On 22 July 2026, Google activated AI Overviews and AI Mode in France, two years after their US launch. France was the last major market without these features, due to the dispute over neighbouring rights for the press. A button in Search Console lets sites opt out of appearing in them, but it is not the comfort setting it appears to be.
Some changes are immediately noticeable, while others slowly transform an entire economy. The arrival of AI Overviews in France belongs to the second category. We asked in a recent article whether AI would kill the web. The French answer to that question begins now.
What actually changes for you
Two distinct features have been activated, and they should not be confused.
AI Overviews display a summary written by Gemini above the classic results, accompanied by links to the sources used. The traditional blue links remain present, but are no longer necessarily the first entry point on the page. Google says this box only appears when its systems judge that an AI-generated answer is useful for the query.
AI Mode is something else: a conversational search experience powered by Gemini 3.5, accessible by text, voice or image. It is designed to handle complex questions that previously required several successive searches. A revealing detail about the shift in usage: Google says an AI Mode search lasts on average three times longer than a traditional search, with users formulating longer, more contextual queries.
The scale is considerable. Google claimed in early June to have 2.5 billion monthly active users for AI Overviews and more than a billion for AI Mode. France is therefore joining a system that has been widely tested elsewhere.
Why France waited two years
The delay has nothing to do with technology. It stems from the dispute over neighbouring rights, the right that allows press publishers to be paid when their content is reused by platforms.
Google has been sanctioned twice by the French Competition Authority on this matter: a €500 million fine in 2021, then €250 million in March 2024, notably because its AI service used press content without informing publishers. This regulatory climate froze the French rollout while the feature expanded to more than 200 countries. The timeline was finally formalised in late June, before an activation faster than most observers expected, who had pencilled it in for September.
The button that looks like a trap
Here is the point that press releases cover in three lines but which deserves three meetings in any newsroom. Google has opened a control in Search Console in France allowing a site owner to decide whether they want to appear in the search's generative AI features and contribute to their answers.
On paper, this is progress: the choice is handed to publishers. In practice, the decisive detail appears in the English version of the June announcement. Google writes there that sites that opt out will receive neither traffic nor impressions from these generative features. So this is not a citation setting, it is a visibility switch.
A publisher has a choice between two uncomfortable options. Accept feeding AI answers, hoping the links displayed compensate for lost visits, with no guarantee of payment. Or refuse, and disappear entirely from surfaces used by billions of people. In an ecosystem where Google concentrates most search traffic, the second choice is less about protection than voluntary withdrawal. That is why many newsrooms talk of a choice that is not really one.
The real impact, measured elsewhere
What do we know about the effects, two years after the rollout in other countries? The data is real but mixed, and one should be wary of any single figure.
In markets where the tool has been running for two years, traffic declines observed at publishers are said to range between 20 and 40%. But the effect varies strongly by content type: according to analyses reported in the trade press, general news content appears less affected than thematic content. Some publishers see a clear decline, others mere stagnation.
This variation has a logic. Content that answers a simple factual question is easily summarised, and therefore easily replaced by a box. Content that brings an investigation, an analysis, a point of view or verifiable expertise holds up better, because the summary cannot stand in for it.
What this changes for us, and we say it plainly
Let us be transparent: Alyriaz is a French online media outlet, and this rollout concerns us just as much as any other site. We are exactly in the position we describe.
Our reading, and it is a deliberate choice, is that the answer does not lie in the Search Console button. Opting out to protect a model that relies on traffic amounts to depriving yourself of visibility without gaining anything in return. The answer lies in the nature of what you publish. An article that exists only to capture a query is doomed, whatever setting you choose. An article that brings an angle, a verification, a perspective that the summary cannot carry keeps a reason to be visited.
It is also for this reason that we take the time, in our articles, to flag what is not verified, to distinguish an announcement from a fact, and to point to our sources. These are precisely the elements a generated box does not reproduce.
The final word
The arrival of AI Overviews in France is neither a catastrophe nor a non-event. It is the acceleration of a transformation already under way elsewhere, with a two-year lag that has at least given us time to observe its effects. The question is no longer whether the web will change, but which content will survive that change.
For the reader, there is a very concrete and almost intimate consequence. Every time you click to read the full article rather than settling for the box, you bring to life the economy that allows someone to write it. This is not a moralising appeal, it is a mechanism. In the web taking shape, attention becomes a vote.