When Fewer Clicks Do Not Mean Better Answers

Zero-click search is becoming one of the defining features of AI-mediated discovery. People search, encounter an AI-generated answer, and increasingly remain on the platform rather than clicking through to the websites behind the information.

There is an appealing explanation for this.

Maybe people aren’t clicking because they no longer need to.

The AI answered the question. The user’s intent was satisfied. Fewer clicks, from this perspective, are evidence of a better search experience.

Sometimes, I’m sure that’s exactly what happened.

But there is an important distinction hiding inside that interpretation:

A click is a behavior. Satisfaction is an outcome. The absence of the first is not automatically evidence of the second.

And new experimental evidence gives us reason to take that distinction seriously.

Does Zero-Click Search Mean Users Got Better Answers?

Not necessarily.

A zero-click search tells us what someone didn’t do. They didn’t leave the search platform for an external website.

It doesn’t tell us why.

Perhaps they found a perfectly sufficient answer.

Perhaps the interface reduced their inclination to explore further.

Or perhaps the answer seemed sufficient even though a source would have provided important context.

Perhaps clicking became less convenient.

Perhaps they gave up.

Or perhaps they reformulated the query.

Several different experiences can produce exactly the same observable behavior: no click.

That makes zero-click behavior a poor stand-alone proxy for successful resolution.

What the Google AI Overviews Experiments Tell Us

This is where recent research becomes particularly interesting.

In a 2026 randomized field experiment, Saharsh Agarwal and Ananya Sen compared people’s real Google searches when AI Overviews were available with searches where they were removed.

Conditional on an AI Overview appearing, the researchers found that AI Overviews reduced outbound organic clicks by 39.8% and increased zero-click searches by 34.5%. Yet they found no measurable improvement in perceived search quality or ease of finding information. Engagement with the websites people did visit was also not measurably different between the groups.

In other words, the AI changed clicking behavior. The experiment did not find evidence that the reduction in clicking came with a corresponding improvement in the measured user experience.

Another preregistered field experiment published in August 2026 studied 1,100 Google users. Removing Google’s AI features increased publisher click-through, while directing users into an AI Mode-only experience reduced publisher clicks and also eroded reported user experience and trust in information found through Google.

Neither experiment settles every question about AI search. The second is a recent preprint, and experimental conditions are not identical to every person’s everyday search behavior.

But together they raise a question worth keeping:

What if platform retention and user value are not the same thing?

Platform Retention Is Not User Resolution

This distinction extends well beyond Google.

Every platform has behaviors it can observe.

Clicks. Time on site. Searches. Sessions. Scroll depth. Return visits. Query reformulations.

These measurements are useful. But they are still proxies for what we actually want to understand.

Did the person find what they needed?

Did they understand it?

And did they trust it?

Did they discover something useful they didn’t know to ask for?

Did they leave better equipped to make a decision?

Those are much harder questions.

And this creates what I think of as the Proxy Problem:

When the thing we can easily measure becomes mistaken for the thing we actually care about.

A platform can measure whether someone stayed.

It cannot infer from staying alone that the person’s need was resolved.

Retention tells us where attention remained.

Resolution tells us whether intent was meaningfully satisfied.

Those are different things.

Google Has a Different Interpretation

Google has publicly argued that AI in Search is generating more queries and “higher quality clicks.” It defines a quality click as one where someone does not quickly return to Google, which it interprets as a typical signal of interest in the destination.

Google also acknowledges that some queries can be satisfied directly in Search. If someone wants to know when the next full moon occurs, there may be little reason to visit another website.

That’s reasonable.

Some searches genuinely should require fewer clicks.

But notice how quickly the measurement problem appears.

If someone doesn’t click, we might interpret that as successful resolution.

If someone does click and doesn’t immediately return, we might interpret that as a higher-quality click.

Both interpretations may sometimes be correct.

Neither behavior, by itself, tells us everything about the user’s experience.

The Agarwal and Sen experiment is especially useful here because when AI Overviews were removed, people clicked more, but the additional downstream visits did not show measurably worse engagement.

More clicks did not simply appear to be worse clicks.

Intent Matters More Than Clicks

This is why I keep returning to the idea behind Intent Is the New Audience.

The most useful question about modern discoverability isn’t simply:

Did someone click?

It is:

What were they trying to accomplish?

Sometimes intent is tiny.

“What time does the store close?”

A direct answer may be all anyone needs.

But consider a different query:

“How should I structure compensation for my first executive hire?”

That isn’t simply a request for a fact.

It may require context, tradeoffs, examples, judgment, source evaluation, and follow-up questions.

A concise AI response may reduce the need to click without fully resolving the underlying intent.

This suggests that the future of search measurement needs something more nuanced than celebrating clicks or mourning their disappearance.

We need to distinguish answer delivery from intent resolution.

What Zero-Click Search Means for Experts

For experts, publishers, and creators, none of this means we should design our work merely to win back clicks.

That would be solving yesterday’s problem.

AI search is clearly changing how information travels. Some questions that once generated website visits will increasingly be answered before anyone reaches the source.

Our job is to create work worth discovering beyond the immediate answer.

Original analysis.

Firsthand experience.

Meaningful distinctions.

Deep explanations.

Useful frameworks.

Intellectual perspective.

Connected bodies of work that allow someone to continue exploring once a quick answer is no longer enough.

The work has never been about maximizing traffic for traffic’s sake. The larger goal is helping valuable expertise become findable, recognizable, and useful.

That requires us to care about what happens beyond the metric.

Fewer Clicks Tell Us Something. Just Not Everything.

Zero-click search is real.

AI Overviews can reduce publisher traffic. The causal evidence increasingly supports that conclusion.

But fewer clicks should not automatically be interpreted as evidence that search has become better.

Sometimes fewer clicks will mean the answer was sufficient.

Sometimes they may simply mean the architecture of discovery changed.

Those are not the same outcome.

And as AI increasingly mediates the relationship between questions, answers, sources, and expertise, that distinction will become more important.

Because the real measure of a discovery system isn’t how successfully it keeps someone inside the system.

It’s how successfully it helps them find what they actually needed.