AI Overviews in France: what the first numbers actually say

France got AI Overviews late and at scale, on 22 July 2026, so it should be the cleanest market to read their effect in. It is not yet. No public figure tells you what AI Overviews cost a French site. What circulates measures something else: a share of keywords, a group of heavily exposed domains watched for nine days, a two-year decline that started long before, and an estimate whose method no public document lets you check.

So do not carry any of these numbers into a budget as they stand. The useful decision is smaller and pays better: measure your own exposure, query by query, and read your click-through rate at constant position. A domain with 5% of its queries triggering a block and one at 30% do not have the same problem, or the same budget to defend.

Did AI Overviews cost French sites 23% of their clicks?

No. That figure is the click-through rate of one subgroup, not French traffic. Ahrefs, an SEO tool vendor, compared the 28 days before 22 July with the 9 days after, across 963 of the domains with the most French clicks (aggregated Search Console data). For domains where over 20% of queries trigger an AI Overview, median click-through rate went from 2.29% to 1.76%, or -23.1%, and 82.2% of them declined. A control group exposed below 2% gained 2.3% over the same period, which puts the adjusted effect at -24.8% according to Ahrefs.

Across the full panel the picture changes: a median decline of 8.8% in clicks and 5.7% in click-through rate, with a median domain AI Overview rate of 5.4%. If you open the source you will also find a -3.4% click-through rate for the studied domains, with no explanation of how that differs from -5.7%.

The window is nine days, on France's largest domains. Ahrefs itself warns that a slope is not a prediction and promises a rerun over a longer window, with no date. The finding most likely to last is another one: positions barely moved in late July. Pages kept showing and got clicked less.

What does each circulating number measure?

Seven numbers, seven different objects.

Details and links in the sources at the end.

None of these lines is a French site's traffic loss attributed to AI Overviews over a meaningful period.

Did the French decline start with AI Overviews?

No. Kevin Indig (Growth Memo, 26 October 2026) estimates from Similarweb data that organic visits fell 24.4% in France between August 2024 and August 2026 (7,730 sites), against 13.2% in the US (8,658 sites). Over 23 of those 24 months predate the French launch, and he writes that the comparison cannot isolate how much came from AI Overviews.

The more useful passage is narrower. Across a matched set of 1,725 sites, organic visits per day fell 5.1% in France and 3.6% in the US between June and August 2026. The gap around launch comes to 1.5 points, on modeled visits and on samples that, he says, do not represent the national web.

French news shows the same difficulty. A Media Leader analysis of ACPM audit data, reported by Siècle Digital, has the websites of 40 news brands down 22.9% from September 2025 to September 2026, with apps down less. The same report says no direct link to AI Overviews can be established, and that Google says it sees no significant decline. Next adds two factors: a Google anti-spam update rolled out at the time of the French launch, and, citing The Media Leader, a base month inflated by France's September 2025 political crisis.

If your organic traffic is down 20% in a year, the July launch cannot explain that alone. Plot the earlier trend first.

Where does the 33 to 38% figure come from?

From an estimate attributed to Arcom, the French media regulator, with no public document found as of 4 November 2026. APIG, the French news publishers' alliance, quotes it in its 11 August 2026 release announcing a referral to the competition authority, without linking a source. Clubic and Génération-NT describe an Arcom study of European countries where the feature was already live, and do not supply it.

The only Arcom passage on the subject that could be found sits in its June 2026 audience barometer. It says that in Italy, Germany or Switzerland the feature "aurait conduit à une diminution sensible des audiences des éditeurs" (reportedly led to a noticeable decline in publisher audiences). In the conditional, with no number.

33% and 38% are also the global and US declines in Google organic traffic that Chartbeat measured across 2,576 sites from November 2024 to November 2025, published by the Reuters Institute, which adds that it is unclear how much is down to AI Overviews. Nothing shows whether this is a coincidence.

Until the document is public, do not present this figure as an Arcom measurement.

What does the randomized experiment prove?

That the effect is real where the block shows, mostly on informational queries. Agarwal (Indian School of Business) and Sen (Carnegie Mellon) randomly hid AI Overviews for 1,065 US desktop Chrome users, across 68,089 searches, from 7 January to 10 February 2026. The working paper could not be opened on SSRN, so these figures come from PPC Land's report.

With the block, 0.37 organic clicks per search; without it, 0.62. That is -39.8%. Zero-click searches rose 34.5%. Informational queries (71% of the sample) carry the effect. On navigational and transactional queries it cannot be told apart from zero, with noisier estimates. Sponsored clicks stayed flat.

Two results get quoted less. User satisfaction did not change. Neither did click quality: back-button use, bounces and time on page showed no clear difference. In August 2025 Google described click volume as "relatively stable" with "slightly more quality clicks", with no figures. In this experiment, the lost clicks are not made up for by better ones.

Scale depends on what you look at. -58% on position 1 of an AI Overview keyword (Ahrefs, desktop, correlation), -39.8% per affected search, about -5% on English Wikipedia's search traffic (Khosravi and Yoganarasimhan, difference-in-differences, October 2026 version). They do not contradict each other. They are not about the same thing.

Why do presence estimates run from 5% to 97%?

Because they do not count the same thing.

The two French figures are compatible. SE Ranking gives 5,000 keywords to each of 20 niches, so business at 87.1% weighs as much as food at 20%. Its title says searches, its measure is keywords, and the page states neither the selection method nor the device. Ahrefs looks at what actually makes up the query portfolio of very large domains.

SE Ranking also makes a telling comparison: 52.63% within two weeks, against first-month rates in 2025 of 27.86% in Germany, 33.23% in Spain and 26.56% in the Netherlands. By the same vendor's count, France landed straight at the level Germany reached later, about 53%. The rule to hold on to: no percentage without unit, country, device and month.

What is AI Overview exposure?

AI Overview exposure is the share of your tracked queries on which Google shows a block, whether or not your brand appears in it.

A query with no block is not missing data. It is a result. Take those queries out and exposure reads 100% everywhere. Read alone, visibility inside blocks can then send the wrong signal.

Arithmetic illustration, fictional brand and figures.

The visibility score falls 20 points while real presence grows 80%. Meanwhile, three times as many queries are exposed to a click decline. Both readings are true, and neither is enough alone.

Which queries should you look at first?

The ones where the stakes differ, in this order.

• Informational, top of funnel. This is where the experiment finds the click loss. Semrush nonetheless reports the informational share of AI Overview triggers falling from 91.3% to 57.1% between January and October 2025 (over 10 million keywords).
• Commercial and comparison queries. On 400 US keywords, Semrush goes from 45.75% in September 2025 to 88% in August 2026, on a small sample with no click data (reported by PPC Land). In France, SE Ranking finds 33.58% in e-commerce and 20.28% in fashion and beauty. The experiment's null on transactional queries dates from early 2026 on a limited sample: do not count on it.
• The most covered sectors. Business (87.1%), relationships (86.04%), insurance (73.62%) and healthcare (71.02%) at SE Ranking. In health, Ahrefs reports a 22.2% AI Overview rate and click-through rate down 20.4%.

SE Ranking also notes that 32.93% of French blocks sit lower on the page. In the US experiment, only blocks at the top of the page cut clicks measurably, according to PPC Land. No public measure of French clicks by placement could be found as of 4 November 2026.

Should a budget move on these numbers?

Not on these numbers. On your exposure. Here is a scenario calculation, not a forecast. Apply the US experiment's effect as it stands: a domain 5.4% exposed would lose about 2% of its clicks, one at 25% about 10%. Use Ahrefs' slope instead (each exposure point costs about 0.9 points of click-through rate change): about -5% and -22%.

The French slope is more than twice what the experiment suggests. Several explanations are possible: exposure counts queries rather than clicks, heavily exposed domains are probably more informational, the window is nine days, the country and devices differ. Nobody has settled it. For a leadership team, that gap is the information: it gives a range, not a number. The method for choosing between channels is in the GEO, SEO and paid budget article.

What will need rechecking?

Almost everything above has a short shelf life.

• The Ahrefs rerun. Promised with no date. The English version of 16 September 2026 rests on the same nine-day window.
• The Arcom document. Publication would show the scope and method behind 33 to 38%.
• A French measure on mobile. The known causal design covers US desktop users.
• Commercial queries. The coverage increase seen in the US, and its effect on clicks, remain to be measured in France.
• The publisher control. Google now lets site owners decide in Search Console whether they appear in these features, and says it is not a ranking signal outside them. The trade-offs are in the AI bots guide.

What should you measure now?

Five readings, on your own data.

1. Freeze a baseline. Export Search Console by query and page. Compare the same weeks year on year and plot the trend before July 2026, so you do not blame the launch for what came earlier.
2. Isolate click-through rate at constant position. Stable position, stable or rising impressions, falling clicks: that is the profile Ahrefs describes. Google counts clicks from these features in the Web search type. Search Console's generative AI report, open to all sites as of 31 August 2026, shows impressions only: how often your URLs appear in these features, not the clicks, and nothing about blocks your site is absent from. So the effect cannot be read directly.
3. Record exposure on a frozen list. Your business queries, with or without a block, noting the absences. For list size, see how many prompts to keep.
4. Build a control. Compare exposed queries with comparable ones that are barely exposed, as Ahrefs does at domain level. The logic is in the article on GEO incrementality.
5. Read visibility next to exposure. Then tie both to conversions, with AI traffic tracking in GA4.

The third reading fits in a spreadsheet for a few dozen queries. It gets heavy across a whole portfolio, week after week. For projects tracking Google AI Overview, Nanga's GEO module gives it a dedicated card, "AI Overview exposure", on the Performances page: the share of your tracked questions on which Google shows a block, with the count with and without one. A missing block never counts as a missing brand, since those questions are excluded from AI Overview visibility metrics. The Models filter then shows who appears inside the blocks: mentions, sources, competitors.

In September 2026, over 14 days and on tracked questions that do not represent the market, the Nanga team found that about 12% of AI Overview readings had no block at all.

To see which questions get a block from Google, and who appears in it: request a demo.

Frequently asked questions

Sources:

• Ahrefs, 1 mois d'AI Overviews en France et déjà -23,1% de taux de clics : qui est concerné ?, August 2026.
• Ahrefs, AI Overviews Cut CTR by 23.1% in France, September 2026.
• Ahrefs, Update: AI Overviews Reduce Clicks by 58%, February 2026.
• Kevin Indig, Growth Memo, Organic traffic is down, but it's not all AI, October 2026.
• Agarwal and Sen, SSRN working paper, reported by PPC Land, Researchers find Google AI Overviews cut publisher clicks 39.8%, June 2026 revision.
• Khosravi and Yoganarasimhan, Impact of AI Search Summaries on Website Traffic, arXiv, October 2026 version.
• SE Ranking, AI Overviews Already Appear on Over 52% of French Searches, August 2026.
• Xu, Iqbal and Montgomery, Measuring Google AI Overviews, arXiv, October 2026 revision.
• Pew Research Center, Google users are less likely to click on links when an AI summary appears in the results, July 2025.
• Semrush, Semrush Report: AI Overviews' Impact on Search in 2025, December 2025.
• Semrush, reported by PPC Land, Google AI Overviews now hit 88% of tracked commercial keywords, October 2026.
• Alliance de la presse d'information générale, Services d'IA de Google : l'Alliance saisit l'Autorité de la concurrence, August 2026.
• Clubic, Résumés IA de Google : craignant de perdre « 33 à 38% » de leur trafic, August 2026.
• Génération-NT, Résumés IA de Google : l'Autorité de la concurrence saisie, August 2026.
• Arcom, Baromètre d'audience des services d'intelligence artificielle, 1re édition, June 2026.
• Reuters Institute, Journalism, Media, and Technology Trends and Predictions 2026, January 2026.
• Siècle Digital, Les sites d'information français perdent près d'un quart de leur trafic, Google dans le viseur, October 2026.
• Next, La consultation des principaux sites d'information a chuté de près de 25 % en un an, October 2026.
• Google, Lancement des Aperçus IA et du Mode IA dans la Recherche Google en France, July 2026.
• Google, AI in Search: Driving more queries and higher quality clicks, August 2025.
• Google Search Central, AI features and your website, read 4 November 2026.
• Google Search Central Blog, generative AI performance reports in Search Console, June 2026, updated 31 August 2026.