5 changes in Google search, explained
Five products inside a single Google Search overhaul. For each one: what it does, who gets it, when, free or paid, and what the headlines leave out.
On May 19, Sundar Pichai — Google's CEO — opened Google I/O 2026, the company's annual product conference, with what many headlines rightly called "the death of the search bar." Behind that line are in fact five distinct product launches. Each one has its own price, its own region, its own timeline, and its own dependency on Gemini, Google's family of AI models. For anyone whose business depends on what users see when they type into Google, the detail matters.
We spent the day reading the keynote, the blog.google product post, the 9to5Google breakdown — a US site that covers the Google ecosystem — and a handful of secondary reports. Here is what Google actually announced, product by product, in order of rollout speed.
The intelligent Search box (US, May 19, free)
The search bar itself is the most visible change. It now expands as you type to accommodate full sentences instead of keyword fragments. It accepts text, images, uploaded files, and even content from open Chrome tabs as input modalities. It surfaces AI-powered suggestions that go beyond autocomplete, attempting to anticipate the intent of the query and reformulate it.
This ships starting May 19, US first. There is no opt-in — no manual toggle to flip. Every US user lands on the new search bar the next time they open google.com. No date has been announced for other regions, and Google did not commit to one in the keynote.
AI Mode, now globally default on Gemini 3.5 Flash (free)
AI Mode is the conversational layer above traditional search. Google launched it in 2024 as an opt-in beta — you had to switch it on yourself. It crossed one billion monthly active users in early 2026 and is now the default conversational layer behind the new search bar.
The model under the hood has been swapped: Gemini 3.5 Flash replaces the prior 2.5 generation. Faster, cheaper to serve, more accurate on multi-turn reasoning according to Google's internal benchmarks. The upgrade is global. Personal Intelligence inside AI Mode, the feature that adapts answers based on your Google account data, is expanding to roughly 200 countries and 98 languages. Free. No subscription.
The AI Mode change matters more than the visible search bar redesign. The bar is the front door. AI Mode is the engine that decides what you read.
Generative UI (free for all, this summer)
This was the most surprising piece of the keynote. Generative UI is a system in which Gemini 3.5 Flash builds custom widgets on the fly to match a query. Ask about mortgage rates, you get a live calculator embedded in the response. Ask about a hiking trail, you get an interactive map with elevation data. Ask to compare two products, you get a generated comparison grid.
The widgets are not pulled from a library of templates. They are rendered from scratch each time, by the model, for the specific question. This is a fundamental departure from the rich-snippet model where Google pulled structured data from your site and rendered it in a fixed card.
Free for all users, rolling out this summer. The implications for organic traffic are not small: a query that previously sent the user to a calculator on your site is now answered inside Google by a freshly-generated calculator. The click never leaves Google.
Information agents (paid, US, this summer)
The agent layer is what Pichai, Google's CEO, positioned as Google Search's most strategic addition. An information agent is a scoped, persistent task the user creates inside Google Search: track apartment listings under 1500 dollars in Brooklyn, alert me when this concert ticket drops in price, summarize what changed in this regulation each week. The agent runs in the background. It surfaces results when the criterion matches.
At launch, restricted to Google AI Pro and Ultra subscribers in the United States. AI Pro is 20 dollars per month, Ultra is 250. The 24/7 background processing is what justifies the paywall: Google is selling continuous compute, not one-shot answers.
No timeline for non-US rollout. No timeline for a free tier.
Mini-apps via Antigravity (paid, coming months)
The most experimental piece. Mini-apps are custom, lightweight applications a user can scaffold directly inside Google Search using Antigravity, Google's no-code agentic builder. Examples cited in the keynote: a wedding planning dashboard, a home-move tracker, a personal medication schedule.
The model writes the app from a natural-language brief. The app persists across sessions, can be edited, can be shared. Paid tier only, AI Pro and Ultra, US first, no firm date beyond "coming months."
What Google did not say
Three silences worth noting.
On ads: not one slide on how generative answers, agents, and mini-apps will be monetized. AI Mode currently shows no ads. Generative UI hasn't been clarified. The ad revenue model that built Google is unaddressed.
On publishers: the keynote did not show data on click-through rates after the redesign. Ahrefs measured in December 2025 that an AI Overview cuts the top organic CTR by 58 percent. The new generative UI will likely worsen that, but Google did not commit to a measurement.
On citation surfacing: the agents and the generative widgets compose answers from sources, but Google did not detail the citation logic. Which sources are surfaced, how prominently, with which click affordance, is left vague.
For an operator whose visibility depends on Search, the practical takeaway is in our previous piece on Google ending its search bar: the marketing minutes that used to feed long-tail SEO now feed name recognition, and name recognition lives on the feeds, not on the result page.