For a long time Google worked as a kind of sorting system. You typed a question, received a list of results, and from there the journey began: a newspaper, Wikipedia, a forum, a university website, someone's blog where ten years earlier they'd had the same problem with their washing machine. The search engine didn't really contain the answer, but it knew how to point you toward where you could find it. Looking something up on the Internet almost always meant entering someone else's site.
With AI Overviews, that distinction is blurring. Google pulls information from multiple pages, synthesises it, and returns an answer above the traditional results. Users can ask follow-up questions without leaving the search, and in the more conversational modes, continue almost as they would inside ChatGPT.
The scale is already enormous. In June, Google reported that AI Overviews had surpassed 2.5 billion monthly users, while AI Mode had passed one billion. From a user's point of view, the advantage is obvious: instead of opening five pages to figure out when to prune a monstera or how a contractual clause works, you can get a summary right away.
The problem is that when the answer arrives before the link, the link loses part of its function. The Pew Research Center analysed the browsing behaviour of nine hundred American adults in March 2025 and found that when an AI Overview appears, users click on one of the regular results in 8 percent of visits. When the summary doesn't appear, that figure rises to 15 percent. The links embedded directly in the AI text are clicked even less: in roughly 1 percent of visits.
These are limited figures: they concern the United States, a specific period, and they don't on their own demonstrate what happens to every type of site. But they point to a concrete change in the way people move through the web. It's not just a question of how many searches Google continues to generate; it's a question of how many of those searches still turn into a visit to an external site.
In France the conflict has just entered a more concrete phase. AI Overviews arrived on 22 July 2026 and the Alliance de la presse d'information générale, which brings together around three hundred publications, is considering a complaint to the antitrust authority. No formal complaint appears to have been filed yet.
The traffic figures also call for caution. The Reuters Institute, using aggregated Chartbeat data from more than 2,500 news sites, found that organic search traffic from Google fell 33 percent globally and 38 percent in the United States between November 2024 and November 2025. The report explicitly warns that it is not possible to attribute the entire decline to AI Overviews: Discover, shifts in social media use, and differences between content types all play a role. Service pages — weather, guides, horoscopes, practical advice — appear particularly exposed to a search experience that answers before redirecting elsewhere.
Google disputes the idea of a generalised traffic collapse. It argues that AI-powered features increase the overall number of queries and that they continue to send billions of clicks to websites every week. The two things are not necessarily mutually exclusive. Google can keep delivering an enormous volume of traffic and, at the same time, make it less necessary to open a page to get what you were looking for.
The question becomes more interesting when you stop looking only at newspapers. A vast share of the Internet's usefulness was built by people who had no economic model to defend: those who answered a question on a forum, wrote a guide to fixing an amplifier, reviewed a campsite, photographed a hiking trail, updated a Wikipedia page, or explained on a blog how to care for a plant.
For decades the web accumulated information because millions of people had a reason — economic or otherwise — to publish it. Search engines organised that material and brought readers to whoever had produced it. Generative systems can use the same material as invisible infrastructure, leaving the user with mostly just the final output.
Loading a page means encountering, at least partly, the context in which a piece of information was produced: who signed it, what kind of site publishes it, what else it contains, how reliable it seems, what interests it might serve. It also means exposing yourself to unplanned detours. You look for one thing, find another, follow a link off to the side, and drift away from the original question.
The generative answer tends to compress that journey. It assembles different materials and returns them in a more efficient form, but makes the sources less visible and reduces that random dimension of browsing that for many years was a defining feature of the Internet.
The conflict between Google and publishers is certainly about copyright, advertising, and traffic distribution. But it anticipates a broader question: if the web becomes primarily the reservoir from which artificial intelligences extract their answers, what incentives will remain to keep feeding it?
News organisations can negotiate commercial deals, demand compensation, or appeal to regulators. What is less clear is what happens to all that informally produced knowledge that made the Internet useful precisely because no one had needed to commission it. An Internet you no longer need to visit can look like simply a better Internet. The paradox is that it will keep working only as long as someone still has a reason to put things online that we will no longer need to go and read.
