the model quotes, one engine at a time.

For two decades, being visible meant one thing: ranking in Google's list of blue links. That world still exists, but a growing share of questions now gets answered by an assistant that writes a paragraph and names a few brands inside it. Your customers ask ChatGPT which supplier to trust, or ask Perplexity to compare two options, and the model answers with names, sometimes yours, often not. AI visibility is simply how present you are inside those answers, and like any performance signal, it can be counted.

What makes a piece of content likely to be picked up is not a mystery. The academic work that named the field, Generative Engine Optimization, formalised in 2023 by researchers from Princeton, Georgia Tech, the Allen Institute and IIT Delhi, found that adding named sources, quantified statistics and direct citations measurably increases how often a passage is reused in a generated answer (Aggarwal et al., arXiv). That matters here for one reason: if inclusion follows measurable factors, then your inclusion can be measured too, and improved. If you want the underlying concept first, we defined what an AI citation is in a dedicated explainer.

Why it is measured differently from an SEO rank

Because AI answers are probabilistic. The same question can name different brands on two consecutive runs, so a single query is an anecdote, not a measurement. You measure a rate across many runs and many engines, never a fixed position.

A Google ranking is stable: search the same term twice and, personalisation aside, you get the same order. Generative models do not work that way. They sample their answer rather than return a fixed list, which means two identical prompts can surface two different sets of brands minutes apart. Treat one lucky answer as proof you are visible and you will draw the wrong conclusion. The unit of measurement is a rate: out of twenty runs of a prompt, how many named you.

The second difference is fragmentation. There is no single scoreboard. A strong presence in ChatGPT tells you nothing about Perplexity, Gemini or Google AI Overviews, because each engine draws on different sources and indexes. Google's own Search Console now reports impressions when your pages appear inside AI Overviews and AI Mode, which is a real step forward, but it covers Google's surfaces only, and for now it shows impressions without clicks (Google Search Central, June 2026). Measure one engine and generalise, and you are optimising blind on all the others.

The metrics that actually matter

Five metrics carry most of the signal: mention rate, share of voice, average position, sentiment and cited sources. Together they tell you not just whether you appear, but how favourably and against whom.

Raw impressions are a start, but they flatten out what actually drives a buying decision inside an AI answer. These are the signals worth logging from the beginning.

The last row is the most actionable. When you know which pages an engine quotes to build its answer, you know exactly where to earn a mention: on those source pages, or on content strong enough to displace them. That is where measurement turns into a to-do list rather than a scoreboard. It is also the bridge between AI visibility and classic SEO, a link we unpack in GEO vs SEO.

The method in 6 steps to measure your AI visibility

Six steps, in order: build a prompt set, pick the engines that matter, run each prompt several times, record the right signals, turn raw runs into metrics, then cross-check and repeat. No tool is required to start.

1. Build a prompt set. List 20 to 50 real questions your customers ask an assistant, from broad category queries ("best X for a small business") to specific buying questions. This prompt set is your measuring stick, so write it the way a customer would type, not the way a marketer would.
2. Pick the engines that matter. Choose the surfaces your buyers actually use: ChatGPT, Perplexity, Gemini, Microsoft Copilot, and Google AI Overviews or AI Mode. You do not need all of them, you need the ones your market lives in. If you sell to people who ask ChatGPT and Perplexity, start there.
3. Run each prompt several times. Because answers are probabilistic, run every prompt 3 to 5 times per engine, ideally in a fresh session with no memory. One run is an anecdote; several runs give you a rate you can trust.
4. Record the right signals. For each run, log whether your brand is named, in what position, with what sentiment, which competitors show up, and which sources the model cites. A simple spreadsheet with one row per run is enough to start.
5. Turn raw runs into metrics. Aggregate your rows into a mention rate, a share of voice against competitors and a sentiment split. These few numbers, per engine, are your baseline.
6. Cross-check and repeat. Compare your findings with Google Search Console impressions for AI features, then rerun the whole measurement on a regular cadence, since models and indexes keep changing.

This is the sequence Cicero Studio runs before we touch a client's content, and it sums up our craft: GEO audit, editorial production and automated semantic meshing. Agency-quality work, software-grade productivity. The measurement is not the deliverable, it is the map that tells us where the deliverable should go.

A spreadsheet or a tool: what to use

Start with a spreadsheet and the free engines to learn what to look for. Graduate to a dedicated tracker once manual runs stop scaling, when you need repeatable measurement across many engines and prompts, week after week.

There is no shame in a manual baseline, and a lot to learn from it: running the prompts yourself teaches you how each engine phrases things and where your competitors keep appearing. But a person can only rerun so many prompts across so many engines before it becomes a full-time job. That is when a tool earns its place. Dedicated trackers such as Ahrefs Brand Radar, Semrush and Profound automate the runs across ChatGPT, Perplexity, Gemini and the Google surfaces, and turn them into trend lines you can watch over time.

Google's own tooling sits alongside them rather than replacing them. The Search Console Generative AI report gives you official impression data for AI Overviews and AI Mode, straight from the source, but only for Google's surfaces, and it is silent on the AI assistants people use off Google entirely (Google Search Central, AI features documentation). The practical setup for most businesses is both: Search Console for the Google side, a tracker or a disciplined spreadsheet for everything else. If you are weighing options, we compared the best SEO and GEO tools in detail.

The mistakes that distort your numbers

Three mistakes recur: measuring a single run, measuring a single engine, and treating mention count as the only metric while ignoring sentiment and the buyer journey. Each one produces a number that looks precise and means little.

The first is measuring once. A single flattering answer feels like proof, but generative variance means the next run may drop you entirely. Without repetition you are reading noise, not signal. The second is measuring one engine and declaring victory: being cited well by Perplexity says nothing about how Gemini or AI Overviews treat you, and buyers rarely stay on a single assistant. The third is the vanity trap, counting mentions while ignoring how you are described. Being named with a lukewarm or negative framing can hurt more than not being named at all, which is why sentiment belongs in the measurement from day one. Deciding what to fix first, once you have the numbers, is a topic in itself, and we lay out a scorecard approach in our GEO audit method.

What measuring will not tell you

For the sake of honesty, and because this is exactly the kind of transparency both readers and AI reward, here are the limits to keep in mind before you start counting.

Measurement is worth doing precisely because it is imperfect: it replaces a guess with a range, and a range is enough to act on. What it cannot do is turn a thin, undifferentiated brand into a cited one. That remains, first and last, a content job.

Alexis Dollé

CEO & Founder of Cicero Studio

Growth and SEO content strategist, I founded Cicéro to help businesses build lasting organic visibility, on Google and in AI-generated answers alike. We measure before we produce, because a number you can see is the only honest starting point.

LinkedIn →

Go further

We document our approach in public, and that is our best proof: the 518 articles published on cicero.studio (272 FR, 246 EN) are there to be read and challenged. Here are the most useful resources to move from measuring to acting.