The headline was too tidy
“Search is over” makes a clean post-keynote take. It also gets the important distinction wrong. Google didn’t announce that the web had vanished from Search; it announced a more AI-forward interface, new agent features, and ways to continue a search as a conversation. Those changes matter. They aren’t the same thing as turning off the web.
The real concern is less dramatic and more useful to name: people may need to visit fewer pages to get what they came for. That’s a traffic question, not proof that SEO stopped working overnight.
Google’s own I/O post says the new Search box accepts text, images, files, videos, and Chrome tabs, and is designed to help people describe what they want. It also says users will continue to get a range of results. A follow-up from an AI Overview can move into AI Mode, with Google saying links and supporting articles become more relevant as the conversation goes deeper. That is not a promise that every publisher gets a visit. It is a description of a Search experience in which AI and links sit together, with the balance changing by query and interaction. (Google)
Search is changing shape, not dropping the web
The Search Engine Journal account of the reaction describes the immediate leap from a new AI-heavy interface to claims that the ten blue links were finished. Google’s response, as reported there, was that traditional results remain accessible, including through the Web tab; Liz Reid also said people would still get a range of results. The distinction is worth holding onto. Results can remain available while becoming less central to the default experience.
That’s a meaningful change for SEO, even if the basic work of making useful pages discoverable remains. Placement, attention, and the likelihood of a click are not guaranteed just because a result still exists somewhere in the product. “The links are still there” is not a traffic forecast.
The product announcements make that tension plain. Google says AI Mode is receiving Gemini 3.5 Flash as its default model globally. It describes a redesigned search box with AI-powered suggestions, multimodal inputs, and a conversational path from AI Overviews. Those are interface choices that encourage richer prompts and follow-up exploration rather than a one-query, one-results-page habit. Google’s explanation is that links and supporting articles can become more relevant as a user explores. That may be good for some kinds of discovery; it doesn’t settle how often people will click through.
Agents change the shape of the visit
The most consequential idea may be the least familiar: information agents that monitor the web in the background and send synthesized updates. Google’s announcement gives examples such as tracking apartment listings against a user’s requirements or watching for a product update. The company says these agents will launch first for AI Pro and Ultra subscribers in the summer. Its post also describes booking capabilities that assemble pricing and availability with direct links to providers, plus other task-oriented features.
There’s a real difference between a person searching, opening a page, and deciding what to do, and an agent checking information and reporting back. In the second case, a source could inform the answer without generating a conventional visit. That possibility is the heart of the traffic risk. It is not evidence that every agent will bypass publishers, or that every query will become agent-mediated. Google’s public description doesn’t establish that outcome.
Still, it changes what a publisher should be watching. Search visibility has never been identical to site traffic, but agent-mediated summaries make the gap easier to see. A page might be useful to a system while the user never opens it. That’s an awkward measurement problem, not a reason to declare the whole discipline dead. For a related discussion of measuring brand presence beyond the click, see our guide.
The economics matter more than the slogan
Search Engine Journal’s analysis argues that straightforward, widely repeated information is more exposed when AI can provide a direct answer. It contrasts that with original data, analysis, and expertise that cannot simply be reconstructed from a bundle of similar pages. Treat that as an editorial argument, not a guarantee that distinctive work will always earn a citation or a click. The point is sound: when an answer can be synthesized without much loss, the user has less reason to visit another page.
That puts pressure on the familiar content treadmill. Rewriting what is already easy to find may still serve a reader in some contexts, but it gives an AI answer little reason to send that reader onward. Original reporting, first-hand experience, and genuinely useful tools offer a clearer reason for a source to matter. None of that is a magic ranking formula. It is a better bet than producing another page whose only distinction is a changed headline.
The uncomfortable part is that visibility and value can diverge. A publisher might be represented in an answer yet see no corresponding click. Or a result may remain available but receive less attention because the interface has given the answer more room. The sources here describe those as risks and emerging product behavior, not measured universal outcomes. Keep the distinction. Forecasts shouldn’t be dressed up as settled analytics.
Measurement is not keeping pace
The SEJ analysis also says Search Console does not provide specific filters to separate AI Mode or AI Overview activity from organic reporting, making it difficult to isolate the impact on a particular site. It raises a further concern about agents: if they consult content and send a synthesis without a visit, ordinary site analytics may not record that consumption as traffic.
That matters because teams are being asked to respond to a changing search experience without a clean view of what changed. Overall impressions and clicks can tell you something about performance, but they don’t necessarily identify which AI surface caused the movement. And a missing visit doesn’t tell you whether a page was ignored, cited in a way that didn’t drive a click, or used by an agent. Don’t claim more certainty than the reporting supports.
For now, keep the measurement practical. Track the outcomes you can see, annotate major Search changes, and avoid treating a single shift in clicks as proof of one cause. Keep an eye on referral patterns and on whether your distinctive pages continue to attract direct interest. That won’t solve attribution. It will stop a team from confusing a plausible story with a demonstrated one.
What I’d do next
Don’t throw out SEO because a conference keynote made the interface look different. Google’s own announcement says a range of results remains, and the reporting on its response says traditional results remain accessible. The work of helping pages be understood and found still has a place. But doing that work while ignoring how AI answers reshape user behavior would be equally careless.
Review the pages that simply restate common facts. Ask what a user gets from your version that they couldn’t get from a short synthesis of many similar pages. If the honest answer is “not much,” improve it or stop multiplying it. Invest instead in material with a specific point of view, first-hand knowledge, original evidence, or a task the page genuinely helps someone complete. Then measure carefully—and admit what you can’t yet isolate.
The loud reaction after I/O was that Search had been replaced overnight. That’s not what Google said. The sharper concern is that Search can keep using the web while sending fewer people to it for some needs. That possibility deserves attention without turning it into a premature obituary for SEO.
Sources
- Google: Search updates announced at I/O 2026
- Search Engine Journal: Google I/O, SEO, and where the risk lies
the headline was too tidy