Google’s bounce clicks defence just collapsed under a test

Google has spent eighteen months telling publishers that the clicks AI Overviews replaced were the clicks that didn't matter anyway. Low-intent visits. Bounce clicks. Users who arrived, looked, left. The remaining traffic, the argument went, was higher quality. More qualified. More likely to convert.

A randomised field experiment has now tested that claim directly. When researchers removed AI Overviews from a subset of queries, organic clicks rose 38%. User satisfaction didn't move.

That's the whole story. If Google's bounce-click framing were true, removing AI Overviews would have produced a measurable user experience cost — frustrated searchers, longer task times, more reformulations, lower satisfaction scores. The trade-off is the entire premise. No trade-off, no premise.

This week, Matt Southern at Search Engine Journal traced how Google's public language about AI Overview clicks has shifted across three phases. It's worth reading in full. But the short version is that Google moved from "no data to share" in May 2025, to "the clicks that remain are higher quality" by late 2025, to Liz Reid's "bounce clicks" framing in October 2025 — and now, this week, to a fourth phase: not arguing about clicks at all, just adding more link surfaces to the AI response.

That sequence isn't a communications strategy. It's a retreat.

What the bounce-click argument was actually doing

The bounce-click defence wasn't really about user behaviour. It was a load-bearing piece of rhetoric designed to make publisher traffic loss feel like a quality improvement rather than a value extraction.

The structure is elegant. Publishers complain about losing traffic. Google reframes the lost traffic as low-quality traffic that the user didn't really want anyway. The remaining traffic, by definition, is higher quality. Therefore the AI Overview is doing publishers a favour. Therefore the conversation is over.

It's a clean argument. It just needed to be true.

The randomised experiment is the cleanest possible test. If the lost clicks were genuinely low-value bounce clicks, removing the AI Overview should have produced two effects: organic clicks rise (yes, it found this — by 38%), and user satisfaction falls (because users would now be doing the friction-laden clicking they previously avoided). The second effect is the one that distinguishes "Google is filtering out bad clicks for everyone's benefit" from "Google is keeping the value for itself."

The second effect didn't show up. That's the bit that matters.

Why Google switched arguments this week

Look at what Google announced this week against the timing. Hema Budaraju shipped five new link features inside AI responses. Inline links beside the claims they support. An "Explore new angles" section pointing to related articles. Visible link surfaces, but no new click data attached to any of them.

The pattern is clear if you read it as a sequence rather than a series of product updates. Argument one (no data) held until the publisher data became impossible to ignore. Argument two (higher quality clicks) held until somebody asked for evidence and nobody had any. Argument three (bounce clicks) held until somebody ran the experiment.

Argument four isn't an argument. It's a pivot. Stop debating the click data. Show more links. Let the visible link count carry the rhetorical weight that the click-quality argument can no longer carry.

This is why I keep telling clients that the link expansion is a confession dressed as a feature. When a company stops defending the underlying metric and starts decorating the surface with more visible references, the metric is the problem.

The defence collapsed. The pivot is the tell.

What this changes for how you talk to clients

The practical implication for anyone running SEO for a UK business is narrower than it sounds, but it matters.

For two years, the polite consultant position has been: yes, AI Overviews are reducing clicks, but Google's own framing suggests the remaining traffic is higher quality, so the picture is more complicated than the raw click numbers suggest. That position was always slightly uncomfortable because it required taking Google's unsupported assertion at face value. Now it's untenable. The assertion has been tested and didn't survive.

If you're a publisher, a service business, or anyone whose model depends on search traffic, you can now say plainly: AI Overviews remove clicks, the lost clicks are not disproportionately low-value, and Google's quality-arbitrage defence has been falsified by a randomised experiment. That's a defensible position. It was a defensible position before, but it required more hedging.

The harder thing this changes is what you tell clients about strategy. The bounce-click framing gave clients a comfortable place to land — *yes, traffic is down, but the remaining traffic is better, so we focus on conversion rate and call it a wash*. That story is now harder to tell. You can still focus on conversion rate (you should), but you can't pretend the click loss is a quality improvement.

What you focus on instead is the same thing that's been true since AI Overviews launched. Brand-led demand. Direct traffic. Channels that don't depend on Google's interpretation of search intent. The clients who took that seriously eighteen months ago are in better shape than the ones who waited for Google's framing to clarify. It just clarified, and the news isn't good.

The honest limit

One randomised experiment isn't a final answer. The methodology, sample, query mix, and duration all matter, and any single study can be challenged or replicated with different results. I'd want to see this finding hold across multiple independent tests before treating it as settled.

What's harder to challenge is the rhetorical pattern. Google has now made three different claims about AI Overview clicks across two years. None of those claims came with data. Each one was abandoned when it became too costly to defend. That's a pattern, and the pattern doesn't depend on any single experiment being definitive.

The close

The most useful thing about this week is what it does to the conversation. For two years, the AI Overview click debate has been a debate about claims. Google says the clicks are higher quality. Publishers say the clicks are gone. Neither side can prove the other wrong without data Google won't share.

A randomised experiment broke that stalemate. Not because it's the final word — it isn't — but because it changed what kind of claim each side has to defend. Google's defence used to be a plausible-sounding assertion. Now it's a plausible-sounding assertion that's been contradicted by the only direct test anyone has run.

That's the loop. Defend the metric, then decorate the surface, then change the subject. We're at "change the subject." If you're advising clients about AI search this quarter, plan accordingly.