AI Overviews France: measuring CTR loss with GSC 28d/9d
Summarize this article with AI
July 22, 2026: France flips, and the cause is not an algorithm
July 22, 2026. Hold on to that date. The Ahrefs study, published September 16, 2026, tracked 963 French domains in Google Search Console. 28 days before the AI Overviews launch. 9 days after. The numbers speak for themselves. The domains most exposed to AI Overviews lost 23.1% of their click-through rate in nine days. Nine days, not three months.
I track my French clients’ curves since that date. The pattern is the same everywhere. The cause is a new surface that captures the click before you do, not an algorithm penalty. The study confirms it: 82.2% of heavily exposed domains lost CTR. At that exposure level, ending up untouched is the exception.
« France had a launch date, and we were already tracking: 963 domains in Google Search Console, over the 28 days before launch and the 9 days after. » — Ahrefs study, Juliette Begue, September 16, 2026
France provided an ideal test bed. A single date. No major core update at the same time. No strong end-of-July seasonality. I could isolate the AI Overviews effect. The overall median is less dramatic: −3.4% CTR in nine days. But that average mixes heavily exposed sites with those that aren’t. The gap is sixfold between the most exposed and the least exposed. The least exposed barely moved. The most exposed fell off.
You can measure your own exposure right now. Google Search Console is already in your hands. The 28d/9d protocol is exactly that measurement. Let’s dive in.
Les impressions montent, les clics baissent. Voici le découplage mesuré par l’étude Ahrefs sur 963 domaines français, ramené en base 100 pour comparer d’un coup d’œil.
Avant / après le 22 juillet : le découplage
Impressions, clics et CTR médians 28 jours avant et 9 jours après (base 100).
Impressions rise, clicks fall: the decoupling is here
Impressions rise. Clicks fall. That’s the great decoupling. The Ahrefs study on France shows that rankings held for almost everyone at the end of July. Pages keep showing up. Sometimes more often. They just get clicked less.
The most extreme case in the French panel shows +183% impressions for −1.9% clicks. Its CTR is divided by more than three. An AI Overview does not remove you from the results page. It makes you redundant inside it.
The overall median is reassuring: −5.7% CTR and −8.8% clicks in nine days. Except that median hides a brutal distribution.
- 48% of sites lost more than 10% of their daily clicks.
- 19% lost more than 30%.
- 5% lost half.
In nine days. And exposure will keep climbing. Today, the median AI Overviews exposure rate among the studied domains is 5.4%. 29% of them are already above 10%. These numbers have only moved one way since 2024.
What slips is the click, not the traffic. So measure your CTR before and after. Not your impressions. Not your positions. Your CTR. On an e-commerce site, a product page that answers an informational query loses qualified visits. On a health blog, an article that explains a symptom gets its click sucked away by the AI box. Pages stay indexed. Clicks evaporate. This decoupling comes from the SERP, not from a bug. Its architecture changed. The click that used to go to your content stops before it reaches you.
Rien de sorcier : cinq étapes dans Google Search Console et vous tenez votre propre mesure. Voici le pipeline exact, à appliquer dès maintenant.
Le protocole GSC 28d/9d en 5 étapes
La marche à suivre pour isoler votre perte de CTR avant / après le 22 juillet.
The GSC 28d/9d protocol: your own measurement, not someone else’s
The 28d/9d protocol does one thing: isolate your CTR loss before and after July 22, 2026. Your data already sits in Google Search Console, ready for you now.
Why these durations? 28 days before gives enough volume for a stable baseline. 9 days after gives enough hindsight to see the immediate impact with minimal noise from other variables. You compare the same set of queries or pages across both periods. Why 9 days and not 30? The effect is immediate, and ranking algorithms also shift. By isolating on 9 days, you limit the noise.
The GSC 28d/9d protocol in 5 steps
- Open Google Search Console, Performance report.
- Filter by country: France (and optionally by device type).
- Before period: June 24 to July 21, 2026. Export: queries, clicks, impressions, CTR, average position.
- After period: July 22 to July 30, 2026. Export the same columns.
- Join the two files by query or URL. Calculate the variation in CTR, clicks, and impressions.
What you’re looking for: queries with stable or rising impressions, but a CTR that drops by more than 10 points. Those are your queries exposed to AI Overviews. In GSC, you don’t see the presence of an AI Overview directly. But the impressions/clicks decoupling remains a reliable signal.
To get your precise exposure rate, use a SERP tracking tool. In Ahrefs Site Explorer, the Organic keywords report lets you filter by SERP feature: AI Overview. You get the exact list of queries where the AI Overview appears. Cross-reference with the GSC protocol. You know which pages to rework.
Run this protocol every 28 days. Exposure keeps increasing. Your measurement should keep up. One file per month. One CTR variation column. You hold a system, not a one-night reaction.
La relation est presque linéaire sur les 963 domaines. Plus votre exposition aux AI Overviews grimpe, plus votre CTR baisse — voici la pente à retenir.
Exposition IA × variation de CTR
Chaque point d’exposition coûte environ un point de CTR (pente −0,9).
One point of exposure costs about one point of CTR: the slope to remember
The relationship is almost linear. If you compare the CTR variation to the AI Overview exposure rate of each of the 963 domains, you get a slope of −0.9. In plain terms: each additional point of exposure costs about one point of click-through rate.
15% of queries that trigger an AI Overview? Expect to lose about 15% of CTR. It’s rough. But it’s actionable. You pull your exposure rate from Site Explorer, and you hold an order of magnitude.
In my view, this slope is not a condemnation. Some heavily exposed sites do well. Others, barely exposed, collapse for reasons that have nothing to do with it. Your exposure rate gives you a risk factor, nothing more.
The Ahrefs study reminds us: a slope is not a prediction. In other words: 15% of your exposed queries does not mean 15% less CTR. But you know where to look. You know which pages to monitor first.
And if you want a dashboard: pull your exposure rate by site section once a month. Cross-reference with the CTR variation from the 28d/9d protocol. Rank pages into three groups: CTR loss < 5%, between 5 and 15%, > 15%. Your efforts go to the last group. Not to the noise. The −0.9 slope guides your analysis.
Health, e-commerce: who takes the hardest hit?
The hardest-hit sector, according to the Ahrefs study, is health. Logical. Simple informational queries attract AI Overviews like a magnet. « Flu symptoms, » « paracetamol dosage, » « normal blood pressure at 60 »: AI answers directly. No click needed.
E-commerce takes the same hit. I observe that product pages ranking on comparative or technical queries suffer the same mechanism. A page that explains « which juicer to choose » gets duplicated by an AI summary. No click. No cart.
For health, the impact is direct: fewer visits to editorial content from labs and medical sites. For e-commerce, it’s more insidious: informational traffic erodes, and with it, assisted add-to-carts.
Measure by sector. Look at page types, not just the domain as a whole. A site can lose 5% overall and 30% on its health pages. The 28d/9d protocol applied by page type shows you this. Filter in GSC by pages containing « /blog/ » or by product directory. You’ll see massive gaps.
The study doesn’t detail other sectors, but the exposure logic applies everywhere. The more your pages answer simple questions, the more exposed you are. The more your pages require a click to compare prices or read a detailed review, the less an AI Overview is enough. But stay alert: AI learns fast. And exposure has been climbing steadily since 2024.
What I’ve been deploying with my clients since July 22
Since July 22, the rule is simple: measure first, then change only what the data says. The DOSE framework I use with Guillaume Attias (BMO Academy) starts with observation. We decide after the data.
With my French clients, I do four things.
- Pull the AI Overviews exposure rate via Ahrefs or an equivalent tool. I want the exact queries where the AI box appears.
- Apply the 28d/9d protocol to GSC data, page by page, query by query. Isolate CTR variations.
- Prioritize pages whose CTR drops more than 10% with exposure above 15%. The others wait.
- Rework content to answer the question before the AI does. The goal: earn the click when the user wants to go deeper.
With some clients, we’ve seen pages lose more than 30% CTR in a week. Traffic didn’t move. The click did. Order of magnitude. Here, structure beats volume. A direct answer box. A comparison table. An expert opinion. The person who read the AI Overview and clicks wants detail. Give it to them.
On health pages, we add medically reviewed opinions. Product pages gain a price comparison at the top. And on a blog post, we start with the answer in 50 words. Blocking the AI Overview? Impossible. What we want is the click from the person who asks for more. The 28d/9d protocol becomes a ritual. Every 28 days, we pull the measurement. We adjust. We measure again. It runs. With full calm.
Measure before you panic. Or change.
963 domains. 28 days before. 9 days after. −23.1% for the most exposed. A slope of −0.9. The numbers are there. The protocol too. It’s up to you to apply it.
You measure erosion, rather than suffer it. Target pages where CTR drops more than 10%. Note what you modify. Adjust. Measure again.
July 22, 2026 moved the line, and everything remains open. It’s one more data point to integrate into your SEO steering. The sites that will thrive are those that built a measurement system before reacting, not those that denied the evidence.
Measurement is within reach. Google Search Console is free. The protocol takes an hour. A €200/month tool is optional: one export, one variation formula, one sort. How many clicks are you losing in silence?
One question: on your high-exposure queries, where is your CTR today? If you lack the number, you know exactly what to do tomorrow morning.
SEO audit: measure your AI Overviews exposure
During the audit, I calculate your exposure rate, apply the 28d/9d protocol to your GSC data, and identify the pages to rework first. You leave with a quantified action plan.
Book a strategic call — 45 minFrequently Asked Questions
How do I know if my site is exposed to AI Overviews?
I open Ahrefs, or any SERP tracking tool, and filter on « AI Overview. » On the GSC side, I monitor queries where impressions hold but CTR drops after July 22, 2026.
Does the 28d/9d protocol work if my site has low traffic?
Yes, but on aggregates. Group your queries by page type or directory. The −0.9 slope gives you an order of magnitude. For an exact prediction, look elsewhere.
What should I do if my CTR drops more than 20%?
I start with pages that answer simple questions. A direct answer box at the top of the page, or a comparison table. And I measure again after 4 weeks.
Do AI Overviews affect mobile as much as desktop?
The Ahrefs study doesn’t break down the numbers by device. The impressions/clicks decoupling, I see it everywhere. Filter by device in GSC to see if mobile suffers more than desktop on your domain.
How long does it take to see a real impact?
Ahrefs numbers are 9 days after launch. Exposure keeps climbing. Run the measurement every 28 days: you see where your CTR goes.
