Eight percent. That's the share of visits in which someone clicked a regular result after seeing an AI summary at the top of the page, according to Pew Research Center data published in July 2025. On visits without a summary, the figure rises to 15%. And clicks on a link inside the summary itself, the one that's supposed to give credit to the source? Around 1%. The sample was 900 American adults tracked in March 2025, and 58% ran at least one search that produced a summary. It isn't a huge pool, and it's a single month. Still, it's one of the few measurements that comes neither from the people selling search nor from the people living off it. ### <br>Google says clicks are stable, publishers say otherwise In August 2025, Liz Reid, who runs Search at Google, wrote in an official post that total organic click volume was relatively stable year over year, and that the clicks that remain are higher quality, meaning people spend longer on the pages they land on. On the other side is Similarweb, which measured zero-click news searches rising from 56% to 69% between May 2024 and May 2025, and organic visits to news sites falling from over 2.3 billion to under 1.7 billion a month. Both can be right at the same time, which is the kind of answer that annoys people. "Total clicks" includes shops, maps, videos, everything. The pie can stay the same size while the slice going to people who publish text shrinks. There's also a detail that gets lost: Google itself said queries in AI Mode are two to three times longer than usual searches, which suggests people are asking more and asking differently, not necessarily clicking more. Google doesn't publish the denominator, so outsiders can only look at samples. Who's right? I don't know, and I distrust anyone who says they do. ### <br>The best defence of the direct answer Who hasn't opened a recipe and waded through a grandmother's life story in Setúbal, two pop-ups and fourteen ads before reaching the amount of flour? Search by links became unpleasant, largely because of the people writing for it. Pages bloated to please the algorithm, "15 best" lists copied from one another, texts that take four paragraphs to admit they have no answer. If a summary saves two minutes and answers the question, the publisher complaining sounds a bit like a coachman protesting against the automobile. It's a strong argument and I find it hard to dismantle entirely. It comes apart in part. It works for questions with a stable answer: how long to boil an egg, what time a shop closes, what an acronym means. It fails when the answer is a judgement made by a person. Why the router drops the connection at 11 pm, whether this laptop will last three years, what a brand's customer support is really like. There a summary can be fluent and vague at once, and you only notice when you've already bought. ### <br>The bill The summary is built from pages that exist because, for over twenty years, someone was paid (badly, often) with the clicks those pages brought. If the click disappears, the machine still needs new text and there's no one left to pay for it. The big players have noticed and are negotiating. Reddit signed a 2024 deal with Google that Reuters estimated at around $60 million a year. In July 2025, Cloudflare began blocking AI crawlers by default on new domains and launched, in private beta, a pay-per-crawl model. In September that year, Penske Media, owner of Rolling Stone and Variety, sued Google, alleging that the summaries cut its traffic and affiliate revenue. A small site has no lawyers and no user data to sell. It has text, and text is precisely the thing that has already been consumed. ### <br>An odd loophole There's one channel that doesn't depend on anyone asking anything: Discover, the feed Google pushes to your phone. Many publishers describe the same pattern in Search Console, impressions rising and clicks lagging behind, and some are betting more on what reaches people without a search. My reading is that this is a nice paradox. Google can answer a question without sending anyone anywhere, but it still needs to show people things they didn't know they wanted to know. Answers serve people who already know what they're looking for. For those who don't yet, someone has to write. I keep getting stuck on the 1%. A link inside an answer, with a tiny icon beside it, is proof that attribution exists and that almost nobody uses it. In March 2027 it would be fair for someone to repeat the Pew study with the same methodology. I bet the 8% won't be 8% anymore. I just don't know which way it will go.