Google’s AI Overviews cut the click-through rate of the most exposed sites by a fifth, in nine days. The lost click is only the visible part. What matters is what the machine does with your message when it reads you without citing you.
On 22 July 2026, Google rolled out AI Overviews in France.
Nine days later, the most exposed sites had lost 23.1% of their click-through rate.
The figure comes from Ahrefs, which tracked 963 domains in Search Console, twenty-eight days before launch and nine days after. The finding is stark. Pages still appear, as often as before, sometimes more. They simply get fewer clicks. One site in the panel shows a 183% jump in impressions for nearly 2% fewer clicks.
You don’t vanish from the results page. You stay there, displayed, beneath a box that answers in your place.
And the shift is only starting. Today, the median AI Overview rate across the tracked sites already sits at 5.4%, and in markets deployed before us the curve has never reversed. In France, the SE Ranking study finds an AI Overview on more than one analysed query in two within the first weeks. Almost twice as fast as in Germany or Spain.
Why isn’t your sector the right lens for AI Overviews?
What decides is the intent behind the query, not your market.
A query that a single paragraph answers is a query you are going to lose. A query that calls for a tool, a quote, an account, a purchase or a human opinion holds up far better, because the box can summarise it but can neither handle it nor close it in your place. Health, the most exposed sector, loses 20.4% of its click-through rate for an AI Overview rate of 22.2%. The same logic applies everywhere.
B2B sits on the front line. In the SE Ranking study, the Business category triggers an AI Overview on 87.1% of analysed queries, one of the highest rates measured.
If your message fits in three lines, Google will summarise it for you. And keep the reader.
So look at your keywords through that lens before you look at your vertical. In Fast Growth Advisors audits, the useful question isn’t “am I in a threatened sector,” it’s “how many of my target queries fit inside a box.”
What are you really measuring when AI reads you?
AI Overviews are only half the problem. Worse, you can’t measure what replaces them.
You might want to track this new channel the way you tracked search. That instrument is broken.
On 27 July 2026, the logs of the Fast Growth Advisors site recorded hits from the ChatGPT-User bot. On the same day, our audience dashboard showed zero visitors from ChatGPT. That zero is not a fact. It’s a blind spot. ChatGPT’s and Claude’s native apps send no referrer header, so real AI visits are structurally undercounted.
What the counter shows is worse than incomplete. It’s false.
Over thirteen days and two of our sites, 876 of 1,129 requests claiming to be AI agents fetching a page for a human were proven impostors. More than three in four were lying.
A counter that reads the user-agent doesn’t count robots. It counts lines of text anyone can write. The gap this opens is not a rounding error, it separates a channel you can actually steer from a number you can only read and hope is true.
We established it request by request, checking each claim against the address ranges publishers publish, and documented it in our analysis 876 impostors: what your AI traffic really hides.
So the consequence is clear. You can’t steer your AI visibility by counting visits.
Your tools count visits. The visit is no longer the unit. You are driving a car whose odometer ignores half the road.
This is exactly what Ahrefs confirms at market scale: impressions are not falling, the click is. Steering by impressions today means reassuring yourself with the one indicator that hasn’t moved.
What is the new unit of visibility?
Being the source the machine repeats, not the link it displays.
In a generated answer, two to five sources are selected, cited, recommended. Nobody else exists. What decides is verifiable authority and extractable structure. Language matters too: Google’s AI Overviews in France cite nearly 90% French-language sources. That proportion holds for this engine and this market, not as a universal rule, because engines do not all handle language the same way, and a market served mainly in English would not produce the same split.
All of it is decided upstream. A search engine cites well only what it grasps quickly. Any content that asserts without proving, or drowns its promise in a dense block, the machine skips for a clearer competitor.
The lever is elsewhere. It lies in the clarity of your message.
Our measurements confirm it. The Fast Growth Advisors Observatory examined 369 French post-funding startups. Their readiness to be cited by AI comes in at 0.91 out of 5. And 75.9% of them stay below the clarity threshold of their own message. Most B2B sites aren’t ready to be cited, because they aren’t ready to be understood. In the end, an engine can say what a company does, not what sets it apart, what we’ve called elsewhere unreadability for the engines.
What should you do, concretely, in B2B?
Three moves, in order.
First, map your queries by intent.
Take “what is lead scoring”: the AI Overview answers in three lines, and the reader stops clicking. Take “which lead scoring tool for a ten-person sales team”: now it takes comparing, pricing, choosing. The click survives, and that’s where your name counts.
Next, make your message extractable.
A subhead like “Our solution” tells an engine nothing. Rewrite it as a real question, “How to shorten a B2B sales cycle,” and answer in the first sentence. Mark up your FAQs, write your figures and comparisons in plain text, never inside an image an engine can’t read.
Finally, measure what you couldn’t see.
Your AI Overview exposure rate, the AI bots crossing your logs, the answers engines already give about your category: Search Console’s generative AI performance reports open part of that view.
Start with the twenty queries that bring you the most clicks. For each one, note whether an AI Overview fires, then sort them into two columns, those a single paragraph settles and those that call for a tool, a price or a decision. The second column is your base, the first measures your real exposure.
And verify the identity of the robots instead of trusting their user-agent. Otherwise you allocate a content budget for visitors who, in large part, don’t exist.
Stop steering by impressions. They have not fallen. It is the click that is going, and behind it, attention.
The French shift has barely started. The question is no longer “where do I rank.” It is “what does the machine say about me, and is it accurate.” Answering that, on your message and your proof, is precisely what a Fast Growth Advisors audit is for.
Sources
- Ahrefs, study of the click-through impact in France after the 22 July 2026 rollout, 963 domains tracked in Search Console, published 19 August 2026. ahrefs.com
- SE Ranking, “AI Overviews Already Appear on Over 52% of French Searches”, dataset collected 7 and 8 August 2026. seranking.com
Frequently asked questions
Do AI Overviews really concern B2B?
Yes, and more than average. In the SE Ranking study, Business-category queries trigger an AI Overview on 87.1% of analysed searches in France, one of the highest rates across the twenty sectors observed. The reason is intent. A large share of B2B queries exists to understand a term, compare two approaches or frame a need before any budget is committed. Those are exactly the questions a six-line box answers correctly. The exposure is therefore highest on the educational content most B2B teams built precisely to attract buyers early, which is where the loss concentrates and where the rewrite has to start.
Should you block AI bots to protect your traffic?
No. Blocking them removes you from generated answers without returning a single click. The point isn’t to be absent, it’s to be cited, and cited correctly. A block does not restore lost traffic, it only removes the one form of compensation still available, your name appearing inside the answer. The question to settle is not access, but what the engine finds once it is in.
How do you measure your AI Overview exposure?
By combining three signals: the share of your queries that trigger an AI Overview, Search Console’s generative AI performance reports, and the AI bot hits recorded in your server logs. None of these signals is enough on its own. The first gives your exposure, the second what Google returns to you, the third what other engines come looking for. At Fast Growth Advisors, we verify bot identity before counting anything.
Will search traffic disappear?
No, but it changes nature. The click thins out on short-answer queries and holds on those leading to a tool, a quote or a decision. Visibility shifts from the link to the citation. In practice, a site can watch its impressions rise, its rankings hold and its traffic fall over the same period, with no technical fault to explain the gap. What changes is not your position, it is what the reader gets before reaching it, and whether your name travels with the answer.