AI search took the click. It didn't repay the user.
Fewer clicks do not automatically mean a better search experience.
Sometimes somebody gets the answer they need without visiting another website. Search researchers sometimes describe this as good abandonment: the click disappears because the search worked.
A preregistered field experiment involving 1,100 US Google users tested that bargain.
After a three-day baseline, participants spent seven days in one of three search conditions. One group used current Google Search. One had AI features reduced. The third was automatically redirected into AI Mode for almost every search.
Assignment to AI Mode reduced external click-through by 18.8 percentage points compared with current Google Search.
The share of people visiting news sites, Reddit and Wikipedia also fell. Participants conducted 0.92 fewer Google search sessions per day, while spending 0.43 minutes longer inside each session.
That might sound like the platform answered the question and kept people usefully engaged.
It was not accompanied by a better reported experience.
Among the 956 people who completed the post-experiment survey, assignment to AI Mode reduced reported trust, usefulness, satisfaction and agency. It also increased the share using a competing search engine by 11.2 percentage points.
That does not prove that every AI result harms every user.
Evidence limitations
This is a preprint and has not yet been peer reviewed. Participants were US-based primary Chrome and Google users, and the sample skewed younger and more highly educated than the wider US population.
AI Mode was imposed on almost every search rather than chosen for a suitable task. The treatment lasted seven days, so the experiment cannot show how people might adapt over a longer period.
The researchers also had incomplete removal of AI Overviews in the separate No AI condition after Google changed its markup. That problem does not invalidate the cleaner AI Mode comparison, but it makes the No AI findings harder to interpret.
The study is interesting experimental evidence, not a universal verdict on AI search.
But it raises a useful measurement problem.
When a platform absorbs a click, we should not automatically treat the missing click as evidence that it successfully completed the person's task.
Did the person visit the source?
Was the replacement experience actually better?
Those are different things.
A platform can retain attention without necessarily creating more value for the person whose attention it retained.
The missing click is only half the measurement.
Inspect the evidence
The paper reports its design, confidence intervals, preregistration, survey measures and limitations in full.