MSME & Global Commerce · established evidence

Ranked First, Chosen Never: What Pew's AI Overviews Study Actually Measured

Last reviewed 2026-07-20. Written by Chandranshu Kumar, Founder, Raveneye Global. · 10 min read

The Pew AI Overviews study, published by the Pew Research Center in July 2025, is the clearest behavioral evidence to date that ranking well and being chosen have come apart. Rather than asking people what they do, Pew tracked what 900 US adults actually did in their browsers during March 2025. When a Google search returned an AI summary, users clicked a traditional result in only 8 percent of those searches, against 15 percent when no summary was present. They clicked a link inside the summary itself in just 1 percent of visits, and they abandoned the session entirely more often, 26 percent versus 16 percent. The page you rank on is increasingly a page the buyer never acts on. This piece reads the study on its own terms: what it measured, what it proves, what it does not, and why the gap it quantifies matters most to the smallest businesses.

The finding that separated ranking from being chosen

For most of search's history, ranking and winning were nearly the same event. A high position on the results page produced a click, and the click produced a visit. The Pew Research Center's July 2025 study measured what happens to that chain when a generated summary sits at the top of the page, and the answer is that the chain breaks in a specific, measurable place.

Across the browsing behavior it tracked, Pew found that a Google search producing an AI summary led to a click on any traditional organic result in about 8 percent of cases, compared with about 15 percent of searches where no summary appeared. The presence of the summary is associated with roughly half the onward clicking. Ranking in the results below that summary did not disappear as an achievement; it stopped reliably converting into the visit it used to guarantee.

This is why the study matters beyond its own numbers. It does not describe a preference or a sentiment. It describes an interruption in a behavior that an entire industry priced its work against for two decades.

Behavior tracked, not opinion surveyed

The methodological point is easy to skip and central to the study's weight. Pew did not ask a panel whether they felt AI summaries changed their habits. It observed the actual browsing of a representative sample of 900 US adults over March 2025 and counted what they clicked. Stated-preference surveys are prone to the gap between what people report and what they do; a behavioral-tracking design closes that gap by watching the behavior directly.

That design is the reason this figure carries more evidential force than the many industry estimates circulating alongside it. The claim is narrow, the observation is direct, and the venue is a non-commercial research institution with no product to sell against the result.

What eight percent versus fifteen percent means for the click-through rate

The headline comparison is a click-through rate on organic links roughly halved in the presence of an AI Overview. It is worth being precise about what that halving does and does not say. It does not say that AI summaries cut every site's traffic in half; effects vary by query, by intent, and by whether the summary answers the question completely. It says that at the level of aggregate browsing behavior, the summary absorbs a large share of the intent that used to flow downward into the list.

For a business, the operational consequence is that a rank which reads as a win inside a rank-tracking tool can be a loss at the moment of decision. The tool reports position; the buyer's browser reports whether the position was ever acted on. Pew's data is a direct measurement of the second thing, and it is the thing that determines whether the ranking earns a customer.

Ranking still matters. It now competes for attention against a synthesized answer placed above it, and the answer is winning a growing share of the clicks that ranking used to collect by default.

The one percent that reframes the entire results page

The most quietly radical number in the study is the smallest one. Users clicked a link inside the AI summary itself in only about 1 percent of visits. The summary is not, in practice, functioning as a launchpad that hands traffic onward to its cited sources. It is functioning as a terminus.

This reframes what it means to be cited inside a generated answer. A citation is valuable, but not primarily as a click. The 1 percent figure suggests that the return on being named is mostly the influence of the mention itself, the fact that the engine spoke your name into the buyer's consideration set, rather than a stream of referral visits. Being the source the answer trusts shapes the decision whether or not anyone taps the link.

That distinction changes the objective. The goal is no longer only to be the result a searcher clicks; it is to be the source an answer names, because for most people the answer is where the query ends.

Abandonment and the rise of the zero click search

Pew also measured how often a browsing session simply ended after the search. When an AI summary was present, users abandoned the session entirely in about 26 percent of cases, compared with about 16 percent when no summary appeared. The summary does not only redistribute clicks within the page; it more often satisfies the query so completely that the person leaves.

This connects the study to the longer-running zero click search literature, which has documented for years that a majority of searches now end without a click to any external site. The AI Overview intensifies a pattern that predates it. The summary is a more capable version of the answer box, the knowledge panel, and the featured snippet that already trained users to get what they needed without leaving the results page.

For a business, an abandoned session is the cleanest possible picture of the risk. There is no ranking to lose to a competitor and no click to compete for. The demand was met and closed before any website entered the picture. The only way to participate in that outcome is to be inside the answer that satisfied it.

Ranked first, chosen never: why the two are now different outcomes

Put the three measurements together and a single mechanism appears. Clicks on organic results fall by roughly half when a summary is present, clicks inside the summary are negligible, and outright abandonment rises by ten points. The results page has quietly split into two competing outcomes: ranking, which is a position, and being chosen, which is being named in the answer the buyer reads first.

A business can now hold the first outcome and lose the second. It can occupy the top organic position for its main query and still be absent from the generated answer written above that position, and Pew's data says the answer is where a growing share of the attention now stops. Ranked first, chosen never is not a slogan; it is the literal description of a page where the winning rank sits below an answer that never mentions the winner.

This is also why a new optimization discipline appeared before the behavior was even fully measured. In 2023, revised in 2024, Aggarwal and colleagues introduced Generative Engine Optimization in a peer-reviewed paper, formalizing the study of which content properties change whether a source is cited inside a generated answer, and reporting that adding cited statistics, quotations, and authoritative sources measurably raised a source's visibility in the engines they tested. The object being optimized had shifted from a rank to a citation, and the research followed the shift.

The narrower funnel for small and local businesses

The Pew figures are engine-behavior averages across all searchers. The businesses that feel the mechanism most acutely are small and local ones, because the generative layer appears to name far fewer of them than classic search ever surfaced. Here the evidence is weaker and must be labeled as such.

Industry monitoring, which is not peer-reviewed and should be treated as directional rather than settled, reports that a generative engine recommends on the order of 1 percent of qualifying local businesses in category queries, against roughly a third surfaced by classic local search. If even the shape of that gap is right, the AI answer is a narrower gate than the ten blue links it sits above, and a business absent from it is absent from the only step many buyers now take.

The stakes scale with how much of the economy this describes. US small businesses number 34.8 million and account for 45.9 percent of private-sector employment and 43.5 percent of GDP, per the US Small Business Administration's Office of Advocacy. A behavioral shift that quietly reprices small-business visibility is not a niche marketing story; it touches a large share of national economic activity, and the firms with the least capacity to measure it are the ones most exposed to it.

Reading the study: what it proves and what it does not

The discipline that gives the Pew figures their authority also bounds them. The study observed one engine's behavior, Google, in one market, the United States, during one month, March 2025. It measures association between the presence of a summary and a drop in clicking; it is careful behavioral description, not a controlled experiment isolating cause. Query mix, seasonality, and the evolving design of the summary itself all sit inside the number.

None of that weakens the core claim, and it is worth saying plainly what the claim is. The claim is that in observed browsing, the presence of an AI summary is associated with markedly less onward clicking and more session abandonment. That is established. What is not established from this study alone is the magnitude of the effect for any single business, any single vertical, or any engine other than the one measured. Those are separate questions that require separate measurement.

This is the correct posture toward every figure in the AI-search conversation, including the confident ones. In 2024 a widely cited forecast projected that search-engine volume would fall 25 percent by 2026 as chatbots absorbed queries. As of this writing that specific collapse has not arrived as stated. The behavioral shift Pew documents is real; the sweeping totals and timelines around it are exactly where confident prediction keeps overreaching. The response is to measure the effect on a specific business rather than to assume the average applies to it.

What the study changes for a business being measured

The practical implication of the Pew data is not a new tactic; it is a new question. The old question was where a business ranks. The question the study forces is whether a business is named in the answer that now sits above its ranking, and how often, across the engines its buyers actually use. That is a measurement no standard rank report produces, because rank reports were built for a page whose behavior the study shows has changed.

Answering it requires observing the answer layer the way Pew observed the click layer: directly, repeatedly, and per engine, since generated answers are not deterministic and a single check proves nothing. The output of that observation is a rate, how often a business is named when a real buyer question is asked, reported with the engine, locale, and date attached. That rate is the AI-answers counterpart to the click-through Pew measured, and it is the starting point for deciding whether any further work is warranted.

The evidence does not counsel panic and it does not counsel denial. It counsels measurement. A business that knows its share of the answer can act on the gap the study describes; a business that does not is exposed to a mechanism it cannot see in any report it currently reads.

The evidence

Key findings, with their sources

When a Google search returned an AI summary, users clicked a traditional organic result in about 8% of searches, versus about 15% when no summary was present.

established Pew Research Center, "Do people click on links in Google AI summaries?", July 2025 (behavioral tracking of 900 US adults, March 2025 browsing).
Users clicked a link inside the AI summary itself in only about 1% of visits, indicating the summary functions as a terminus rather than a launchpad to its sources.

established Pew Research Center, "Do people click on links in Google AI summaries?", July 2025.
Users abandoned the browsing session entirely more often when an AI summary appeared, about 26% versus about 16% without one.

established Pew Research Center, "Do people click on links in Google AI summaries?", July 2025.
Adding cited statistics, quotations, and authoritative sources measurably raised a source's visibility inside generated answers in the engines tested, defining a new optimization object: the citation, not the rank.

established Aggarwal et al., "GEO: Generative Engine Optimization", arXiv:2311.09735, 2023 (rev. 2024), peer-reviewed.
Generative engines appear to recommend roughly 1% of qualifying local businesses in category queries, against roughly a third surfaced by classic local search, a funnel about an order of magnitude narrower.

contested Industry monitoring summarized via Entrepreneur.com / GoodfellasTech / PushLeads, 2026 (not peer-reviewed).
US small businesses number 34.8 million and account for 45.9% of private-sector employment and 43.5% of GDP.

established U.S. SBA Office of Advocacy, "2024 Small Business Profile", November 2024.

Calibration

What is proven, what is promising, what is unproven

Evidence tier	Tactics	What the evidence says
Established	Treat the click-through drop, the ~1% in-summary click rate, and the higher abandonment as measured behavior for Google AI Overviews in the US, March 2025. Measure your own AI-answer presence per engine rather than assuming the average.	Pew Research Center behavioral-tracking study, July 2025; GEO peer-reviewed paper, arXiv:2311.09735.
Emerging	Optimize content and entity signals for citation inside answers, using the GEO levers (cited statistics, quotations, authoritative corroboration) as directional, engine-tested guidance that is still generalizing.	Aggarwal et al. GEO, 2023 rev. 2024; early behavioral corroboration from Pew.
Contested	Read the "generative engines name ~1/30th of the businesses classic search does" claim as directional, not settled; verify against primary measurement before repeating a specific percentage as fact.	Marketing-industry monitoring, 2026, not peer-reviewed or standards-body audited.

Reference

Glossary

AI Overview: Google's generated summary that appears above the traditional list of results, synthesizing an answer from multiple sources and naming a few of them.
Click-through rate: The share of searches that produce a click on a result. Pew measured this for organic results with and without an AI summary present.
Zero click search: A search that ends without a click to any external website because the query is satisfied on the results page itself, a pattern the AI summary intensifies.
Share of answer: How often a business is named in the generated answers to a fixed set of real buyer questions, measured per engine. The AI-answers counterpart to a ranking position.
Generative Engine Optimization: The study, introduced in a 2023 peer-reviewed paper, of which content properties change whether a source is cited inside a generated answer.

Straight answers

Frequently asked questions

What did the Pew AI Overviews study actually measure?

It measured real browsing behavior, not stated opinions. The Pew Research Center tracked what 900 US adults did in their browsers during March 2025 and counted their clicks. It found that when a Google search returned an AI summary, users clicked a traditional result in about 8 percent of searches versus 15 percent without one, clicked links inside the summary in only about 1 percent of visits, and abandoned the session more often, 26 percent versus 16 percent.

Does ranking first still matter if AI Overviews reduce clicks?

Ranking still matters, but it no longer guarantees the click it once did. Pew's data shows the click-through rate on organic results roughly halved when a summary was present. A business can hold the top position and still be absent from the answer written above it. Ranking and being named in the answer are now two different outcomes, and a growing share of attention stops at the answer.

What is a zero click search and how does the Pew study relate to it?

A zero click search is one that ends without a click to any external site because the results page satisfies the query. Pew found that AI summaries raised outright session abandonment from about 16 to about 26 percent, intensifying a zero-click pattern that featured snippets and knowledge panels had already established. The summary is a more capable version of the on-page answer that keeps users from clicking through.

Is the Pew study proof that AI is killing search traffic?

No, and the study does not claim that. It observed one engine, Google, in one market over one month, and it measures association rather than isolating cause. What it establishes is that the presence of an AI summary is linked to markedly less onward clicking and more abandonment. The magnitude for any single business or vertical is a separate question that requires separate measurement, which is why we measure rather than assume.

How can a business tell if it is being named in AI answers?

It has to be observed directly, because no engine publishes this and normal reporting does not show it. The method mirrors how Pew measured clicking: run a fixed panel of real buyer questions repeatedly across each engine and record how often the business is named, as a rate with the engine, locale, and date attached. That share-of-answer read is the AI-answers counterpart to a ranking report and the starting point for any plan.

Provenance

Sources

Pew Research Center, "Do people click on links in Google AI summaries?", July 2025 (behavioral tracking of 900 US adults, March 2025) (established)
Aggarwal, P., Murahari, V., Rajpurohit, T., Kalyan, A., Narasimhan, K. R., Deshpande, A., "GEO: Generative Engine Optimization", arXiv:2311.09735, 2023 (rev. 2024) (peer-reviewed, established)arxiv.org
U.S. SBA Office of Advocacy, "2024 Small Business Profile", November 2024 (established)
Industry monitoring on generative-engine local recommendation rates, summarized via Entrepreneur.com / GoodfellasTech / PushLeads, 2026 (contested, needs primary data)
Gartner, press release forecasting a 25% decline in search volume by 2026, 2024 (contested, used as a forecast-accuracy check)gartner.com

Every figure above is attributed to a real, dated source and tagged with its evidence tier. Where a claim could not be verified to a primary source, it is not stated as fact.