June 4, 2026·5 min read

Google's AI Search Crossed 1 Billion Users. What That Means For Everyone Else's Website.

TL;DR

Google AI Mode passed 1 billion monthly users about a year after launch, and AI Overviews now reach 2.5 billion monthly users worldwide.
Pew Research found people click a traditional result link on only about 8% of visits to a page with an AI Overview present, versus roughly 15% without one.
Google's Liz Reid argues AI features are filtering out low-value bounce clicks, not real ones. Industry data pushes back on that framing.
A growing body of research, including a Nature paper on recursive training, suggests models trained increasingly on AI-generated web content degrade over generations. That's the harder problem underneath the traffic numbers.

Ask Google a question today and there’s a real chance you’ll get a full answer before you’ve scrolled past the first inch of the page. No click required, no link opened, no site visited.

At I/O 2026 on May 19, Google made that experience the default. Gemini 3.5 Flash took over as AI Mode’s standard model globally, the search box became multimodal enough to accept text, images, files, video, and open Chrome tabs at once, and Google introduced an Information Agent that crawls the web continuously on the user’s behalf. AI Mode hit a billion monthly users roughly a year after launch. AI Overviews, the summary box that now sits above search results by default, reaches 2.5 billion monthly users on its own.

How many people are actually skipping the click now?

Pew Research Center measured it directly: on searches where an AI Overview appears, users click through to a traditional result link about 8% of the time. Without one, that number is roughly 15%. That’s not a small dip. It’s close to half the click-through rate disappearing on exactly the queries where AI Overviews show up.

For a site that depends on search traffic to exist at all, that’s the whole business model getting quietly cut in half, one query at a time, with no single dramatic event to point to.

What does Google say about this?

Liz Reid, Google’s VP and head of Search, has called AI Mode and AI Overviews “the most significant upgrade of the Google Search experience ever” and maintains that internal data shows people searching more, not less, and feeling better about the results they get. Her specific framing on lost clicks is that AI Overviews mostly filter out low-engagement “bounce clicks,” the kind where a user opened a link, didn’t find what they needed, and left within seconds anyway.

If that framing were fully accurate, the click-through drop wouldn’t matter much: those visits weren’t worth anything to the site losing them either. It’s a real argument, not just a talking point, and it deserves to be taken seriously before assuming the worst.

Is that framing actually holding up?

Industry data disagrees with it. Reporting on ongoing search-visibility tracking has found AI Overviews cutting into clicks more broadly than Reid’s “bounce clicks only” explanation accounts for, including on queries where the underlying link was a genuinely useful destination, not a dead end. The Pew numbers don’t distinguish between a wasted click and a valuable one either, so neither side of this argument currently has the granular data to fully prove its case.

What’s not in dispute is the direction. Whether the clicks being lost were valuable or not, fewer of them are happening, and that’s true across a search engine that still controls the large majority of global search volume.

What happens to the web if people stop clicking?

Design, layout, and interaction work only pay off if someone experiences them. A publisher, blogger, or small business site that used to earn a visit now competes for a mention inside an AI-generated summary instead, and there’s no direct revenue in being cited the way there was in being clicked. The economic argument for investing in a website at all gets weaker every time the AI answer alone is enough.

Sites that do get cited by name inside an AI Overview fare better than ones that get summarized without attribution, but citation is a much smaller prize than a visit used to be, and it goes to fewer sites than used to earn traffic the old way.

Is this actually a bigger problem than lost traffic?

Underneath the business-model question sits a harder one. Web content trains the next generation of AI models. If AI-written answers increasingly replace the human-written pages those models used to learn from, the training data itself starts to shrink and homogenize.

A peer-reviewed Nature paper documented this directly: models trained repeatedly on recursively AI-generated data degrade over successive generations, losing the diversity and accuracy present in the original human-authored data. Ahrefs’ 2025 analysis found that 74.2% of newly published web pages already contained AI-generated material, and researchers at Epoch AI have projected the web could run out of new, usable human-written text sometime between 2026 and 2032. Apple researchers separately found that large reasoning models trained under this kind of recursive pattern can suffer what they called complete accuracy collapse on complex tasks.

None of that is settled science yet, and mitigations exist, curated human-only datasets and stricter data filtering both help. But the mechanism is no longer theoretical, and it points at something bigger than any one publisher’s traffic graph: if the web that trains these models is increasingly built out of what the models themselves already said, there’s less and less new ground for either humans or AI to actually learn from.