France's First Month With AI Overviews: A Verdict That Doesn't Hit Everyone Equally

TL;DR

• One month after the July 22, 2026 French launch of AI Overviews, the ten leading news sites all lost organic referrals, between 11.3% and 29.6% (Le Monde around -29.6%, Le Parisien -26.6%, Le Figaro -23.4%). E-commerce and real estate mostly grew: Fnac +4.2%, Temu +21.9% (Similarweb estimates analyzed by Aleyda Solís, Aug. 2026).
• The split has a mechanism: 99.2% of keywords that trigger an AI Overview are informational (Ahrefs). Information gets absorbed into the answer. The transaction still needs the click.
• Belgium is the preview, sixteen months ahead: on 554 HubSpot keywords with strictly stable rankings across Belgium, Switzerland and Canada, clicks fell 44% in a year (Ahrefs, 2026). The ranking didn't move. The click did.
• And since August 24, ChatGPT shows ads in 31 European markets. The moment where opinions form is moving into generated answers; the moment of purchase stays at the click, and that click is now the ad inventory of both Google and OpenAI.
  

What did the first French month actually show?

A split, not a wave. Google switched on AI Overviews and AI Mode in France on July 22, 2026. One month later, Aleyda Solís compared organic traffic to major French sites before and after, using Similarweb estimates. The pattern is unusually clean:

Source: Similarweb estimates analyzed by Aleyda Solís, August 2026.

Exposure was immediate and massive: SE Ranking measured AI Overviews on 52% of French searches within weeks of launch. And one month of data does not prove causality; Aleyda has the rigor to say so herself, and an algorithm shift can never be excluded. But when all ten sites in one category move in the same direction, at that scale, in the exact month a new answer layer covers half the country's searches, and the pattern matches what every previously exposed market has shown, the burden of proof starts to shift sides.

The counterpoint worth taking seriously

Honest analysis means airing the strongest objection, and it comes from Le Monde itself. Seven weeks after launch, Le Monde Group CEO Louis Dreyfus said the paper's total audience had not dropped abruptly, seasonal effects aside. His argument: the figures circulating describe search traffic, not total audience. Apps delivered 36% of the group's traffic in August, and search drives only part of the rest, so a sharp fall in Google referrals dilutes into a much smaller total decline. Meanwhile, the French press alliance APIG has asked the competition authority to enforce Google's commitments, citing the regulator Arcom's estimate of a 33% to 38% search-traffic loss in European markets where AI Overviews already run.

Both things can be true, and probably are: search referrals to French news are falling hard, and diversified publishers with apps, subscriptions and direct audiences absorb the shock far better than search-dependent ones. Dreyfus added one number that should reframe the whole debate for every publisher: visitors arriving from ChatGPT convert into paying subscribers at 26 times the rate of visitors from Facebook. Fewer clicks, radically better clicks: it's the same trade we documented in Authority Beats Awareness, now confirmed from inside a newsroom.

Why is information absorbed while the transaction is spared?

Because AI Overviews are built for questions, not carts. Ahrefs' analysis of the keywords that trigger an AI Overview found that 99.2% are informational in intent. Purchase-intent queries rarely trigger one. The asymmetry follows mechanically: when the answer to "what is", "how to" or "which is better" appears above the links, the click becomes optional. Pew Research measured it: users click a result in only 8% of visits when an AI Overview appears, versus 15% without one. When the intent is to buy, the answer can't complete the job. The click survives because the transaction requires it.

That's the first mechanism behind the French split. News sites sell exactly what the answer layer absorbs: understanding. Retailers sell what it can't: the product.

The demand for information is also leaving Google

The second mechanism operates outside Google entirely. As we detailed in our analysis of the Belgian OpenAI Signals data, the number one use of ChatGPT in Belgium is practical guidance, at 36.3% of messages: how to do something, what to choose, how to decide. Add information seeking at 20.4%, and the majority of what 3.5 million Belgians ask an AI has nothing to do with a search box. Understanding, comparing, forming an opinion: that increasingly happens inside a conversation, before Google is ever opened, if it's opened at all.

So publishers and brands face a pincer. Inside Google, the answer layer absorbs informational clicks. Outside Google, the informational questions are migrating to assistants. The common consequence: the moment where an opinion forms is detaching from the visit.

The third fact: the ads have arrived

Since August 24, 2026, ChatGPT shows advertising in 31 European markets, the 27 EU member states plus Norway, Iceland, Liechtenstein and Switzerland; Belgium and France included. The format: sponsored links beneath AI-generated answers, shown only to Free and Go tier users, non-personalized at launch for GDPR compliance, with self-serve access through Ads Manager since August 31 (OpenAI, 2026).

Read in isolation, it's a news item for advertisers. Read next to the French traffic data, it's something else: a map of where the value now sits. The moment where people form opinions is moving into generated answers, on Google and in assistants. The moment where they buy stays at the click. And that click, the one commercial moment the answer layer cannot absorb, is precisely what both Google and OpenAI are now monetizing. The organic informational click is evaporating; the transactional click is becoming the industry's most contested paid inventory.

For brands, the strategic reading is uncomfortable but clarifying: influence is earned upstream, in the answers, where ads don't decide who gets cited. What remains buyable is the last step, and everyone will be bidding on it.

Belgium is the preview: sixteen months ahead, and one number to remember

Belgium has lived with AI Overviews since March 2025, sixteen months before France. Which makes francophone data from exposed markets the closest thing France has to a forecast. Ahrefs built exactly that: it tracked HubSpot's keywords across Belgium, Switzerland and Canada, isolating 554 keywords whose Google position stayed strictly stable over a full year.

The result: clicks on those keywords fell from 3,481 to 1,932. A 44% decline, at rankings that did not move at all (Ahrefs, 2026). And this is HubSpot, a company with a dedicated team working on SEO and AI visibility. The broader data points the same way: Ahrefs' 300,000-keyword study now measures a 58% lower click-through rate for the top-ranking page when an AI Overview is present, up from 34.5% just eight months earlier (Feb. 2026).

The ranking didn't move. The click did.

That sentence should end every SEO reporting meeting in France this quarter. Position tracking, the metric two decades of digital marketing were built on, no longer measures what reaches you. A pattern we see confirmed weekly in the GEO monitoring we run across francophone markets: visibility inside the answers and traffic to the site have become two separate variables, moving on separate curves, and only one of them appears in a rank tracker.

The bilingual wrinkle: two languages, two SERPs, two clocks

Belgium adds a layer France doesn't have, and it matters for anyone selling across the language border. The Belgian rollout itself was staggered by language. At the March 26, 2025 launch, AI Overviews on Belgian SERPs triggered only on English-language queries; Dutch-language queries began triggering them around May 10, 2025, with French confirmed in the same period (observations documented by Belgian agencies Onlinemarketing Monkey and Universem, 2025). Two languages, one country, different clocks from day one.

Coverage then grew explosively. Across an analysis of more than 8 million keywords for 70 Belgian websites in 7 sectors, the share of queries triggering an AI Overview went from 3.37% in May 2025 to 14.53% in August 2025, nearly fivefold in three months, while the share of queries where those sites appeared as a cited source fell from 2.43% to 1.48%. And 86.8% of the AI Overviews were triggered by informational queries (Onlinemarketing Monkey, Semrush data, Sept. 2025), the same intent asymmetry driving the French split.

Full transparency on what the data does not yet say: to our knowledge, nobody has published a current, quantified comparison of French versus Dutch trigger rates within Belgium. Practitioners observe that the two SERPs don't move in lockstep, and the staggered rollout proves the two languages have never been on the same clock. But until someone measures it at scale, treat the FR/NL coverage gap as a hypothesis to test on your own keyword set, per language, not as an established figure.

Two facts that are measured, and that already reshape the budget arbitrage for a bilingual e-commerce player:

• Not translated means invisible. A study of AI Overview citations found that translated websites gain up to 327% more AI Overview visibility for queries in languages the site wasn't originally available in; AI systems overwhelmingly prefer sources available in the searcher's language (Clonable study, Jan. 2026). A bilingual site with a thin NL (or FR) side is forfeiting the answer layer on that side, whatever its rankings.
• Belgian answers cite Belgian sources. Across 44 Belgian queries in French and Dutch (440 sources analyzed), 70% of the sources AI engines retrieved were Belgian (.be) and only 2% came from France (.fr) (Nexuro Digital, July 2026). French-language authority built for the French market does not automatically carry into Belgian French answers. Each language-market needs its own local corroboration.

The operational conclusion: measure your AI Overview trigger rates per language on your own keywords, budget FR and NL as two distinct answer markets with their own local sources, and never assume that data from France describes francophone Belgium.

What should you do with a month one verdict?

Treat it as a map, not a panic. Four moves follow directly from the data:

1. Split your measurement. Track impressions versus clicks in Search Console and watch the gap widen on keywords where an AI Overview appears. Track citations inside the answers separately. Two curves, two dashboards.
2. Reassign your informational content. Its job is no longer to harvest clicks; that model is what just broke for French news. Its job is to earn citations: to be the source the answer is built from, which is what feeds high-intent visits like Le Monde's 26x-converting ChatGPT subscribers.
3. Fortify the transactional layer. The purchase click survives, and it's about to get more expensive. Use-case pages, documented outcomes, reviews that hold up under verification: the assets that convert the visit the answer triggers.
4. Build the two sides of the signal now, while the francophone answer space is thinly contested. On-site structure the engines can extract; off-site corroboration they can trust. The brands cited when the answer layer matures will be the ones who built for it during month one.

France just received its verdict. Belgium received it sixteen months ago. The question is no longer whether the click economy is being redrawn.

It's whether you're building for the answers, or still reporting on the rankings.

FAQ

What happened to French websites after AI Overviews launched?

In the first month after the July 22, 2026 launch, all ten leading French news sites lost organic referrals, between 11.3% and 29.6% (Le Monde ~-29.6%, Le Parisien -26.6%, Le Figaro -23.4%), while e-commerce and real estate mostly grew (Fnac +4.2%, Temu +21.9%), per Similarweb estimates analyzed by Aleyda Solís. One month doesn't prove causality; SE Ranking measured AI Overviews on 52% of French searches within weeks.

Why do AI Overviews hurt news sites more than e-commerce?

99.2% of keywords triggering an AI Overview are informational (Ahrefs); purchase queries rarely trigger one. Information is absorbed into the answer, the transaction still needs a click. Pew measured clicks in 8% of visits with an AI Overview versus 15% without.

What does the Belgian market show about long-term impact?

Belgium has had AI Overviews since March 2025. On 554 HubSpot keywords with strictly stable rankings across Belgium, Switzerland and Canada, clicks fell 44% in one year (Ahrefs, 2026). Ahrefs' 300,000-keyword study now measures -58% CTR for the top result under an AI Overview.

What changed with ChatGPT ads in Europe?

Since August 24, 2026, ChatGPT shows sponsored links beneath answers in 31 European markets (EU27 plus Norway, Iceland, Liechtenstein, Switzerland), for Free and Go users only, non-personalized at launch, with self-serve Ads Manager since August 31 (OpenAI). Opinion-forming moves into answers; the purchase click becomes paid inventory for Google and OpenAI alike.

Do AI Overviews behave the same in French and Dutch in Belgium?

The rollout was staggered by language: English-only at the March 2025 launch, Dutch queries triggering from around May 10, 2025, French confirmed in the same period. Belgium-wide coverage then grew from 3.37% to 14.53% of queries in three months (Onlinemarketing Monkey, 8M+ keywords, 2025). No published study yet quantifies a current FR versus NL trigger-rate gap within Belgium, so test it on your own keywords per language. Two measured facts frame the arbitrage: translated sites gain up to 327% more AI Overview visibility in the searcher's language (Clonable, 2026), and 70% of sources cited in Belgian AI answers are Belgian (.be) versus 2% French (.fr) (Nexuro Digital, 2026).

-----

HOW THIS WAS WRITTEN: Subject, angle, convictions: mine. AI assists with the writing. I remain in control.

Sabrina Bulteau is the founder of PingPrime.ai, GEO Expert and specialist in Narrative Authority in AI Search. She helps brands, institutions and media become the reference AI engines trust, cite, and repeat, not just one option among many, by working both sides of the signal: on-site (structure, narratives, architecture) and off-site (earned media, platforms, co-citations, sector press). Co-founder of Be Connect (acquired by iO Group) and Sench, active within CEC Belgium, she brings 25 years of experience in digital growth and strategic positioning.