What Pew's 68,879 Searches Actually Show About AI Overviews and Clicks

Why this study, and not the usual chart

Claims about AI search circulate as round numbers with no method attached. Traffic is "down 40 percent," AI "ate the click," search is "dead." Most of these figures come from single vendors measuring their own denominator, and they vary by an order of magnitude depending on what exactly was counted. Against that backdrop, one source stands apart on rigor: the Pew Research Center, a nonpartisan research institution, published a study in July 2025 built on observed behavior rather than a survey of opinions or a proprietary traffic estimate.

The distinction matters. Pew did not ask people whether they thought they clicked less; it watched what they actually did, with consent, through a browsing panel. That method, tracking real navigation across a representative group, is the same approach used in the most credible clickstream research, and it is why this study is the strongest single anchor available on the question of AI Overviews and clicks.

What Pew actually measured

The design is worth stating precisely, because the precision is the point. Pew analyzed the browsing activity of 900 US adults who agreed to have their web use recorded. Across March 2025, those panelists generated 68,879 unique Google searches. Pew then compared what happened on searches that returned a Google AI summary against searches that did not.

Two properties give the study its weight. First, it is behavioral: the unit of analysis is a real search by a real person, not a recalled impression. Second, it is bounded and disclosed: one engine (Google), one month (March 2025), one country (the United States), and a stated sample. Those boundaries are not weaknesses hidden in a footnote; they are the terms that let a reader judge exactly how far the finding travels.

The headline: about 8 percent versus about 15 percent

The central result is a difference in click-through rate. On searches where an AI summary appeared, panelists clicked through to a traditional search result in about 8 percent of cases. On searches where no summary appeared, they clicked through in about 15 percent of cases.

Read carefully, this is a comparison of two conditions, not a before-and-after of the same searches over time. The accurate way to state it is associational: the presence of an AI summary is associated with roughly half the traditional click-through rate. Pew did not run a controlled experiment that randomly assigned summaries, so the study cannot by itself isolate the summary as the sole cause; queries that trigger summaries may differ systematically from those that do not. What it can establish, on primary data, is that the two populations of searches behave very differently at the moment a link would be clicked.

The 1 percent that says more than the 8 percent

The number that reframes the whole debate is the smallest one. When an AI summary appeared, panelists clicked a link cited inside that summary only about 1 percent of the time. Pew also reported that 88 percent of AI summaries cited three or more sources, and only 1 percent cited a single source.

Put those together and a structural fact emerges. Being cited inside an AI answer is common for the answer (most summaries name several sources) but rarely converts to a visit (about 1 percent). For a business, that means citation and click are now two separate outcomes. You can be named in the answer a buyer reads and still receive no visit from it. The exposure still happens, the buyer still forms an impression, but the traffic signal a business would normally use to detect that exposure barely registers. This is precisely the gap that ordinary analytics cannot see.

Behavior, not just clicks: the session-ender

Pew captured one more behavior that clicks alone would miss. Panelists were more likely to end their browsing session entirely after landing on a page with an AI summary (about 26 percent) than after a standard results page (about 16 percent). The study also found that AI summaries appeared on roughly 18 percent of all Google searches studied, and that 58 percent of panelists encountered at least one summary during the month.

The session-ending figure is the behavioral counterpart to the click figure. It suggests the summary is not merely diverting a click to a different destination; it is more often satisfying the query on the spot, after which the person stops. That is consistent with a longer-running pattern, the rise of the zero-click search, which predates AI summaries and gives the Pew result a context that a single month cannot supply on its own.

What the study does not prove

A rigorous read is defined as much by the claims it refuses as the ones it makes. Four limits should travel with every citation of this study.

• It is one engine and one month. The data is Google, March 2025. It does not describe ChatGPT, Perplexity, Gemini, or Copilot, and it does not establish a trend line from a single observation window.
• It is associational, not causal. The 8 percent versus 15 percent gap compares searches with and without summaries; it does not randomize the summary, so it cannot alone prove the summary caused the lower click-through.
• It is aggregate, not vertical. The figures describe all searches by a general US panel. They say nothing specific about high-consideration local queries (a med-spa, a home-services firm, a dental or legal practice), where the buyer, the intent, and the citation behavior may differ.
• It is a population average, not your number. An 8 percent click-through across 68,879 searches is not an estimate of how often your business is named, cited, or chosen inside an AI answer. That is an entirely separate measurement that this study does not make for anyone.

Reading Pew against the wider record

The Pew finding does not stand alone, and placing it against the rest of the record is what separates evidence from hype.

On the corroborating side, independent clickstream analysis from SparkToro, using Datos and later Similarweb panel data, found US zero-click searches rose from 58.5 percent in 2024 to about 68 percent in early 2026, and that AI Overviews, now appearing on more than 20 percent of Google searches, were associated with a drop in click-through rate of close to 60 percent when present. Two independently sourced panels pointing the same direction raise confidence that the Pew result is not an artifact of one method.

On the cautionary side, the same domain contains forecasts that did not hold. In February 2024 Gartner predicted that traditional search-engine volume would fall 25 percent by 2026 as AI chatbots absorbed queries. Two years on, that collapse has not materialized as stated: Google still holds more than 90 percent of the search market, and the actual pattern is search evolving (AI answers embedded inside Google's own results) rather than volume draining away to competitors. The click is changing; total search volume is not collapsing on the timeline that was forecast. Holding both facts at once is the discipline this field requires.

The operational conclusion: measure, do not assume

The reason a study like this belongs in a marketing conversation is not that it hands you a number to repeat. It is that it removes the option of assuming. If AI summaries appear on roughly a fifth of searches, if being cited rarely converts to a click, and if the behavior differs by engine, query, and vertical in ways no single study resolves, then the only accurate statement any business can make about its own AI-answer visibility is a measured one.

Herbert Simon named the underlying economics in 1971: a wealth of information creates a poverty of attention. When attention is scarce and fragmented across surfaces, the competitive question stops being "where do I rank" and becomes "where do I actually stand across each surface that now decides." That question has an answer, but it has to be observed, not presumed, because the traffic signals owners are used to reading are exactly the signals an AI answer suppresses.