What AI Overviews did to traffic in France, in nine days

When Google switched on AI Overviews in France on July 22, 2026, the median French site lost 8.8% of its daily clicks within nine days, and sites where more than a fifth of queries trigger an AI summary lost 23.1% of their click-through rate, according to an Ahrefs study of 963 French domains published on August 17, 2026.

France is worth your attention even if you never sell there. Every other major market got AI Overviews gradually, mixed in with algorithm updates and seasonal noise. France got them in one day, two years late, which makes this the cleanest before-and-after measurement available on what AI summaries do to a website's traffic.

Why is the French rollout a better experiment than the others?

The French launch produced a control group, which is what previous studies lacked. Ahrefs tracked the 963 highest-click French domains using aggregated Google Search Console data, comparing the 28 days before July 22, 2026 with the 9 days after. It then split them: heavily exposed domains, where more than 20% of queries trigger an AI Overview, against a control group where fewer than 2% do.

The control group gained 2.3% in click-through rate over the same period. Heavily exposed domains lost 23.1%. Because the control group tells you what every site would have experienced anyway, seasonality included, Ahrefs puts the true effect of AI Overviews at 24.8%. That gap between two groups on the same days, in the same market, is the strongest evidence yet that the losses come from the AI summary and not from an algorithm update.

How much did an average site actually lose?

Across the full Ahrefs panel, the median was a 5.7% drop in click-through rate and an 8.8% drop in clicks. That number is far less alarming than the headline, and it is also far less useful, because the damage concentrates rather than spreads. Here is how the 963 domains split by daily clicks lost.

• 48% of the panel lost more than 10% of their daily clicks.
• 19% lost more than 30%.
• 5% lost more than half.
• 82.2% of heavily exposed domains saw their click-through rate fall, so at that level of exposure, surviving is the exception.

Whether you sit in the calm half or the damaged half is not luck, and it is not your industry. It is the share of your queries that triggers an AI summary.

Why did rankings hold while clicks collapsed?

AI Overviews do not push pages down the results, they make clicking on them unnecessary. Ahrefs found few ranking losses in France in late July 2026: pages still appeared as often as before and sometimes more often, they were simply clicked less. The most extreme case in the panel recorded 183% more impressions alongside 1.9% fewer clicks, a click-through rate cut by more than three.

This is why the problem goes unnoticed for months. A dashboard tracking average position shows nothing wrong. A dashboard tracking impressions shows growth. Only click-through rate reveals the loss, and it is the metric most companies check last. The underlying behavior is documented: the Pew Research Center, analyzing real browsing data from 900 US adults across 68,879 Google searches in March 2025, found users clicked a result on 8% of visits with an AI summary against 15% without, and clicked a source cited inside the summary on just 1% of visits.

How exposed is your own site?

The number that matters is your exposure rate: the share of your queries that triggers an AI Overview. In the Ahrefs panel the median was 5.4%, and 29% of domains already exceed 10%. Ahrefs found a consistent relationship between exposure and loss, with a slope of 0.9, meaning each additional point of AI Overview rate costs roughly one point of click-through rate. If 15% of your queries trigger a summary, expect to lose somewhere around 15% of your click-through rate.

Treat that slope as an order of magnitude, not a forecast. Ahrefs is explicit that some heavily exposed sites hold up while some barely exposed ones fall for reasons unrelated to AI. Exposure is a risk factor to cross-check against your own Search Console numbers. Any visibility audit worth paying for measures eligibility and risk, never a guaranteed outcome, which is the distinction drawn in What an AI visibility scan actually measures, and what it cannot.

Which queries are gone, and which survive?

Intent decides your exposure, not your sector. Health is the most affected vertical in the Ahrefs study, with a 22.2% AI Overview rate and a 20.4% fall in click-through rate, because health queries are informational, short-answer and carry no transaction at the end. The answer fits inside the box, and the reader leaves with what they came for.

The working rule is one sentence long: a query a single paragraph can answer is a query you are going to lose. Audit your keywords through that lens before you look at your industry.

• Absorbed by the summary: definitions, opening hours, unit conversions, symptoms, general comparisons, anything with one factual answer.
• Holds up well: queries asking for a price, a quote, a tool, an account, availability, a human opinion or a purchase decision.
• Grey zone: brand comparison queries, where the AI Overview cites sources and being one of those sources becomes the whole game.

Does this mean search traffic is finished?

No, and treating it as a collapse leads to the wrong spending. Impressions did not fall in the French panel, they often rose, which means people are still searching just as much. What changed is the value of ranking on a question the summary can answer by itself. Your classic search work keeps its value on transactional queries, the ones that hold up, and the informational traffic drying up rarely carried immediate commercial value in the first place.

The practical shift is that answer engines become a separate surface to monitor, with its own signals and its own failure modes, alongside the search rankings you already track.

What should you do, and what does it cost?

The first useful step costs nothing. Open Google Search Console and compare your clicks and click-through rate query by query across a period before and after AI Overviews reached your market. If your click-through rate is steady, you are barely exposed and the rest of this section can wait. If impressions climb while clicks fall, you have found the problem and measured it for free.

The second step is making your site readable and citable by answer engines, since the queries that survive are also the ones where being cited pays. At Biencité the scan is free, monitoring starts at €24 per month on Starter and €39 per month on Pro, and a managed installation is a one-off €290 or €490 depending on the platform. The three axes and nineteen signals are listed in full on the method page. For how those figures compare with agency pricing, see How much does an AI visibility audit cost in 2026.

When is doing nothing the right call?

Three common situations make this spending premature. If your exposure rate is near zero because your queries are local, transactional or highly specialised, your budget belongs elsewhere. If your site has no factual content worth citing, no technical signal will compensate, because an answer engine cannot cite what does not exist. If your company has no reputation, no reviews and no outside mentions, the priority is commercial rather than technical.

There is one genuinely urgent case: a site where a meaningful share of revenue depends on informational queries, and where the split between rising impressions and falling clicks is already visible. Everyone else should measure now, keep a record of the before, and reassess in three months as exposure rates climb.