Key takeaways:
- No single method suffices. You cross three sources: Search Console for exposure, an LLM tracking tool for share of voice, a manual test in neutral conditions for control.
- Search Console measures impressions, not clicks, and groups AI Overview and AI Mode without separating them.
- Three indicators truly matter: trigger rate, citation, share of voice. A tool that only gives citation leaves you half-blind.
- Cited sources change weekly: AI Mode renews 56% of its sources every week (SISTRIX, 2026). Monthly tracking captures only noise.
You can rank first on Google and be invisible in its AI Overview. No tool hands you a single reliable figure: accurate measurement combines three methods, each with its blind spots.
Does ChatGPT recommend your brand?Measure your presence and identify the brands cited in your place. No credit card required.
What exactly should you measure in an AI Overview?
Three indicators, not one. Most business owners ask “do I appear?”, when that question covers only a third of the picture. Useful measurement answers three distinct questions.
- Trigger rate: on your strategic queries, does an AI Overview appear at all? If Google generates none on your topic, the rest is moot.
- Citation: when the AI Overview appears, is your site cited as a source? This is the question everyone asks first.
- Share of voice: among the cited sources, where do you stand against the named competitors? Appearing alone or appearing last behind five competitors are not the same.
A tool that only surfaces citation leaves you blind on the other two. Trigger rate tells you whether the surface exists for you. Share of voice tells you whether you win or lose it. Without those two bounds, an isolated citation score cannot be interpreted.
Does Search Console measure presence in AI Overviews?
Yes, since 3 June 2026, and it is the only first-party source. Google added a performance report dedicated to generative AI, which isolates for the first time a site’s visibility in the generative features of Search.
The report offers a “search” view, covering AI Overviews and AI Mode together, and a Discover view. It exposes a single metric, the impression, broken down by page, country, device and date, with granularity down to the hour.
The limit is at the heart of the subject. The report provides no clicks, no click-through rate, no position, no per-query breakdown. It measures exposure, the fact of being displayed, without saying how many visits that exposure generates. Google indicates it may add clicks later.
What Search Console gives you, and does not
- Gives: AI impressions per page, per country, per device, per date.
- Does not give: clicks, the AI Overview / AI Mode split, per-query detail, your share against competitors.
Two practical consequences. First, a high impression volume can coincide with falling traffic, since the user gets their answer without clicking. Second, because AI Overview and AI Mode are grouped, Search Console does not tell you which surface you perform on. For that, you need a second tool.
Are third-party tracking tools reliable?
They complement Search Console on what it ignores, provided you know what they actually measure. Two families exist: SEO suites that added an AI visibility module, and platforms dedicated to tracking in generative engines.
They do not all measure the same thing. Some track mentions, when your brand is named without a link. Others citations, when a link points to your site. Others share of voice or mention sentiment. The number of engines covered matters less than the fit between the metric and your goal.
One technical limit deserves your attention. Most of these tools query the models via their API, not via the engine actually used by the public. The gap with real experience is mechanical: an API response is not what a user sees in their browser.
Google adds a blunt warning in its documentation: no third-party tool has access to its internal ranking systems. A tracking tool observes produced responses; it does not read the algorithm. Any claim of “internal” metrics is marketing.
In my view, the right buying reflex is to invert the usual question. People compare engine counts first, when the real question is: does this tool measure trigger rate and share of voice, or only citation? A tool tracking ten engines but surfacing only raw presence is worth less than one tracking three engines with competitive share of voice. On the accounts I run, that last criterion separates useful tools from decorative dashboards.
Read also: How to Appear in Google AI Mode in 2026?
Is the manual test still worth anything?
Yes, as a control, never as an isolated measurement. Typing your strategic queries and observing whether your site is cited stays free, immediate and irreplaceable for your high-value queries. But a single test is worthless, because model responses are not deterministic: the same prompt, minutes apart, can cite you then ignore you.
A reliable test respects five conditions:
- Log out of your Google account or switch to private browsing.
- Neutralise location where possible.
- Vary the phrasing of a single query.
- Repeat each test several times.
- Renew the measurement weekly, not monthly.
The most common error is testing from your own account, logged in, in your usual location. The result is skewed by personalisation. In neutral conditions, a site judged absent often reappears across several responses.
How often should you measure your presence in AI Overviews?
Every week, because cited sources renew at a pace no classic SEO surface has ever known. This is the point most guides overlook, and it changes everything.
SISTRIX tracked 82,619 queries over 17 weeks, across three platforms and six countries including France. The finding is clear: each AI Mode response contains a “fixed core” of a few stable domains, present in 86% of cases, and a “carousel” of sources rotating around it. That carousel renews by 89% per week. Across all sources, AI Mode changes 56% each week, ChatGPT up to 74%.
“A visibility score reads as a trend, not a stable ranking.”
A second study, run by Stacker and Scrunch across more than 3 million citation events, measures an average half-life of 4.5 weeks for a cited domain: within a month, half your ChatGPT citations have vanished. ChatGPT cycles fastest (3.4 weeks), Perplexity is the stickiest (5.7 weeks).
The operational consequence is direct. A monthly reading on a surface that renews by half each week does not measure your presence, it captures a random snapshot. The right cadence is weekly, with a success metric based on presence over time, not on an isolated reading.
| Surface / engine | Source renewal | Recommended cadence |
|---|---|---|
| Google AI Mode | Around 56% per week | Weekly |
| ChatGPT | Up to 74% per week | Weekly |
| Google AI Overview | Stable for about half of queries | Weekly to fortnightly |
| Perplexity | Half-life of about 5.7 weeks | Fortnightly |
Sources: SISTRIX (17 weeks, 82,619 queries, 2026) and Stacker/Scrunch (3 million citation events, 2025-2026).
Which errors distort AI visibility measurement?
Four traps produce false conclusions, and all are avoidable.
Confusing exposure and traffic. Search Console counts impressions, not visits. A click from an AI Overview arrives with a google.com referrer and blends into organic traffic. GA4 does not isolate it. Its “AI Assistant” channel measures chatbot traffic like ChatGPT or Gemini, but not that of AI Overviews.
Taking GA4’s AI traffic for a total. Between 35 and 70% of AI-referred sessions arrive without a referrer and fall into direct traffic. Any figure obtained is a floor, never a total.
Reading a score as a fixed position. Model responses vary from one run to the next. A rising score indicates a trend, not a secured rank.
Confusing being cited with being recommended. This is the subtlest trap. Being cited says nothing about how your brand is presented. A Demand-Genius study, run on ChatGPT in a B2B context, observes that at constant visibility, a model can steer its recommendation towards a competitor depending on the angle of the question. A rising citation score can hide an unfavourable recommendation.
In my view, this last point will become the real subject of the next eighteen months. We measure presence today because it is measurable, but presence is only a prerequisite. The question that matters to a business owner is not “am I cited?”, it is “when a potential customer asks for the best solution, does the AI name me, and in what terms?”. The tools that learn to measure the meaning of the recommendation, not just its presence, will take the lead.
How do you build complete tracking today?
By crossing the three sources rather than seeking the single tool that does not exist. The French rollout of AI Overviews, live since 22 July 2026, opens a window to establish your baseline before these responses become widespread.
A tracking setup rests on four building blocks:
- Search Console for real exposure on Google’s side, exported weekly to build your impression history per page.
- An LLM tracking tool for share of voice and competitive comparison on your strategic prompts, across generative engines.
- Audience analytics for chatbot traffic, keeping in mind it underestimates the total.
- A weekly manual test in neutral conditions on your twenty to thirty priority queries.
On tracking generative engines, this is the need I set out to cover by co-founding Cockpyt AI, which today tracks your presence in ChatGPT, Perplexity and Gemini, with share of voice against the brands cited in your place. Dedicated tracking of Google’s AI Overviews and AI Mode is in development and will soon join those three engines.
Disclosure: I am a co-founder of Cockpyt AI. I mention it with first-hand knowledge of its current scope and limits, and I distinguish what the tool does today from what is coming.
Does ChatGPT recommend your brand?Measure your presence and identify the brands cited in your place. No credit card required.
Frequently asked questions
Does Search Console separate AI Overviews from AI Mode?
No. The AI performance report groups both surfaces in a single “search” view. It does not let you know which one you perform on. Only a third-party tracking tool, or a manual test surface by surface, provides that distinction.
Can you see the clicks generated by an AI Overview?
Not in the Search Console AI report, which exposes only impressions. In GA4, those clicks arrive with a google.com referrer and blend into organic traffic, with no native way to isolate them. Google has indicated it may add clicks to the report later.
Is a site not cited in an AI Overview really absent?
Not necessarily. Model responses vary on each run, and a single test from a logged-in account is skewed by personalisation. In neutral conditions and across several passes, a site judged absent often reappears.
Do you need a paid tool to measure AI presence?
No, to start. Search Console and the manual test in neutral conditions are free and cover exposure and control. A paid tool becomes useful for competitive share of voice and automating tracking across many queries.
Do Search Console AI impressions inflate my traffic?
No, they are exposures, not visits. A high impression volume can even coincide with a traffic drop, since the user gets their answer in the AI Overview without clicking through to your site.
When will French data become reliable?
The Search Console report started surfacing French data from the 22 July 2026 rollout, initially on a subset of queries. Allow a few weeks of hindsight for a usable history, which makes a baseline established now all the more valuable.
Primary sources
- Google Search Central, “Introducing Search generative AI performance reports in Search Console”, 3 June 2026, developers.google.com
- Google Search Central, “Optimizing your website for generative AI features on Google Search”, updated 10 July 2026, developers.google.com
- SISTRIX, “AI Citation drift: How stable are sources in AI search results?”, May 2026 (82,619 queries, 3 platforms, 6 countries including France, 17 weeks), sistrix.com
- Stacker & Scrunch, citation decay study, September 2025-March 2026 (over 3 million citation events; average half-life of 4.5 weeks)
- Demand-Genius, “Your AI Visibility Metrics Are Lying to You”, study on ChatGPT in a B2B context, demand-genius.com
