You measure AI visibility in three layers: mention rate (how often an AI model names you), share of voice (how you compare to your competitors) and AI referral traffic (what it ultimately delivers). That split is necessary because it otherwise looks unmeasurable: AI answers do not sit in any dashboard, they change every time, and no classic SEO tool reveals them. It is the measurable side of GEO, the broader field of getting cited by AI. This guide explains that KPI structure, backed by research, and shows how to start today.
Why measuring AI visibility is urgent now
Because a large share of searches ends without a click, whether an AI answer names your brand helps decide whether you are considered at all, and no classic SEO dashboard shows it. Research firm Gartner has described generative AI since 2024 as "substitute answer engines": systems that take over searches that used to happen in Google. The direction is clear.
The figures support that, though they come largely from international (mostly US) research and are indicative for any market, not exact:
- According to Bain & Company (February 2025, survey fielded December 2024), roughly 80% of consumers rely on zero-click results in at least 40% of their searches, and around 60% of searches end without a click through to another website.
- Google AI Overviews reach around 2 billion users per month, according to TechCrunch (July 2025).
- Click-through to a regular search result fell from 15% to 8% when an AI summary appeared, in US data from the Pew Research Center (July 2025), and Ahrefs measured a 58% lower click-through rate for the top organic result when an AI Overview is present. The full picture is in what GEO is.
In short: a large share of your audience gets its answer directly from an AI, without ever clicking a link. Whether your brand is in that answer helps decide whether you still exist in the decision. And you can only know that if you measure it. For the wider context, read the difference between GEO and SEO.
What makes AI visibility so hard to measure?
AI visibility is hard to measure because AI answers are non-deterministic: the same question returns a different answer every time. This is the single most important insight in the whole discipline, and the reason a one-off check is worthless.
A 2026 study (Don't Measure Once) put identical prompts to the same AI search engine repeatedly within a 24-hour window. The cited sources overlapped only 32 to 43% between runs (Jaccard similarity). In other words: ask once today whether ChatGPT names your brand, and the answer can already be substantially different later the same day. From one day to the next, only about 35% of the cited sources overlapped.
The practical consequences are profound:
- A single measurement says nothing. Your AI visibility is not a fixed position but a probability distribution. You only measure it reliably by asking the same prompt repeatedly.
- Brand mentions are more stable than individual source URLs. In the same study, around 65% of cited sources changed from one day to the next, while brand sets stayed more stable (45 to 59% overlap). So steer on brand mentions first, not on individual links.
This is exactly why automated, repeated monitoring matters. Veesie measures every prompt weekly. Claude and Gemini, the noisiest sources, are asked up to three times per question with the majority answer counting; ChatGPT, Perplexity and Google AI Overviews are asked once per measurement. You read the trend over several weeks. In Veesie's own test on one site (12 questions, five runs, all five sources) the result stayed between 84 and 88 while 23% of the cells flipped between runs; according to a calculation on those five runs it falls to about 8% with a majority of three measurements, a lower bound. In practice Veesie repeats only Claude and Gemini, so the gain is smaller. See what Veesie measures per source.
It also reframes what you should measure. Rand Fishkin puts the warning bluntly:
Measuring your brand's presence in AI answers with precision is a fool's errand.
Rand Fishkin, co-founder/CEO of SparkToro, in SparkToro (January 2026)
Measuring is not pointless, though. Chasing a precise ranking position inside an AI answer is the illusion, because that number is unstable from one run to the next. What does hold up, and what Fishkin himself calls a reasonable metric, is your visibility percentage across dozens to hundreds of prompt runs. That is the mention rate you measure repeatedly.
The three layers of measuring AI visibility
The most reliable way to measure AI visibility is a layered approach, because no single tool sees the full picture. Marketing experts converge on three layers that together answer the question: are you named, how do you compare to competitors, and what does it return? (Search Engine Land, October 2025).
Layer 1: Mention rate, are you named at all?
Mention rate is the base metric: in what percentage of relevant AI answers does your brand appear? This is the floor of AI visibility. If you are not in there, there is nothing to optimise.
Make an important distinction here that tools like Semrush also use (AI SEO metrics): a mention is a textual reference to your brand, a citation is a reference where the model cites your domain as a source. Both count, but a citation also drives traffic.
Measure your mention rate across a fixed, representative set of search queries. A broad set of high-intent prompts gives a more reliable picture of your market than a handful of one-off checks. Veesie measures 20 to 50 questions, depending on the plan (Starter 20, Plus 30, Pro 50 shared across two websites; the trial has 15), so pick the questions your buyers actually ask. For that, Veesie provides ready-made question templates per sector and language, plus suggestions from a scan of your website, as a starting point for your own set of questions.
Layer 2: Share of voice, how do you compare to competitors?
Share of voice measures your slice of the AI answers relative to your competitors. A 30% mention rate sounds thin, but if the market leader sits at 35%, you are in a strong spot. Without a competitor comparison, your number lacks context.
Good share-of-voice measurements are position-weighted: being named near the front of the answer counts more heavily than a mention right at the end. Note: these scores are not yet an industry standard. Every provider calculates them slightly differently, so compare your own evolution and your relative position, not absolute numbers across tools. Veesie's competitor tracking shows which competitors are named more often than you, and which are gaining or losing ground in your niche.
Layer 3: AI referral traffic and conversion, what does it return?
The bottom layer measures what your AI visibility delivers in concrete terms: visitors and inquiries. Since 2026, Google Analytics 4 has a built-in "AI Assistant" channel that recognises traffic from AI assistants such as ChatGPT, Gemini and Claude via the referring domain.
Two caveats are crucial:
- Underestimation. A lot of AI traffic arrives without a referrer, for example from apps or copied links, and lands in the Direct channel. Your real AI traffic is therefore higher than GA4 shows.
- AI Overviews do not count. The AI Assistant channel captures chatbots, not Google AI Overviews or AI Mode. For those, Search Console's Generative AI performance report shows impressions over time, by page, country and device. Google says it has rolled the report out to all websites worldwide as of 31 August 2026, but a site sees no data if it has too few impressions in generative AI features on Google Search or has excluded itself from them.
So complement that measurement with your branded search in Google Search Console. The typical pattern has two steps: someone discovers your brand in an AI answer and then searches directly for your name. A rise in branded queries is often the first tangible effect of better AI visibility.
Sentiment, model coverage and position: the extra dimensions
Beyond presence, you measure three quality dimensions that determine how valuable a mention is. Whether you are named is the first question, but not the only one.
- Sentiment: are you described positively, neutrally or negatively? A negative mention can hurt rather than help.
- Model coverage: across how many of the major models (ChatGPT, Claude, Gemini, Perplexity) are you named? Visibility in one model is fragile.
- Position in the answer: where do you rank when an answer lists several brands? The academic research into Generative Engine Optimization (Princeton, IIT Delhi and others, KDD 2024) explicitly weights visibility based on where your citation sits in the answer.
Veesie's GEO Score bundles these dimensions into a single number: mention rate counts for 50%, sentiment 25%, model coverage 15% and position 10%. So at a glance you see whether a rise comes from more mentions or from a better position.
How often should you measure AI visibility?
More often than a human can keep up by hand. Because AI answers differ every time, a single measurement is unreliable. The same research into non-determinism recommends at least seven measurements per prompt per day for a stable read on your brand mentions, and at least eight if you also want to track the cited sources.
Do the math: 300 prompts, five AI sources, seven repetitions a day. That is more than 10,000 measurements a day, every day again. Doing that by hand is impossible, so you need automation. Veesie takes a lighter route: it measures weekly and asks Claude and Gemini up to three times per question, so a single week still carries some noise. A weekly report gives you a trend over time instead of a random snapshot, and the trend is what you steer on.
What tools and methods exist?
There are broadly two ways to measure AI visibility, each with its limits.
Testing prompts by hand. Free and low-threshold: you put your questions into ChatGPT or Perplexity yourself and note whether your brand appears. The drawback is fundamental: it is a snapshot, without history, without competitor comparison and without repetition. Because of the variation between runs, you cannot draw reliable conclusions from it.
Automated trackers. AI visibility tracking tools like Veesie, Semrush, Otterly, Profound and SE Ranking send your prompts systematically and repeatedly to multiple models, and bundle the results into scores and trends. This kind of AI search monitoring runs on a fixed schedule, so the variation between runs is averaged out instead of misleading you. Important nuance: every provider uses its own, non-interchangeable measurement method. So compare your evolution within one tool, not absolute scores across tools.
Veesie measures per language (English, Dutch and French separately, set per question), up to five AI sources (all five from the Plus plan), and weekly. It also publishes a GEO benchmark with sector averages for Belgium and the Netherlands, a snapshot, for marketers and agencies who want to compare their score.
Start measuring in four steps
Measuring AI visibility does not have to be complicated. With a structured approach you are up and running quickly:
- Assemble your prompt set. Collect the questions your audience actually asks: definition, comparison and recommendation questions. Aim for a representative set of high-intent prompts.
- Measure repeatedly across multiple models. Put every prompt to ChatGPT, Claude, Gemini and Perplexity repeatedly, so the variation is averaged out. Each fetches sources differently, as how to appear in AI chatbots and ChatGPT SEO explain.
- Track all three layers. Mention rate, share of voice and AI referral traffic together, not one isolated number.
- Report weekly and adjust. Compare your trend, not your snapshot, and tie every optimisation to the evolution.
Concrete optimisations that raise your score are in 8 concrete tips to improve your GEO Score. If you stay invisible despite everything, first check your technical AI-Readiness, because what AI cannot read, it cannot measure. To see which myths about GEO you can let go, read 8 GEO myths marketers still believe. And to tick off your technical foundation, use the GEO audit checklist. Want your baseline straight away? Start the free 7-day Veesie trial and set up your first measurement.
Conclusion: AI visibility can be measured with a different method
AI visibility feels unmeasurable because it does not appear in the familiar SEO dashboards and because the answers move. But once you break it into mention rate, share of voice and AI referral traffic, and you measure repeatedly instead of once, it becomes just as steerable as classic SEO.
The difference between brands that appear in AI answers and brands that go missing lies in measuring, understanding and adjusting, week after week. Veesie covers mention rate and competitor comparison for marketers and agencies and, from the Pro plan, AI referral traffic through its GA4 link, so every decision rests on a measurement.