Measuring your AI visibility: what each tool actually sees
Three free sources measure your visibility in AI answers at the source, rather than by replaying questions: the AI report in Bing Webmaster Tools, Topic Insights in Microsoft Clarity, and the Generative AI report in Search Console. Together they cover most of what is measurable today. A paid tool earns its place on one precise need: tracking several models in a single table, over time.
- No source covers everything. Bing sees Copilot, Search Console sees Google surfaces, and neither one sees ChatGPT.
- A score built on replayed questions moves on its own: only 25.6% of cited domains overlap between two ChatGPT reasoning modes on identical prompts.
- The Search Console Generative AI report gives impressions, pages, countries and devices. It gives no queries and no clicks.
- Your server logs are the only source you own, and the only one that shows what a model came to read at the moment it answered.
- A paid tool becomes legitimate when comparing models over time turns into a monthly decision rather than a curiosity.
On this page
AI visibility measurement
AI visibility measurement covers the methods used to find out whether a company is cited in the answers produced by assistants and by the generated summaries of search engines. Two families stand apart. The first asks models a set of questions and counts the citations obtained: it produces a score, and that score depends on the questions chosen. The second reads what the platforms themselves report about real citations: it covers only part of the surfaces, but it rests on no reconstruction.
Why an AI visibility score moves when nothing changed on your site
Most tools sold between $200 and $500 a month work the same way. They pick a list of questions, put them to the models at regular intervals, record the brands and domains cited, and derive a visibility score. The method is transparent and it is defensible. Its limit comes from a property of the models themselves: the same question asked twice does not return the same sources.
Semrush measured that gap with strategist Kevin Indig and published the result on June 30, 2026. One hundred prompts, spread across twenty buyer journeys in four categories, were put to ChatGPT twice each: once in minimal reasoning, once in high reasoning. Between the two passes, 25.6% of cited domains overlap. Close to three cited sources out of four change, for an identical question and a gap of a few seconds.
The scope of that study is worth naming, because it is almost always dropped when the figure travels. It covers ChatGPT, one hundred prompts, four categories. It says nothing about Google, nothing about Perplexity, and nothing about your market. What it establishes solidly is that a score built on replayed questions measures a distribution rather than a state.
The practical consequence is simple. A few points of variation from one month to the next tell you nothing about what you changed. What stays readable in these tools is the trend over several months and the relative comparison with the companies cited next to you. That is already useful, as long as the number is not presented to a leadership team as a performance measure.
An AI visibility score is neither wrong nor useless. It is unstable by construction, which rules out reading it the way you read an average position in Search Console. It reads like a poll: the margin of error is part of the result.
The three first-party sources, and what each one sees
The three sources below reconstruct nothing. They report citations actually served, counted by the platform that served them. They are free, they require domain verification, and they remain underused.
Bing Webmaster Tools, the AI report
Microsoft opened its AI report on February 11, 2026. On March 24, the tool began connecting the grounding query, the one a model formulates to go and find information, to the page on your site that was cited. On June 16, four capabilities were added, including the measure called Citation Share.
Citation Share is the most interesting of the four, because it gives a share rather than a count. The formula Microsoft publishes divides your citations on a query by the total citations displayed for that same query. Three citations out of the ten an answer drew on give you a 30% share. It is the first time a major platform has published a citation share rather than a volume, and it is what tells you whether you are gaining ground or the whole market is.
Microsoft itself calls the measure observational. The wording comes from its own documentation and it is worth keeping: the figure describes what was observed on Microsoft surfaces, it predicts nothing.
What the report covers: Copilot, Bing summaries, and the experiences built on the Bing index. What it does not cover: ChatGPT, Perplexity, and Google's generative surfaces.
Microsoft Clarity, Topic Insights
Clarity added Topic Insights on July 9, 2026. The feature groups your citations by topic rather than by query, and it shows the domains cited alongside yours on those same topics. It is the most direct free answer to the question paid tools charge for: who else is in the answer.
Microsoft states that the feature is meant for directional monitoring. In other words, the volumes serve to compare topics against each other rather than to be summed. Two further limits matter before you build a monthly report on it: the reading remains that of Microsoft surfaces, since grounding runs through their infrastructure, and the tool applies a quota of reports per project per week, which rules out treating it as a daily dashboard.
Search Console, the Generative AI report
Google announced the report in June 2026 and completed its worldwide rollout on August 31, 2026. It is therefore out of beta, and available on every verified property, including yours.
The report covers AI Overviews and AI Mode in Search, plus generative features in Discover. It gives impressions, the pages receiving them, countries, devices and dates. It gives no query and no click, and that absence is structural: you will know which pages of your site appear in Google's generative surfaces, never on which questions.
One check to run on first opening: in the property settings, the control that excludes generative results from the main performance report must not be enabled if you want to compare the two curves. Many teams read a decline in the classic report without knowing that the setting produced it.
What that report supports as a conclusion, and what it does not, is covered in our note on measuring AI answers in Search Console.
What each source covers, and what nobody covers
The table below is the reason this article exists. It answers the only question that matters before choosing: what will I not see.
| Surface | Bing Webmaster Tools | Clarity | Search Console | Your logs |
|---|---|---|---|---|
| Copilot and Bing summaries | Yes | Yes | No | Partly |
| Google AI Overviews and AI Mode | No | No | Impressions only | Partly |
| ChatGPT | No | No | No | Grounding requests |
| Perplexity | No | No | No | Grounding requests |
| Gemini outside Search | No | No | No | Partly |
Two readings follow. First: the richest free tool on the market, Microsoft's, sees neither ChatGPT nor Google. Second: the only row where ChatGPT appears is your own server logs, which is why the next section exists.
The fourth source is yours: your server logs
Every time an assistant goes to the web to build an answer, it sends a request to your server, and that request leaves a line. It carries an identifiable agent name, a page address and a timestamp. Nobody can take that access away from you: it is your infrastructure.
The distinction to hold is between two visits of a different nature. A training crawler passes to build a corpus, with no connection to a question asked at that moment. A grounding crawler passes because someone has just asked a question and the model is looking for material to answer it. The second visit is a signal of real demand, and it is the only trace available for the surfaces that publish no report.
What logs give that nothing else gives: the exact pages models come to read, how often, and the moments when that frequency changes. What they do not give: the citation. A visit is not an answer, and a page heavily read by crawlers may never be cited.
Setup, the agents to recognize and the trade-off on which ones to let through are covered in our analysis on blocking or allowing AI crawlers.
Confusing a crawler visit with a citation is the most common error right now, and it shows up in agency reports. A rising chart of crawler visits presents well, and it says nothing about your presence in answers. Both are measurable, separately.
When a paid tool is justified, and which one
Free sources are enough as long as the question is: do we appear. They stop being enough when the question becomes: are we gaining ground against the companies cited next to us, on a defined set of questions, across several models at once. Three situations cross that line.
Your market is decided across several models
Your buyers use ChatGPT and Google, and the question of where you are cited comes up every month. No free source brings the two into a single table.
Comparison with other vendors becomes the point
Knowing you are cited is no longer enough: you need to know who else is, on which questions, and since when. That is a competitive reading, and it is what tracking tools do best.
You track a stable set of questions over time
A single question cannot be measured. A set of fifty to two hundred questions, recorded every month, gives a trend that survives the instability of any one run.
The tool Falia tested and kept is Peec AI. What it does better than the free sources comes down to three points: tracking several models in a single table, comparison with the domains cited alongside yours, and tracking over time on a defined set of questions rather than an isolated query.
The limit that follows applies to this tool as to every tool in its category, including those showing a more flattering score. The measurement rests on replayed questions. It gives a trend and a relative comparison, never an absolute truth. That is why we cross it with first-party sources rather than reading it alone.
One element makes it possible to verify the method rather than take it on trust: the team behind Peec publishes its research. The analysis of 1,056,727 citations released in March 2026, which we cite elsewhere in our articles, sets out its sample, its models and its definitions. A vendor who publishes its data accepts being contradicted, and that is the sturdiest trust criterion available in this market.
On the amount, a rule of proportion beats a threshold. A $300 monthly subscription is one fifth of a $1,500 monthly content budget: at that level, you are funding measurement at the expense of production. The same subscription against an $8,000 monthly organic budget is under 4%, and the question becomes legitimate.
Which combination fits your situation
The choice depends on two variables: the number of surfaces where your buyers search, and how often the question comes up internally.
| Your situation | What is enough |
|---|---|
| Under a hundred pages, local market, question asked twice a year | The three free sources, checked quarterly |
| Competitive market, several models, question asked monthly | The three free sources plus a tracking tool such as Peec |
| Large catalog or high page count, technical team in place | The three free sources plus regular server log reading |
A vendor offering you an AI visibility score without naming the surfaces covered is selling a number. One who starts by having you verify your domain in Bing and Search Console is working with what exists.
Deciding what to measure before buying a tool is exactly the kind of trade-off a 90-minute consultation settles, on your data and your market.
On an engagement, we look at these sources in a fixed order, and the order has a reason. We verify the domain in Bing Webmaster Tools and Search Console first, because the data does not backfill and every week lost is a week of history lost. We then read the server logs over thirty days, to find out which models already come and on which pages. Clarity goes third, for the reading by topic. The paid tool comes last, once the first three have shown what they do not cover. What stays in-house is the list of questions your buyers actually ask: nobody on the outside has it, and it decides the value of everything else.
Checking your measurement setup
The general framework for visibility in answer engines is set out in optimizing for answer engines. What actually separates classic search from this work is covered in SEO and GEO, what changes and what does not. Building these measures into a dashboard that survives is covered in the SEO indicators worth tracking.
Installing this measurement and acting on it is at the heart of the Strengthen your visibility in AI answers goal.
Already running a marketing team? See how we plug in as reinforcement on answer engine optimization.
Frequently asked questions about measuring AI visibility
How do I know if my company is cited by ChatGPT?
No free source reports it directly. The only trace available sits in your server logs, as the grounding requests sent by OpenAI agents when someone asks a question. For regular tracking rather than a one-off check, you need a paid tracking tool, accepting that its measurement rests on replayed questions.
Are prompt tracking tools worth the money?
They are worth it when comparing models over time has become a monthly decision, and when the subscription stays a small fraction of your organic budget. They are not worth it when they serve to produce a number for a report, because that number moves from one run to the next without anything changing on your side.
Does the Search Console Generative AI report give queries?
No. It gives impressions, pages, countries, devices and dates. It gives no query and no click. You will know which pages appear in Google's generative surfaces, never on which questions they appear.
Why does my AI visibility score change every week?
Because the measurement rests on replayed questions and models do not cite the same sources from one run to the next. Semrush measured 25.6% overlap between two ChatGPT reasoning modes on identical prompts. The trend over several months stays readable, a week-to-week variation does not.
Should you verify your domain in Bing when you do not target Bing?
Yes, and it is the first thing to do. The Bing Webmaster Tools AI report is currently the most detailed free source on citations, including citation share per query. The data does not backfill: a domain verified today starts accumulating history today.
Do AI crawler visits prove that I am cited?
No. A crawler visit shows that a model came to read a page, not that it cited it in an answer. Both are measured separately, and a report presenting a crawler visit curve as proof of visibility is mixing two distinct things.
- Semrush and Kevin Indig, Only 25% of cited sources overlap between ChatGPT's different reasoning modes, 100 prompts across 20 buyer journeys, published June 30, 2026.
- Google Search Central, Introducing Search generative AI performance reports in Search Console, June 2026, worldwide rollout completed August 31, 2026.
- Search Console Help, Generative AI performance report, accessed September 2026.
- Microsoft Bing Webmaster Tools, AI Performance report released February 11, 2026, grounding query to page mapping since March 24, 2026, Citation Share since June 16, 2026.
- Microsoft Clarity, Introducing Topic Insights, July 9, 2026.
- Zyppy, AI citation ranking factors, 23 factors scored, May 7, 2026.
- Peec AI, AI visibility tracking tool, and its public research on the content types most cited by LLMs, 1,056,727 citations analyzed, March 2026.
Geneviève puts the strategy for your engagement into action. She leads all our web development projects: Shopify, WordPress and the new ways of building a site with AI. She manages our team of developers and translates your business needs into technical language. She runs your organic search (SEO), your visibility in AI answers (GEO) and your site's conversion rate optimization (CRO). Her work is at the heart of three goals: Attract customers with SEO and AI, Improve your site's conversion, and Strengthen your visibility in AI answers. With Gabriel, she also builds the landing pages for your advertising campaigns. She writes mainly about SEO, AI visibility and web design.
About Falia →
SEO
Paid advertising
Conversion
AI visibility
Web design