What happened
Researchers from the University of Pennsylvania and Northeastern University ran a randomized experiment splitting 1,100 tracked participants into groups that searched with Google's AI Mode on, with AI Overviews stripped out, or with AI features left untouched, then surveyed 956 of them afterward. The paper, "AI in Search Reduces Publisher Referrals Without Improving User Experience: Experimental Evidence", was posted to arXiv in August 2026 and got wider trade coverage on September 21 via Search Engine Journal. The routing worked: 94.7% of the AI Mode group's searches actually landed in AI Mode, and 51.1% of AI Overviews were successfully hidden in the stripped-down condition. With that controlled, the researchers could isolate the feature's actual effect instead of guessing from aggregate traffic logs, which is what most AI-search debates have relied on until now. That matters because Google has publicly maintained, through executives including search head Liz Reid and CEO Sundar Pichai, that AI features expand what people search for without hurting the broader web, a claim this experiment was specifically designed to test rather than take on faith.
Why AI Mode Reduces Clicks matters now
MagnetizeX builds founder visibility systems for B2B firms.
Founders who built an inbound engine on organic search have been told the click drop is a wash, that AI summaries just move the same traffic somewhere else on the page. This study says otherwise: AI Mode cut both search sessions and clicks to outside sites, and removing AI features increased clicks without hurting how satisfied people felt. Heavy search users, the closest behavior pattern to a B2B buyer doing multi-session vendor research, showed the strongest drop in trust and relevance ratings. Users in the AI Mode group were also significantly more likely to bail to a competitor engine like Bing, DuckDuckGo, or Yahoo entirely. If your pipeline still assumes a blue link gets clicked the way it did two years ago, this is the data that says budget that channel lower and start treating an AI citation as its own distinct goal.
- Audit your click-through data by query typePull your Search Console data for the last 90 days and segment by query intent. If informational queries are losing clicks faster than transactional ones, that's the AI Mode effect showing up in your own funnel, not a ranking problem.
- Stop optimizing purely for position oneA top ranking that gets summarized and never clicked is worth less than it used to be. Shift some effort toward getting cited by name inside the AI answer itself, which still drives brand recall even without a click.
- Diversify away from a single discovery channelWith AI Mode pushing some searchers to competitor engines entirely, make sure your content and citations show up on Bing and in AI answer engines too, not just in classic Google rankings.
By the numbers: 94.7% of AI Mode group searches were successfully routed into AI Mode; 51.1% of AI Overviews were hidden in the no-AI condition; AI Mode's negative effects on trust and satisfaction were most pronounced among heavy search users, the closest proxy this study has for a researching B2B buyer.
What to do this week
Open Google Search Console this week and compare click-through rate on your top 20 informational-intent pages against the same period last quarter. If CTR is down while impressions and position hold steady, that's AI Mode eating your clicks, not a content problem, and it's a signal to invest that content budget into answer-engine citations instead of another round of on-page SEO tweaks.