AI search overviews reduce external clicks by 38% in Google test

Google’s AI Overviews sent fewer people to outside websites in a randomized field test, cutting outbound organic clicks by 38% on queries where they appeared. (searchenginejournal.com) The working paper came from Saharsh Agarwal and Ananya Sen of the Indian School of Business and Carnegie Mellon University, and Search Engine Journal reported it on April 28, 2026. (searchenginejournal.com) The researchers built a Chrome extension and randomly assigned 1,065 U.S. participants to normal Google Search, a version with AI Overviews hidden, or Google’s AI Mode. The study ran for two weeks per participant between January and February 2026. (searchenginejournal.com) AI Overviews appeared on 42% of queries in the experiment. When they showed up, zero-click searches rose from 54% to 72%, meaning many users got an answer and stopped without visiting another site. (searchenginejournal.com) Position mattered. The click drop was strongest when the AI Overview sat at the top of the page, which happened 85% of the time in the test, and removing those top-position boxes nearly doubled outbound clicks. (searchenginejournal.com) The user-experience result was different. Survey responses from the normal-search group and the group with AI Overviews removed were nearly identical on satisfaction, perceived information quality, and ease of finding information. (searchenginejournal.com) That result lands in a search market already shifting toward answer boxes instead of blue links. Pew Research Center found that 58% of 900 U.S. adults in its March 2025 browsing panel saw at least one Google search with an AI summary, and users were less likely to click links when one appeared. (pewresearch.org) Other studies have pointed in the same direction, though with different methods. Ahrefs said in April 2025 that position-one pages on queries with AI Overviews saw a roughly 34.5% lower click-through rate than comparable informational queries without them. (ahrefs.com) One academic paper on Wikipedia traffic found a smaller effect — about a 15% daily traffic decline for exposed English articles — but that paper was later withdrawn on March 26, 2026, pending a broader sample. (arxiv.org) The newer randomized test is narrower than all of Google Search because it covers active U.S. desktop Chrome users, not mobile users or the full web. But it adds one concrete number to a fight publishers, marketers, and Google have been having for months: answer-first search appears to keep more searches inside Google. (searchenginejournal.com)