Does Google AI Mode Reduce Clicks to Websites?

Yes, by a lot, at least in the first controlled test. Researchers moved a group of people into Google AI Mode for a week. Compared with people using normal Google search, their click-through rate to outside websites fell by 18.8 percentage points. If fewer people click, clicks alone can't tell you how visible your brand is anymore.

That's the gap topgeo.ai is built to close. topgeo.ai tracks how often, and how favorably, AI engines name your brand compared with your competitors, so you can see your visibility even when nobody clicks. Then we publish AI-optimized pages on your own domain to improve it, and tag each enquiry they bring in with the AI engine that likely sent it.

What did the study do?

It's a preregistered field experiment by researchers at the University of Pennsylvania and Northeastern University, posted on arXiv on August 18, 2026. About 1,100 US adults who use Chrome and Google took part, recruited in March 2026.

After three normal days, each person was randomly put into one of three groups for seven days:

• Current Search: Google as usual.
• No AI Search: AI Overviews hidden and AI Mode blocked.
• AI Mode Search: every search sent to Google AI Mode.

A browser extension handled the switch, and recorded what people searched and clicked.

What did it find?

• Fewer clicks: the AI Mode group's click-through rate to outside websites fell by 18.8 percentage points.
• Hardest hit: clicks to Reddit fell 21.2 points, news sites 12.5 points and Wikipedia 9.9 points.
• Less trust and satisfaction: people in AI Mode rated their searches as less useful, less satisfying and less trustworthy.
• More switching: the share of people who also searched on Bing, DuckDuckGo or Yahoo rose by 11.2 points.

What doesn't it prove?

The authors are clear about the limits. It's a preprint that hasn't been peer reviewed yet. People were forced into AI Mode, while real users choose it, and only for seven days. The sample was US-only and skewed younger and more educated than the population. And it's not set up to say much about high-stakes topics like health or finance.

So treat 18.8 points as strong early evidence, not a forecast for your own traffic.

If clicks shrink, what should you measure?

When an AI answer satisfies the question, the buyer may never visit your site. But they still read which brands the answer named. So alongside traffic, track:

1. Whether you're named: ask AI the questions your buyers ask and note if your brand appears.
2. How you're described: is the mention accurate, and is it positive?
3. Who's named instead: which competitors keep showing up where you don't.
4. The visits you do get: check your analytics for AI referrals, and treat them as a floor, not the total.

Repeat the same questions every few weeks. The answers change, and the trend tells you more than any single check.

---

Source: Stephanie T. Wang, Jeffrey Gleason, Yakov Bart, Christo Wilson and Danaé Metaxa, "AI in Search Reduces Publisher Referrals Without Improving User Experience: Experimental Evidence", arXiv preprint, August 18, 2026.