There is a quiet moment in most digital strategy conversations when someone asks whether the traffic they are seeing from AI-driven sources is real, or whether it will settle back down to something more familiar. I stopped treating that question as rhetorical some time ago. The honest answer, as of Google I/O 2026, is that it is not only real but is now structurally embedded in the most-used search engine on the planet, and the architecture around it changed materially last week in ways that most marketing teams have not yet sat with properly.
The headline figure from Sundar Pichai's keynote at I/O was worth reading twice. AI Mode, the conversational layer Google introduced to Search roughly a year ago, has surpassed one billion monthly active users. Queries have been more than doubling every quarter since launch. Last quarter, total search queries across the product reached an all-time high. The conventional wisdom, repeated with some confidence through 2024 and into 2025, was that AI chat interfaces would cannibalise traditional search volume. That has not happened. What has happened instead is that the ceiling on what people think they can ask has risen, and they are asking more of everything.
Understanding why that matters starts with the change Google made to the search box itself. It sounds modestly described in a press release. In practice, the upgrade is the most significant signal Google has sent about its intent in a generation. The box now dynamically expands as you type. It anticipates intent rather than completing keywords. It accepts text, images, files, video, and Chrome tabs as simultaneous inputs. Liz Reid, Google's VP of Search, described it as the biggest upgrade to the Search interface in over twenty-five years, and that framing is precise rather than promotional. The single-line keyword field was the foundational grammar of the internet economy. Every content strategy, every SEO framework, every PPC account structure built over the past quarter century has been written in dialogue with that box. Replacing it with something that behaves more like a conversation partner than a form field is not an incremental update. It changes what kind of inputs get rewarded. The model now powering all of this is Gemini 3.5 Flash, which Google made the default in AI Mode globally on the day of the keynote. The benchmarks are interesting reading. 3.5 Flash outperforms Gemini 3.1 Pro across almost all standard measures, operates at four times the output speed of comparable frontier models, and does so at less than half the cost. For e-commerce brands, the practical consequence is that the AI reasoning happening inside a search about, say, a dining table or a luxury clothing item is now both faster and more capable than it was last month. The model can hold more context, reason across more signals, and return more nuanced synthesis. That raises the bar on what your content infrastructure needs to deliver if it wants to be legible to that process.
The deeper structural change announced at I/O sits inside what Google is calling information agents. These are persistent AI agents you can configure directly within Search, operating in the background twenty-four hours a day, monitoring the web for conditions that match your stated criteria. The example Google used in the keynote was apartment hunting: you describe your requirements once, and your agent scans continuously, notifying you when listings appear that meet them. The commercial version of that same pattern is straightforward to see. A shopper defining a product specification or a price threshold no longer needs to return to Search repeatedly and hope the right result surfaces. They instruct an agent and wait for the signal.
For brands, this has two simultaneous implications, and it is worth sitting with both rather than rushing to the optimistic one. The optimistic reading is that high-specificity products gain reach they could not previously achieve through passive discovery. If a shopper has a detailed requirement and your product satisfies it precisely, an agent has the patience to find you in a way that a three-keyword query never would. The uncomfortable reading is that brand equity at the awareness layer erodes further. If the agent is doing the comparison work, it is doing so against criteria the user defined, not against the positioning language your creative team spent months crafting. The ability to interrupt and reframe someone's consideration set in the moment of discovery shrinks considerably when there is no moment of browsing to interrupt. The brand names that will hold ground in this environment are the ones whose product data, structured descriptions, and schema implementation are accurate and complete enough to be selected confidently by a reasoning system rather than noticed by a browsing human.
Google also announced what it is calling generative UI in Search: the engine can now construct bespoke visual layouts, interactive tools, simulations, and dashboards on the fly in response to complex queries. A question about a fitness routine could return a working tracker. A question about mortgage options could surface an interactive calculator. A research query could return a structured layout that aggregates, contrasts, and visualises data from multiple sources simultaneously. This capability, built on the Antigravity development platform and scheduled for broad release across Search this summer, means that the concept of a search result as a ranked list of ten blue links is now formally retired. What replaces it is something closer to a commissioned piece of work, assembled specifically for the question asked. The question worth asking now is whether your content, your data, and your structured metadata are good enough to be commissioned.
The Personal Intelligence expansion announced at I/O is easier to contextualise within that frame. Google is extending its ability to connect AI Mode to a user's Gmail, Google Photos, and, shortly, Google Calendar, to more than 200 countries and across 98 languages, without requiring a paid subscription. The engine now knows not just what the web says about a topic but what this specific person has historically cared about, purchased, asked, and saved. Personalisation at that depth changes the competitive surface for e-commerce in ways that are still being mapped out, but the direction is clear: generic content optimised for generic queries has a diminishing return in an environment where every query is being contextualised against the known history of the person asking it.
What I keep returning to, sitting with the full scope of what was announced at Mountain View last week, is that none of this is happening at the margins. AI Overviews now has 2.5 billion monthly active users. The Gemini app has grown from 400 million to 900 million monthly active users in twelve months. Google is processing over 3.2 quadrillion tokens a month across its surfaces, a figure that was 480 trillion at I/O 2025 and 9.7 trillion the year before that. The scale of adoption means these are not features being tested on an adventurous minority. They are the mainstream search experience for an enormous portion of the global shopping population.
For a brand trying to make sense of what an appropriate response looks like, I would offer the same principle I return to in most data conversations: start from the decision, not from the platform. The question is not how to optimise for AI Mode or how to get cited by information agents. Those are outputs. The input is whether your product data is accurate, your descriptions are specific, your metadata is structured, and your content answers the questions your actual customers are asking with enough precision that a reasoning system finds you credible. Brands that have invested in that foundation are already in a stronger position than they know. Brands that have not are now operating with a more urgent timeline than they may have appreciated before last Tuesday.
The search box changed. Everything that feeds into it has to change too.
P.Lorenc - Director of Data & Insights
Editorial Note
Some changes we talked about were live globally on 19 May itself. Gemini 3.5 Flash as the default model in AI Mode rolled out "for everyone globally" on the day of the keynote. The new intelligent Search box began rolling out that same day, "in all countries and languages where AI Mode is available." Conversational follow-ups from AI Overviews also went live "across desktop and mobile, worldwide." And Personal Intelligence (connecting Gmail, Google Photos, etc.) expanded to nearly 200 countries across 98 languages, no subscription required.
The more powerful agentic features are a different story, and they are explicitly US-first for now. Agentic booking for local experiences and services, including Google calling businesses on your behalf, rolls out "to everyone in the U.S. this summer." The generative UI capabilities (custom layouts, interactive simulations) are also coming to Search "this summer" with no specific mention of the UK. A reasonable working assumption would be late 2026 for the UK to see information agents and agentic booking, though that could accelerate if adoption signals in the US are strong.