Google AI Overviews Cut Publisher Clicks 40%, Study Finds

The first causal evidence

For two years, publishers have argued that Google's AI Overviews are draining their search traffic. Google has generally pushed back, and the debate has mostly relied on before-and-after traffic charts that can't separate AI summaries from everything else changing in search. A new field experiment adds harder evidence to that argument.

Saharsh Agarwal of the Indian School of Business and Ananya Sen of Carnegie Mellon University's Heinz College built a custom browser extension and installed it for 1,065 desktop Chrome users. The extension randomly decided whether each participant saw Google's AI-generated summaries at the top of results pages. 1 Because assignment was random, the authors say they can attribute differences in behavior to the AI Overviews themselves rather than to seasonality, algorithm updates, or shifting user habits. 1

The paper was first posted to the Social Science Research Network on April 3, 2026, and revised June 17. It found that when an AI Overview appeared, outbound organic clicks fell 39.8% and zero-click searches rose 34.5%. 1 Clicks on sponsored results did not change. 1 The authors also report that the summaries did not improve users' search experience. 1 On that reading, users clicked less, but they were not more satisfied.

Citations don't close the gap

Google's standard answer to traffic concerns is that AI Overviews link to their sources and send visitors onward. Digital Content Next, the trade group for premium publishers, says the experiment undercuts that defense. Even when Overviews cite sources, those links recover only a small share of the lost traffic. 3 DCN frames the result as a structural problem. When the answer sits on the results page, most users have little reason to go further, and source links work more as attribution than as a way to send readers onward. 3

The split between organic and paid clicks matters here. If AI Overviews reduce clicks to publishers but leave ad clicks untouched, the cost lands on the open web while Google's core revenue stays intact. That is an inference from the reported numbers, not a finding the authors state about motive. Still, the pattern is hard to ignore.

The aggregate data points the same way

The experiment measures individual behavior under controlled conditions. Industry analytics show what that looks like across the news business. Chartbeat's 2026 Publisher Playbook reports that Google Search referrals to news and media sites fell 40.2% between July 2025 and July 2026. 2 Across Chartbeat's network, Google Search's share of all pageviews dropped from 9% in July 2024 to 5% in July 2026, a 46% relative decline over two years. 2 Chartbeat points to AI-generated search summaries as the main cause. 2

The two figures are not directly comparable. One is a per-search effect among desktop Chrome users. The other is a year-over-year aggregate across a news-heavy network that includes mobile traffic and other factors. Even so, both land at roughly 40%. A controlled experiment and real-world traffic data arrive at similar numbers, which is about as strong a case as this kind of research can make.

Where publishers and marketers are turning

The responses fall into two groups. News organizations are leaning harder on audiences they don't have to rent from Google. Chartbeat's data shows publishers relying more on direct visits, loyal readers, and private "dark social" sharing through messaging apps and other untracked channels. 2 DCN argues that publishers need new routes to audience growth because search referrals are no longer dependable. 3

The other group is a new optimization industry. Marketers are already selling "Answer Engine Optimisation" and "Generative Engine Optimisation," which promise visibility inside AI answers from Google, ChatGPT, and Perplexity rather than in traditional result rankings. 4 One New Delhi-based consultant, for example, pitches these services on the premise that customers now ask chatbots to recommend accountants or compare software instead of browsing multiple websites. 4

Read against the experiment, that pitch has a limit. Earning a citation in an AI answer may help a brand get noticed. The research suggests it is a weak source of traffic. For a local business that wants a name mentioned, that may be acceptable. For publishers whose revenue depends on pageviews, appearing in the summary does not replace a click.

The takeaway

The Agarwal-Sen study turns a long-running complaint into a measured effect. AI Overviews cut organic clicks by about 40%, leave ad clicks alone, and do not clearly improve search for users. Cited links recover only a small part of the loss. Publishers waiting for AI referrals to make up the difference are unlikely to see it. The more realistic options are building direct reader relationships, which they control, and pressing for policy or commercial terms on how their content is used, which they cannot win through optimization alone.