Do AI Search Summaries Reduce Clicks to Websites?

The Question

If an AI-generated summary answers the searcher’s question directly, does the searcher still visit the websites behind the answer?

For a service business, that question matters because organic traffic has traditionally been one of the clearest signs that search visibility is working. A page ranks, someone clicks it, the website explains the service, and some portion of those visitors call, submit a form, or book.

AI-mediated search can interrupt that sequence.

A business may still be visible. Its information may still contribute to the answer. The customer may even see the company mentioned or cited. But none of those outcomes necessarily means the customer visits the site.

What the Research Examined

Agarwal and Sen conducted a randomized field experiment on live Google Search using a custom browser extension. Participants were assigned to either standard Google Search with AI Overviews or a version in which the AI Overview was removed when it otherwise would have appeared. [1]

Conditional on an AI Overview appearing, the researchers found that outbound organic clicks fell by 39.8% and zero-click searches increased by 34.5%. Sponsored clicks and overall search frequency did not significantly change. The researchers also reported no measurable improvement in perceived search quality or ease of finding information. [1]

That is unusually important evidence because the design was experimental rather than simply comparing traffic before and after AI Overviews became more common.

The research can support a causal statement within the study conditions: showing the AI Overview reduced outbound organic clicking. [1]

It cannot support the much broader claim that AI Overviews reduce every website’s organic traffic by 39.8%.

What the Research Found

The Difference Between a Search-Level Effect and a Website-Level Effect

This distinction is easy to lose.

The 39.8% figure describes outbound organic clicking when an AI Overview appeared in the experiment. It is not a forecast that an individual plumber, law firm, consultant, dentist, or roofing company should expect its total organic traffic to decline by the same amount. [1]

A company’s actual traffic depends on which searches produce AI summaries, how often the business appears in those searches, what kind of intent those queries represent, how the interface changes, and whether customers still need information that requires visiting a website.

The experiment tells us that the search interface can absorb clicks.

It does not tell us exactly how much traffic any one business will lose.

Real Browsing Data Pointed in the Same Direction

Pew Research Center analyzed one month of browsing behavior from 900 U.S. adults covering tens of thousands of Google searches. [2]

Users clicked traditional search-result links less often on result pages containing an AI summary than on result pages without one. Clicking a source cited directly inside the AI summary was considerably rarer, and users were also more likely to end their browsing session after a page containing an AI summary. [2]

This was observational rather than randomized evidence, so it cannot by itself prove that the AI summary caused the difference. Searches that trigger AI Overviews may differ from searches that do not. [2]

The value of the study is that representative browsing behavior pointed in the same general direction as the experimental evidence.

Clicking the Citation Is Not the Default Behavior

One of the more consequential findings for businesses is how infrequently people clicked sources directly from AI summaries in the browsing data. [2]

That should make us cautious about a common assumption in AI-search strategy:

“If the AI cites us, the traffic will follow.”

Citation and visitation are different outcomes.

A business can benefit from being represented accurately in an AI answer even if the user never visits its site, but that benefit is harder to measure through conventional web analytics. The opposite is also true: being cited does not automatically create the same opportunity that an organic click once created.

The Evidence Is Not Completely Uniform

Not every study has found a large reduction in website engagement. [3]

Controlled AI-assisted search research has reported cases where AI-generated summaries attracted greater visual attention without creating a statistically significant difference in downstream website engagement. [3]

That is important because good evidence reviews should not discard findings that complicate the story.

The studies use different interfaces, tasks, samples, and experimental conditions. A controlled task-based experiment is not the same environment as a field experiment in which people use live Google Search for their own needs. [1] [3]

The disagreement does not mean nothing can be concluded.

It means the magnitude of the traffic effect is contextual.

Newer Experimental Evidence Strengthens the Traffic Concern

Additional preregistered field experimentation has examined what happens when exposure to AI search features changes during normal Google use. [4]

Researchers reported that removing AI search features increased click-through to outside publishers, while routing users through more AI-mediated search reduced publisher click-through and affected several user-experience measures. [4]

This work is current academic preprint evidence rather than mature peer-reviewed consensus. It should be classified accordingly. [4]

Its value is not that it gives businesses another percentage to put on a slide. It is that independent experimental work is increasingly testing the same underlying question and finding that generative answers can redirect attention away from external websites.

Generative Search Is Not Just Traditional Search With a Bigger Snippet

There is another reason to avoid thinking about AI search only through the lens of click-through rate.

Peer-reviewed research comparing traditional organic search with generative search systems has found meaningful differences in retrieval behavior, source diversity, reliance on internal versus external knowledge, and output stability. [5]

That means the old model—

rank → impression → click

—is no longer sufficient to describe every discovery path.

Generative systems may retrieve, synthesize, cite, omit, or recombine information before the customer ever encounters a traditional list of pages. [5]

For a business, the discovery layer itself has changed.

What This Does—and Doesn’t—Tell Us

The evidence increasingly supports the conclusion that AI-generated search answers can reduce outbound clicking to external websites, particularly when the answer is presented prominently enough to satisfy some portion of the searcher’s information need. Randomized field evidence supports a causal reduction in organic clicks under the conditions studied, while representative browsing data shows similar real-world patterns. [1] [2] [4]

What the research does not establish is that AI search eliminates the value of a website or makes traditional search optimization irrelevant.

It does not tell us that all queries are equally affected. It does not show that a lower volume of clicks necessarily produces fewer qualified leads. It does not establish whether a citation, brand mention, recommendation, map result, phone call, or direct visit should be valued the same way.

And it does not tell us how to guarantee that a specific business will be cited or recommended by an AI system.

The practical change is more fundamental:

Visibility, website traffic, and business outcomes can no longer be treated as the same thing.

What This Means for Your Website

A business should not respond to AI search by assuming the website matters less.

The opposite may be more useful.

If fewer search interactions result in a website visit, the visits that do happen need to accomplish something the search result or AI summary could not. The website has to provide enough depth, proof, specificity, usability, and business context to justify the additional step.

At the same time, information published on the website may increasingly have value before the customer visits it. Clear service descriptions, location information, company details, proof, policies, and other factual material can become part of the broader information environment from which search and AI systems build answers.

That does not guarantee inclusion.

It does mean the website has two jobs now: serve the visitor who arrives and clearly represent the business to systems that may mediate the interaction before the visit happens.

Track the concepts separately.

Was the business represented in the answer? Was it cited? Did someone visit the website? Did the visit produce a qualified inquiry? Did the inquiry become business?

Those are different stages of the journey.

The measurement infrastructure for some AI surfaces is still immature, so a business may not be able to observe every stage cleanly. That limitation is itself important. If website analytics is the only dashboard you watch, some discovery may now happen outside the dashboard entirely.

A person asking a factual question may be completely satisfied by an AI summary and never need the source page. A person deciding which contractor to hire may still need reviews, service-area details, photos, credentials, pricing context, financing information, scheduling, or a way to contact the business.

Those are different tasks.

The current research demonstrates changes in click behavior. It does not establish that every disappearing click would otherwise have become a customer.

That distinction should affect how performance is evaluated.

The website should help with the parts of the decision that require more depth.

Who actually provides the service? What makes the company appropriate for this problem? What does the process involve? Where does the business operate? What proof supports the claims? What happens next? How can the customer call, book, or request help?

For a service business, those questions are often closer to the transaction than a generic informational answer.

Useful content can still help establish what the business knows, clarify services, answer customer questions, support traditional search visibility, and provide material that discovery systems may retrieve or synthesize.

The objective should not be to publish information solely because it once generated traffic.

Publish information when it helps the customer or accurately represents the business.

That is a more durable standard.

“How do we get the traffic number back?”

Ask what changed underneath it. Which queries lost clicks? Were they informational or commercial? Did qualified inquiries decline too? Did branded searches change? Are calls, forms, bookings, or sales moving in the same direction?

A traffic decline matters when it reduces useful customer opportunities.

Traffic is an input.

The business outcome is the objective.

The Practical Takeaway

AI search can satisfy part of a customer’s information need before that person ever visits a website. Recent field experiments and browsing data suggest that this can reduce outbound clicks, but the effect is not uniform enough to justify treating every lost visit as a lost customer.

For service businesses, the useful response is not to panic about “zero-click search” or chase a guaranteed AI-visibility tactic.

Separate discovery from traffic and traffic from conversion. Make the business easy to understand wherever its information is encountered, and make the website valuable enough to advance the decision when the customer does visit.

The goal is not to preserve every click. It is to preserve the business opportunity behind the search.

Where This Applies at CTG

AI Search

We look at whether the business is represented clearly across the information it controls and distinguish visibility, citation, website traffic, and business outcomes rather than treating them as interchangeable measures.

Local SEO

Local customer journeys can involve traditional results, maps, profiles, AI-generated answers, business websites, reviews, and direct calls. We evaluate discoverability across the path rather than assuming every valuable search interaction has to become an organic website session.

Conversion Optimization

If traffic patterns change, the website still has to convert the qualified demand that reaches it. We use behavioral and business data to determine whether the issue is fewer visits, weaker traffic quality, onsite friction, or some combination.

New Websites

A new website should clearly explain the business even when its information may be encountered indirectly. Services, locations, proof, company details, and next steps should remain understandable both on the site and when information is extracted into broader discovery experiences.