benchr
What Google actually shipped
At I/O 2026, Google described Search moving beyond the summary box:
- Information agents. Users can create and customize agents that track topics and run continuously in the background — designed to keep you informed without re-searching the same thing daily. Google said these launch first for AI Pro & Ultra subscribers in summer 2026.
- Agentic booking. Search is expanding its ability to act on tasks — local experiences and services — where the user shares criteria and Search assembles current pricing and availability.
- A rebuilt Search box. Google called it the biggest upgrade to the Search box in over 25 years, reorganized around AI rather than around ten blue links.
- Gemini 3.5 Flash as the AI Mode default, globally — the same agent-tuned model covered in our look at the current model lineup.
Put together, these describe a layer that takes actions on the user's behalf. That is a different product than a results page with an AI paragraph on top.
The click data, with its sources kept separate
The agent shift sits on top of a trend that is already measurable: when AI answers appear, fewer people click through. The honest way to present this is study by study, because the methods differ.
- Pew Research ran a controlled study of 68,000 queries and found a 46.7% relative decline in clicks when an AI Overview appeared — users clicked a traditional result in about 8% of searches with an AI summary versus 15% without. Clicks on links inside the summaries were rarer still, around 1% of visits.
- Ahrefs (December 2025) reported that the presence of an AI Overview correlated with a 58% lower clickthrough rate for the top-ranking page.
- Seer Interactive, across 2.43 billion impressions, found organic CTR dropped about 61% on queries where AI Overviews appeared.
These are different populations and different definitions, so they do not collapse into a single number — but they point the same way. We keep the deeper breakdown in what zero-click search did to the web.
Why "which pages survive" is the right question
If Search summarizes, the pages that lose are the ones whose entire value was the quick fact a summary now states for free — definitions, conversions, simple how-tos. If Search also acts, a second category is exposed: pages whose value was being the destination for a transaction the agent can now complete inline.
What tends to survive both pressures is content an agent cannot substitute for: original data, first-hand testing, opinion and judgment, primary reporting, and anything that requires trust in a specific author or brand. The strategic response is not to fight the summary — it is to be the source the summary and the agent both have to cite or send a user to. Our guide to getting cited inside AI answers covers the tactics; the short version is to be quotable, specific, and demonstrably first-hand.
Practical takeaways for site owners
- Audit your pages by substitutability. Sort your top URLs into "a summary replaces this," "an agent transacts past this," and "only we have this." Invest in the third bucket.
- Measure impressions, not just clicks. As clicks fall on AI-answer queries, being named as a source still carries brand value. Track citations and brand search alongside organic CTR.
- Treat being a "source" as a goal. Structure content so a model can lift a clean, attributable claim from it — see get cited by AI search and compare how different engines surface sources in AI search engines compared.
- Plan for agentic intent. If users can book, buy, or compare without leaving Search, your conversion path may need to move upstream — into the data and reputation that feed the agent.
Analyst caveat: Google's agent features are rolling out to paid tiers first and on Google's timeline. The click-loss trend is here now; the agent layer is arriving. Plan for both, but do not assume the most aggressive agentic features are universal yet.