There is a measurement window that closes permanently once AI answers reach a market, and almost nobody uses it. Since 3 June 2026, Google Search Console has exposed your impressions inside AI Overviews, AI Mode and Discover — but with no clicks, no click-through rate and no queries. Here's what that report proves, what it can't, and how to measure the impact anyway. France, which flipped on 22 July, is the live case.
Why France is worth watching
On 22 July 2026, Google switched on AI Overviews and AI Mode in France. It was the last major market to go, held up for two years by the dispute with French press publishers over neighbouring rights — a fight that cost Google €500 million in 2021 and €250 million in March 2024.
That makes it the closest thing the industry has to a controlled experiment: a large, mature search market flipping late, with everyone watching, and rollout completing by the end of September. If you operate anywhere that already flipped, France is showing you in real time what your own numbers did before you were paying attention. If you operate somewhere still waiting, it's a rehearsal.
Either way the practical lesson is the same, and it has a deadline: the "before" period only exists until it doesn't.
What Google opened up
Until June 2026, measuring your presence in generated answers meant improvisation — screenshots, third-party tools, manual sampling. Google finally published its own numbers.
The report lives in Search Console under Performance, via the dedicated generative AI filter in the search appearance criteria. It covers three surfaces: AI Overviews, AI Mode, and generative features in Discover.
That's the good news. Now the gap between what people expect to find in it and what's actually there, which is considerable.
What the report gives — and what it doesn't
| Data | Available? | What it lets you do |
|---|---|---|
| Impressions in AI surfaces | Yes | See whether you're retained as a source, and how often |
| Cited pages (canonical URLs) | Yes | Identify which content the AI actually selects |
| Breakdown by country | Yes | Isolate one market — useful during a rollout |
| Breakdown by device | Yes | Compare mobile and desktop behaviour |
| Change over time | Yes | Track a trend, if you have history |
| Clicks from AI surfaces | No | — |
| Click-through rate | No | — |
| Average position | No | — |
| Triggering queries | No | — |
The omission matters. You learn where you appear, not whether it earns you anything. Google has said it is working on adding those metrics, with no published timeline. Until then the report answers "am I visible?" and not at all "is this working?".
A second, quieter limitation: the history is short. This data only exists from spring 2026. You cannot compare against "before AI" — that reference period was never recorded.
The interpretation trap
An AI impression does not mean the user saw your name.
It means your page was retained as a source for building an answer. Depending on the layout, the link to that source may be visible immediately or hidden behind a panel the user must expand — and the Pew Research Center study measures clicks on links inside a summary at just 1% of visits.
So: a rise in your AI impressions is an excellent signal that the model is selecting you, and that your AI search optimization work is landing. It is not a promise of traffic. Confusing the two leads to celebrating a rising curve while visits fall — a situation that is about to become very common.
The method, in four steps
Since Google withholds AI clicks, impact has to be measured by difference. Nothing complicated, but the order matters.
1. Record your baseline — now
This is the step with a deadline. Export from Search Console, page by page: clicks, impressions, CTR and average position, over a stable period (the last three months, avoiding your seasonal peaks). Put that file somewhere safe.
Why the urgency: rollouts are progressive and take weeks. France started on 22 July and completes by the end of September. Every week that passes contaminates a little more of your "before". Three months from now, without that file, you will not be able to separate an AI effect from ordinary seasonality — and you'll be left arguing from impressions.
2. Segment your pages by intent
A single traffic number says nothing, because the decline isn't uniform. Sort your pages into three families:
- Informational — guides, definitions, how-tos, advice articles. This is where it lands: the summary replaces the page.
- Transactional — pricing, booking, ordering, "near me". Barely touched: the user wants to arrive somewhere.
- Navigational — your brand name. Untouched.
This segmentation is worth more than any dashboard: it tells you in advance which pages to watch and which to ignore.
3. Compare like for like
Compare September to September, not September to July. A clothing shop comparing back-to-school against midsummer will conclude either disaster or miracle, neither related to AI. If you have more than a year of history, use it — it's your best protection against false causality.
4. Cross-reference with sessions and conversions
Always finish on the number that actually matters: organic sessions and, above all, calls, bookings or enquiries. A site can shed 30% of informational visits without losing a single customer — a common scenario for a local business. The reverse also happens, and that's the one that should alarm you.
What to watch monthly
Once the baseline is set, four indicators are enough:
- AI impressions, filtered to your market — your visibility inside generated answers.
- Cited pages — which content the AI keeps. Often surprising, and the best editorial guide you'll get: write more of what's already being selected.
- Organic CTR on informational pages — the first place the effect shows.
- Organic conversions — the only number that decides whether there's a problem.
And keep the manual habit: once a month, ask AI Mode the three questions your customers really ask, and note who gets cited. Search Console tells you that you appear; only the manual test shows you how.
Frequently asked questions
Where can I see if my site is cited in AI Overviews?
In Search Console, Performance report, via the generative AI filter. Announced by Google on 3 June 2026, covering AI Overviews, AI Mode and Discover, with breakdowns by page, country, device and date.
Does it show clicks and CTR?
No. It is confined to impressions, cited URLs and their breakdown by country, device and date. Clicks, CTR, average position and queries are unavailable for AI surfaces. Google says it is working on adding them, with no announced timeline.
Does an impression mean I was visibly cited?
Not necessarily. It means your page served as a source. The link may be shown immediately or hidden behind an expandable panel — and clicks on links inside a summary account for roughly 1% of visits according to Pew Research Center.
How do I measure impact without click data?
By comparison: a baseline recorded now, pages segmented by intent, like-for-like period comparison, then cross-referencing with sessions and conversions.
Do I need a paid AI citation tracking tool?
For a small business, not at first. Search Console plus a monthly manual test gives you most of the signal. Citation-tracking platforms become useful when you manage many pages or several locations, not for watching three neighbourhood queries.
Want that baseline set up properly, and interpreted? Ask for a diagnostic: I record your reference numbers before the rollout makes it impossible, test your real queries inside AI answers, and tell you honestly where you stand — including if there's nothing to change.