AI Search Summaries Reduce Click-Throughs to Health Sources by Half

Health organizations face a shifting landscape as AI-generated search results and delayed fact-checking systems reshape how people encounter and share medical information online, according to new research compiled by KFF's Health Information and Trust Initiative.

AI overviews dramatically reduce source visits

When Google displays AI-generated summaries at the top of search results, users click through to the underlying sources just 1% of the time, according to a Pew Research Center analysis that tracked the browsing behavior of 900 U.S. adults over one month in 2025. The presence of these AI Overviews also reduced clicks on traditional search results below them — users clicked links 8% of the time when AI summaries were present, compared to 15% when they were absent.

This shift matters because roughly two-thirds of adults seek physical or mental health information from search engines each year. People are now receiving information from health organizations without directly encountering those sources, potentially disrupting the relationship between providing information and building institutional recognition or trust.

Some health care and pharmaceutical companies have already begun using Generative Engine Optimization (GEO) techniques designed to increase their visibility in AI-generated answers. One study showed these methods boosted a source's appearance in AI responses by up to 40%, though results varied across different subject areas.

Fact-checks work, but timing determines spread

Separate research on social media corrections reveals that fact-checks can effectively reduce belief in false health claims, but their impact on limiting viral spread depends heavily on speed. Community Notes on X, a crowd-sourced fact-checking feature, reduced subsequent shares of misleading posts by 61.2% once added. However, these corrections appeared an average of 63 hours after the original post, while most posts had already received half of their first 36 hours of reposts within just six hours of publication.

The research, first reported by KFF, indicates that crowd-sourced fact-checks can be as effective as expert corrections in reducing confidence in misinformation. But the delay means Community Notes are better at limiting further spread than preventing the initial viral wave. This suggests corrections must appear early to meaningfully contain false claims before they reach wide audiences.

Why it matters

The convergence of these trends creates a strategic challenge for health communicators. As AI tools reduce direct engagement with source websites, organizations lose opportunities to build trust through repeated exposure. Meanwhile, the effectiveness of misinformation corrections depends on speed that current systems struggle to achieve. Health organizations may need to rethink both their technical optimization strategies and their approach to rapid response if they want to maintain influence in an AI-mediated information environment.

These findings were compiled by KFF's Health Information and Trust Initiative, which tracks health misinformation and its impact on the American public.