Study Finds Google AI Mode Reduces Website Clicks and Worsens User Experience
A randomized field experiment by researchers from the University of Pennsylvania and Northeastern University found that Google's AI Mode significantly reduces click-through rates to external websites and lowers user satisfaction, directly contradicting public statements made by Google executives.
A draft research paper published in August 2026 concludes that Google's AI search features reduce traffic to publisher websites and deliver a user experience poor enough to drive some users toward rival search engines, findings that directly contradict claims made by Google's leadership.
The paper, titled AI in Search Reduces Publisher Referrals Without Improving User Experience: Experimental Evidence, was produced by researchers from the University of Pennsylvania and Northeastern University, Massachusetts. It is based on a randomized field experiment that tracked real Google searches conducted by 1,444 enrolled participants, of whom 1,100 had their searches monitored over a seven-day period. Of those, 956 completed a post-experiment survey.
Methodology
Participants were randomly assigned to one of three conditions: Forced Regular Google Search, in which AI Overviews were hidden and AI Mode searches were redirected to standard results; Natural Google Behavior, in which Google operated normally; and Forced AI Mode Search, in which all searches were routed into AI Mode. Researchers tracked behavior through a browser extension that recorded search activity, browsing history, and snapshots of search results pages.
The researchers noted difficulties maintaining the experimental conditions. According to the paper, "During the experiment period, 94.7% of searches in the AI Mode group were successfully routed to AI Mode. In the No AI condition, 90% of AI Overviews were successfully hidden on the first intervention day, but the success rate declined to 0% as the experiment progressed. The cause of this breakdown was a change to AI Overview HTML rolled out by Google during the study period that broke our extension's logic for identifying and hiding AI Overviews. Overall, 51.1% of AI Overviews were successfully hidden and the median participant in this condition had 50% of AI Overviews hidden."
Key Findings: Clicks and Competitor Usage
The study found that users assigned to AI Mode were substantially less likely to click through to external websites. Conversely, when AI Overviews were removed, clicks to external websites increased, with no detectable decline in users' search experience. The researchers also found that AI Mode caused some users to turn to rival platforms, observing that "Assignment to AI Mode Search significantly increased the fraction of users searching on a competitor engine (Bing, DuckDuckGo, or Yahoo) and intention to switch to Bing."
Search Session Frequency
Contrary to the researchers' own predictions, AI Mode did not increase users' search activity. Participants randomly assigned to AI Mode conducted fewer search sessions. The researchers also noted: "We do not find evidence that exposure to No AI Search impacted search sessions per day." This indicates that while AI Mode reduced search activity, the absence of AI features had no measurable effect in either direction.
User Satisfaction and Trust
AI Mode broadly worsened how users perceived Google Search, reducing trust, perceived usefulness, satisfaction, and personal relevance. By contrast, removing AI features produced no detectable change in user perceptions. The study found no clear difference in user perceptions between AI Mode and regular Google Search specifically for news-related searches. The negative effects of AI Mode on search activity were more pronounced among heavy users of Google Search than among lighter users.
Contradiction With Google's Public Statements
Google's CEO Sundar Pichai and other company representatives, including Liz Reid, have publicly stated that users are enthusiastic about AI Mode and that it is sending more users to the open web than ever before. The findings of this study directly contradict both claims. Google had not publicly responded to the paper's findings at the time of publication of this article.
Prepared with AI assistance and reviewed by the editorial team.