Google's own data shows AI Mode cuts clicks to your website

A randomised field experiment with 1,100 Google users found that forcing people into Google's AI Mode cut clicks to outside websites by 18.8 percentage points, while also lowering trust and satisfaction with search. For founders and marketers, this is the clearest causal proof yet that AI-first search is a structural threat to referral traffic, not a temporary side effect.

What changed?

Researchers ran a preregistered field experiment that randomly assigned real Google users to either normal search or forced AI Mode, using a custom browser extension to track actual clicks rather than relying on surveys. The study found fewer publisher visits, lower trust and satisfaction, and greater use of rival search engines when people were pushed into AI Mode, according to Search Engine Land.

The numbers are specific. When participants' searches were forced into AI Mode, the share of searches that led to an external website fell by 18.8 percentage points compared with normal Google use. People in AI Mode also searched less overall, running 0.92 fewer Google search sessions per day.

The damage wasn't spread evenly. The share of users who clicked news sites fell by 12.5 percentage points, clicks to Reddit fell by 21.2 points, and clicks to Wikipedia fell by 9.9 points.

Crucially, the researchers also tested what happens when AI features are removed rather than forced. Blocking Google's AI Overviews increased external click-through rates by an estimated 8.8 percentage points. That result comes with a caveat worth noting for anyone citing this study: Google changed the HTML structure of AI Overviews during the experiment, which weakened the researchers' blocking mechanism, and the intervention only succeeded in hiding about 51% of AI Overviews, as reported by Search Engine Land.

This builds on earlier work from the same field. A related preregistered study from researchers at the Indian School of Business and Carnegie Mellon University's Heinz College had already found that when an AI Overview appeared, outbound organic clicks fell by 39.8%, zero-click searches rose by 34.5%, and clicks on sponsored results stayed flat, according to PPC Land. Separate coverage of the same study put the organic click decline at 38%, according to Search Engine Journal. Together, the two studies form the most rigorous causal evidence yet that AI search surfaces, not just a correlation with broader behaviour shifts, are actively diverting traffic away from the open web.

Why does this matter for your business?

Most of the traffic-loss data circulating in the industry until now has been observational: before-and-after comparisons that critics could dismiss as confounded by seasonality or algorithm changes. This is different. The study used real random assignment, which means the drop in clicks can be attributed directly to AI Mode itself, not to noise or shifting user habits.

It also undercuts a common defence of AI search features: that users simply prefer instant answers and the trade-off is worth it. The data says otherwise. Removing AI Overviews increased external clicks without producing any detectable decline in users' perception of Google Search, and AI Mode actually reduced trust, usefulness, satisfaction and perceived relevance, with especially strong effects among heavy Google users, according to MarketingProfs. In plain terms, AI Mode is not delivering a better experience that justifies the traffic it takes from publishers. It's simply keeping people inside Google for longer.

For any business that relies on organic search to bring in leads, this confirms a trend that's been building all year. Users who go to AI Mode are less likely to ever reach your product pages, blog, case studies or pricing page, because the answer they wanted already appeared on Google's own interface.

There's also a measurement problem hiding inside this shift. Google Search Console's AI Overview reporting can distort traditional ranking metrics, treating an entire AI Overview as a single block and assigning the same position to every link within it, regardless of where the link actually appears. That means a prominent citation and one hidden behind a "Show more" control can both be recorded as position one, as noted in the same MarketingProfs roundup. Your Search Console dashboard may be telling you your rankings look fine, while your actual referral traffic quietly erodes.

What should you do now?

First, stop treating Search Console position data as a proxy for real visibility. If you're ranking well inside an AI Overview but that citation is buried under a "Show more" toggle, your reported position tells you almost nothing about whether a human will ever see or click your result.

Second, start tracking two separate metrics: how often your brand is mentioned or cited inside AI answers, and how much actual human traffic you receive from search, AI-assisted or otherwise. These numbers are diverging, and conflating them will lead to bad decisions about where to invest content effort.

Third, build content and brand signals that work even when nobody clicks through. If AI Mode answers the question before a user reaches your site, your brand's name, reputation and data still need to be the ones Google's model chooses to cite and recommend. Citation without a click is still worth something, just not the same thing as a visit.

This is exactly the gap that's pushing more founders to check their AI visibility directly rather than guessing from search rankings. Running a free audit with Sited will show you whether AI systems are actually naming your brand, independent of whether anyone clicks through to your site.

Finally, diversify where you expect discovery to come from. Clicks to Reddit fell 21.2 points and clicks to Wikipedia fell 9.9 points in this study, which means even the third-party authority sources that have historically funnelled traffic are being squeezed by the same dynamic. Community platforms, earned media and direct relationships matter more when generic search referral is in structural decline.

Frequently asked questions

Is this study just about AI Mode, or does it also apply to standard AI Overviews?

The main 1,100-person experiment specifically tested forced AI Mode against normal Google Search. A related, earlier study by researchers at the Indian School of Business and Carnegie Mellon found that Google's AI Overviews reduce organic clicks to external websites by around 38% to 39.8% on queries where they appear, so both surfaces show similar click-suppression effects, measured through different methodologies, according to Search Engine Journal and PPC Land.

Did users actually prefer the AI Mode experience, even if it reduced clicks?

No. The research found the opposite: forced use of AI Mode lowered users' satisfaction with Search, and participants were less likely to click external links, ran fewer search sessions, and were more likely to try Bing, DuckDuckGo or Yahoo. This weakens the argument that reduced clicks are an acceptable trade-off for a better user experience, according to Search Engine Land.

Why can't I just trust my Google Search Console rankings to judge my AI visibility?

Because Search Console's reporting wasn't built for AI Overviews. It treats an entire AI Overview as a single block and assigns the same position to every link within it regardless of where the link actually appears, meaning a prominent citation and one hidden behind a "Show more" control can both show as position one. You need separate tools to understand actual AI citation visibility.

Does this mean SEO is pointless now?

No, but the goalposts have moved. Ranking well still matters for being considered as a citation source inside AI answers, but it no longer guarantees traffic the way it used to. Businesses need to measure brand mentions in AI answers and website traffic as two distinct outcomes, not one.

What should small businesses with limited resources prioritise first?

Start by understanding whether AI systems are citing or recommending your brand at all, separate from your Google ranking position. That diagnostic step matters more than any single tactic, because it tells you whether your content problem is visibility, structure, or simply not yet being a trusted source.