What changed
On August 18, researchers released a preregistered randomized field experiment involving 1,100 participants using Google Search in ordinary browsing. Participants were assigned to one of three conditions: current Search, Search with AI Overviews and AI Mode removed, or an AI Mode-only experience. The researchers report that removing Google’s AI search features increased click-through to third-party publishers, while the AI Mode-only condition reduced publisher referrals and also lowered reported trust and user experience relative to current Search.
Why it matters
Publishers and search marketers have had abundant observational evidence that AI answers coincide with fewer outbound clicks, while Google has argued that AI Search can generate more queries and higher-quality referrals. This experiment is materially stronger evidence because the search experience was randomly manipulated in the field rather than inferred from traffic trends. It does not tell an individual publisher how much traffic will change, but it strengthens the case for treating AI-search exposure and referral traffic as separate distribution metrics and for reducing dependence on traditional Google click-through as AI Mode expands.
The experiment changed the search experience, not the publisher
The researchers used a browser extension to randomly assign participants to three Google Search conditions during normal browsing: current Search, Search with AI Overviews and AI Mode hidden, or a version redirecting searches to AI Mode. That design helps isolate the causal effect of the search interface from seasonality, publisher changes and many of the confounders that affect observational traffic studies.
Removing AI features increased outbound publisher clicking
The study reports that participants in the no-AI condition clicked third-party publishers more often than users seeing current Google Search. Conversely, the AI Mode-only condition reduced external click-through. The direction is consistent with the substitution concern: more information is consumed inside Google rather than through visits to the sites supplying source material.
AI Mode did not offset the traffic loss with better reported experience
The researchers also report that the AI Mode-only experience reduced trust in information found on Google and worsened user-experience measures relative to current Search. That result challenges a simple trade-off in which lower publisher traffic is assumed to buy a clearly superior search experience, although the effects may vary by task and population.
This is stronger evidence, but not a universal traffic forecast
The paper is a newly released preprint rather than a peer-reviewed final publication. Its 1,100 participants and field design improve causal inference, but the results should not be translated into a fixed percentage loss for every site, query class or country. Google’s products, ranking systems and AI interfaces also continue to change, so the experiment is best treated as evidence about the direction and mechanism of attention shift rather than a permanent coefficient.