Google said its AI Mode crossed one billion monthly users in its first year, swapped Gemini 3.5 Flash in as the default model, and redesigned the Search box for the first time since the early 2000s. Liz Reid, vice president and head of Search, called it “the biggest upgrade to our iconic search box since its debut over 25 years ago.”
Personal Intelligence, the connector that pulls Gmail and Google Photos into Search results, is rolling out free in 198 countries and 98 languages, four months after debuting for paying subscribers only. That global free expansion, more than the redesigned box, is the I/O change publishers and SEO teams will be modeling against this week.
Gemini 3.5 Flash Becomes the AI Mode Default
Reid told reporters at a press briefing that AI Mode queries have “more than doubled every quarter” since launch and hit an all-time high in the last reported quarter. The model swap goes live globally starting Tuesday, according to the company’s I/O Search update post.
This continues an upgrade cadence Google has been running in public. Gemini 3 became the default behind AI Overviews in January. The Gemini app itself moved to Gemini 3 Flash in December. AI Mode was the largest surface still riding an older default; Tuesday closed that gap.
Flash is the cheaper, faster half of Google’s tiered model family. Slotting the newer Flash variant into AI Mode lets the company push higher-end Pro models toward the agentic surfaces announced later in the keynote, where token spend per session is going to climb. Crossing a billion monthly users in twelve months puts AI Mode on a faster ramp than any prior Google product apart from Search itself.
The Search Box Rewrite Lands Globally
Most users will see the box first. It expands dynamically as a query grows, accepts images, files, videos, and open Chrome tabs as inputs, and offers AI suggestions instead of plain autocomplete. Standard blue-link results still render alongside the AI panels, which Google said is by design.
Reid described the change as the literal entry point shifting from a keyword input to a conversation starter:
We are entering the era of Search agents, where you can easily create, customize and manage multiple AI agents.
Rollout begins Tuesday in every country and language where AI Mode is available, which after this week’s expansion is functionally most of the world. A separate change lets users ask follow-up questions directly from an AI Overview, with the conversation context carrying into a full AI Mode session. That bridge ships immediately on desktop and mobile.
The mechanical effect is subtle and important. Users who would have left an Overview to click a blue link can now keep typing inside the same surface, and Google keeps the session.
Search Agents Move Tasks off the Results Page
The agentic block is where Sundar Pichai’s “agent manager” language from April starts taking concrete form. Google introduced three flavors of agent behavior inside Search, with different access tiers and rollout windows.
Information agents run in the background, monitoring blogs, news sites, social posts, and Google’s own real-time finance and shopping feeds, then ping users when something matches a saved brief. Example use cases in the briefing: an apartment-hunting watch and an alert tied to an athlete’s brand collaborations. The feature lands this summer behind Google AI Pro and Ultra, the company’s two paid Gemini subscription tiers.
Agentic booking expands beyond restaurants into local experiences and services. Users give a price band and a calendar window and get back options with live pricing. For categories where merchant inventory is offline, including home repair and pet care, Google said the assistant will phone businesses on the user’s behalf. Booking rolls out to every US user this summer, no subscription required.
Shopping picked up its own set of agentic capabilities, detailed in a separate Shopping post the company published alongside the keynote.
Here is how the agent stack maps against access and timing:
| Feature | Who Gets It | Where | When |
|---|---|---|---|
| Information agents (web monitoring, alerts) | AI Pro and Ultra subscribers | Global, where AI Mode lives | Summer 2026 |
| Agentic booking and calling (home repair, pet care, beauty) | All users, no subscription | United States | Summer 2026 |
| Generative UI in Search | All users, no subscription | Global, where AI Mode lives | Summer 2026 |
| Antigravity mini apps (custom dashboards, trackers) | AI Pro and Ultra subscribers | United States first | Coming months |
Generative UI and Mini Apps Reach Search
The Antigravity platform, Google’s agent-first developer environment, is being threaded into Search. The coding capabilities in Gemini 3.5 Flash let the engine generate a custom visual tool, simulation, or interactive widget in direct response to a query.
Persistent versions of those generations resemble mini apps, the kind a user might return to over weeks for a single repeating task. Tracking a health routine. Managing a household move. Watching a fantasy lineup. Generative UI, the on-the-fly visual response, will be free for everyone in Search this summer. The longer-lived mini apps are starting with US AI Pro and Ultra subscribers in the coming months.
For Search interaction time, the implication is large. A query that previously sent a user to a tracker site or a spreadsheet template now resolves inside Search with a generated dashboard the user can revisit, extending the typical session length and keeping the next several follow-up queries inside Google’s surface.
Personal Intelligence Goes to 198 Countries Without a Paywall
Personal Intelligence is the feature that lets AI Mode read your Gmail and Google Photos library (with Calendar flagged as coming) and weave that personal context into search responses. It launched on January 22 for US AI Pro and Ultra subscribers, expanded to free US users on March 17, and as of Tuesday is live in 198 countries and territories across 98 languages, no subscription required.
Four months separated paid-only beta from global free release. Pricing is the giveaway. Google has decided personalized context belongs in the baseline AI Mode experience and is willing to absorb the inference cost worldwide to keep parity with assistants from OpenAI, Anthropic, and Meta. That is a four-month arc the company would not run if it expected the per-query economics to swing the wrong way at scale.
The connection model is opt-in. Users link Gmail and Photos through a permissions flow, and Google has said the feature does not train on the connected inboxes or photo libraries. The data access surface is still wide, and the read scope (rather than the training question) is what security researchers have been flagging since the January launch.
Robby Stein, Google’s vice president of product for Search, put the value differently at the January launch, saying personalized context lets the system give answers based on what users have already shared rather than forcing repeat explanations. Calendar integration remains undated; once it lands, Search will hold most of the surfaces a personal assistant would need to plan and act on a user’s behalf.
The Click Economy Reads the Memo
Google said queries hit an all-time high last quarter. Independent measurement has been telling a parallel story: queries that show an AI Overview send fewer of those users out to the open web than queries that do not.
The CTR Math Already Moved
An Ahrefs study on AI Overviews and click-through rates published in February analyzed 300,000 keywords drawn from before and after the AI Overviews rollout. The headline numbers from that study and adjacent measurement work:
- 58% reduction in click-through rates for the top-ranking page on queries that trigger an AI Overview, after controlling for general search trends.
- From 7.3% to 1.6% drop in position-one CTR for those queries.
- 33% fall in Google search referrals across 2,500 news sites tracked by Chartbeat over 2025.
- 8% vs. 15%, per Pew Research data on click behavior, was the rate at which users clicked a traditional result on queries with an AI summary present versus without one.
Publishers File Briefs
Penske Media Corporation filed a federal court memorandum in February opposing Google’s motion to dismiss its antitrust suit, citing internal data on the AI Overviews traffic effect. Digital Content Next has reported that most of its member publishers saw traffic declines between 1% and 25%, with some exceeding 75%. About a third of publishers in one industry survey said they would block Google’s AI crawlers entirely if they could do so without breaking organic indexing.
Pichai, on the Cheeky Pint podcast in April, said Search will function as “an agent manager” with users running long-running tasks in parallel rather than browsing results. He pointed to 2027 as the inflection year for non-engineering workflows.
Tuesday’s announcements are the first stack of products that fit that description. A Search session can now begin with a multimodal input, run an agent in the background, generate a custom interface to track the result, and pull from a user’s mailbox without ever directing the user to a third-party site.
If publisher traffic data over the next two quarters shows another step down at the same time AI Mode usage keeps doubling, the antitrust calendar gets shorter and the SEO playbook gets rewritten faster. If clicks hold roughly steady while the new agentic surfaces ramp, Google buys itself the runway to keep building the post-link Search it described Tuesday, on its own clock.
