Google's AI answers and the clicks that stop arriving: what changes for a business

Anyone running a company website usually notices it in the numbers before they notice it in the news: organic search traffic drops, but the keywords the site ranks for stay the same. The suspicion is almost always the same one: the AI-generated answers Google shows above the list of traditional links. The useful question isn't whether the phenomenon is real, but how well it's actually measured, and what changes in practice for a business that uses content to be found rather than sells it.

What the data say, and where they stop

The most cited starting point is a Pew Research Center study, published on July 22, 2025 and checked for this article on October 10, 2026. Researchers tracked the browsing of 900 US adults in March 2025, across 68,879 Google searches, of which 12,593 (about 18%) produced an AI-generated summary. The result: a traditional result link gets clicked in 8% of visits when a summary appears, versus 15% when it doesn't. Only 1% of visits with a summary produce a click on a link inside the summary itself, and 26% of pages with a summary end the browsing session with no further step, versus 16% of pages without one.

These are solid numbers, but with a limit the authors state themselves: the sample is a panel of users who agreed to have their browsing tracked, not a random sample of the whole population, and the study covers only Google, not other search engines. It's observational data: it shows the two things move together, not on its own that one causes the other. Google has publicly disputed the study, citing a flawed methodology and an unrepresentative set of queries, but without publishing its own data to support that objection.

The proof an observational study can't give

Isolating the cause needs an experiment, not just an observation. Two researchers, from the Indian School of Business and Carnegie Mellon University, ran one using a browser extension that randomly shows or hides the AI summary for a group of over a thousand users recruited in early 2026, observing nearly 70,000 searches over two weeks. It's still unpublished in peer-reviewed form, so it should be read as a strong signal, not a final word.

The central finding concerns informational searches, the ones where someone wants an explanation rather than a product or a service: removing the AI summary from these searches increases clicks by a statistically significant amount. Across all searches combined, the study measures a 39.8% drop in organic clicks and a 34.5% increase in searches that end with no click at all. For navigational and transactional searches, the ones where people search to buy or book, the collected sample is smaller and the result less certain: the authors themselves flag this as a limitation, not as proof those searches come through unscathed.

It isn't the same for every search

The reason the strongest data concern informational searches is simple: that's where the AI summary appears almost every time. In the same experiment, the summary is shown or triggered on 53% of informational searches, versus 15% of navigational ones and just 6% of transactional ones. An independent Ahrefs analysis of 300,000 keywords, based on real Search Console data, confirms the direction for informational searches: the click on the top organic result for these searches fell from 7.3% to 2.6% between March 2024 and March 2025.

What the data don't allow, at the current state of the research, is calling commercial searches safe. They're simply less measured, because the AI summary shows up there less often. Assuming your own site is protected just because it sells something, rather than explains something, is a conclusion the available data don't yet support.

What actually changes for a business running a website

The practical point isn't to stop writing content, but to look at which content is actually still carrying the traffic. An article that answers a generic question, the kind an AI summary can reconstruct on its own by combining public sources, is the most exposed type of content: that's what the data describe, not an impression. Content that instead requires direct experience, a concrete case, a piece of data the business owns that nobody else can summarize, stays harder to replace with a generic summary.

The second thing that changes is where you look to judge whether the site is working. Search traffic on its own can drop even when the site is doing fine, simply because whoever already had the answer before clicking no longer reaches the page: it's the same reason why nobody can guarantee a fixed position on Google, and now not even a fixed click rate at the same position. What matters for a business is how many contact requests, bookings or sales come from the site, not how many people pass through the search results page. It's also one reason, not the only one, why keeping a website under observation over time, instead of leaving it alone after launch, pays off: the way people find a business online changes more often than the site itself does.

For Italy, the measurement is missing, not the phenomenon

For the Italian market, there is currently no public, independent measurement able to isolate how much of a site's traffic drop depends on AI summaries versus other factors, such as seasonality, a search algorithm update, or waning interest in the topic. The traffic-decline figures that circulate for Italian sites, where they exist, lump several causes together without separating them. It should be said with the same honesty as the rest: that's a limit of the available measurement, not proof the phenomenon doesn't exist in Italy.