Google yesterday announced the most sweeping set of changes to its Search product in years, replacing the default AI model inside AI Mode, introducing persistent background agents that monitor the web without user prompts, and rebuilding the search box interface for the first time in more than 25 years.

The announcements came on May 19, 2026, during the Google I/O keynote. They cover five distinct areas: the model behind AI Mode, the search box itself, a new category of autonomous search agents, an expansion of agentic booking to cover local services, and a major extension of Personal Intelligence to nearly 200 countries. Most features require no subscription.

SEO consultant Marie Haynes, who analysed the announcements in a video published on May 20, 2026 at https://www.youtube.com/watch?v=wjxjWqQxiGk, described the significance plainly: "Some of the changes that are happening are not good for some websites and others are really good for those who are on the cutting edge in creating agentic experiences, creating tools that agents can use."

AI Mode and the Gemini 3.5 Flash model upgrade

The most immediate change for anyone using Google Search is a model swap. Google replaced the previous default inside AI Mode with Gemini 3.5 Flash as of May 19, covering every country where AI Mode is currently available.

The speed characteristics matter. According to Marie Haynes, Sundar Pichai stated at I/O that "Gemini 3.5 Flash tends to be about four times faster than most of the Frontier models" and, inside Google's Antigravity infrastructure, "it's 12 times faster than that." Haynes attributed the performance gains in part to Google's custom chip programme, which uses an AI system called AlphaChip to iteratively design chips for both training and inference workloads. One version handles model training; a separate version handles inference - serving the AI at scale.

The speed gain is not incidental. It is what makes persistent background agents technically feasible at the scale Google is now pursuing. Faster inference means an agent monitoring the web for a user's specific question can process, synthesize, and deliver results in a practical window.

On the model quality, Haynes noted from the benchmark data shown at I/O that Gemini 3.5 Flash "for most things is actually better than Gemini 3.1 Pro," adding that it also outperforms competing models in most benchmark categories. That claim - a Flash-tier model surpassing a previous-generation Pro model - is the key technical assertion Google made at I/O, and it underlies the decision to make 3.5 Flash the default across Search.

AI Mode itself has grown rapidly since it first appeared in March 2025 as an experimental feature limited to Google One AI Premium subscribers. Google opened it to all US users without a waitlist on May 20, 2025, then expanded it internationally in October 2025 to more than 40 countries and 35 new languages. According to the company, AI Mode has now surpassed one billion monthly users. Queries in AI Mode run on average three times as long as conventional search queries, and the total volume has more than doubled every quarter since launch.

Elizabeth Reid, VP of Search at Google, described the I/O announcements as "the next step in our journey to bring together the best of a search engine with the best of AI," according to the keynote. Earlier in the same presentation, Reid stated directly: "We are entering the next chapter of Google search where incredible AI features aren't just in search. Google search is AI search through and through." Haynes, citing that statement in her video, noted it "is not going to make a lot of people happy" but reflected the direction Search has been moving for some time.

The search box rebuilt

The visual change most users will notice first is the redesigned search box, which Google calls the biggest upgrade to that interface in more than 25 years. The changes are functional rather than cosmetic.

The box now expands dynamically to accommodate longer queries. It accepts text, images, files, videos, and Chrome tabs as inputs from a single entry point. Haynes described the new interface as "kind of like Google Lens on steroids," noting that the multimodal capability goes well beyond what Lens currently offers. AI-powered suggestions are embedded in the box to help users formulate more complete questions. According to Nick Fox from Google, the search box "offers AI-powered suggestions to help you formulate your whole question" - a step beyond traditional autocomplete that predicts the next word in a short keyword string.

Conversational continuity also changed. Follow-up questions can now be asked directly from within an AI Overview, flowing into a back-and-forth exchange in AI Mode without the user needing to navigate away. Context accumulates across turns, and the sources surfaced become more relevant as the conversation deepens. Haynes observed a concrete consequence of this design: "A lot of SEOs were concerned that Google might announce at Google IO that AI mode is becoming the default and this is not the case but boy are we getting close to that." Her concern is grounded in observed behaviour - standard searches already surface AI Overviews, and follow-up questions pull users directly into AI Mode, where, as she noted, "I didn't have to click on any of these websites."

Search agents: background monitoring without a prompt

The most structurally novel announcement is the introduction of search agents - persistent autonomous agents that run within Search and operate without requiring the user to submit a new query.

The first type is information agents. These run continuously in the background, watching the web for changes relevant to a question a user has previously specified. According to Liz Reed's remarks at I/O, as quoted by Haynes, the agents are "operating in the background 24/7," and they "intelligently reason across information to find exactly what you need at exactly the right moment." Reed also stated that a user's agent will "intelligently look across everything on the web like blogs, news sites, social posts, plus our freshest data such as real-time info on finance, shopping, and sports to monitor for changes related to your specific question," before sending "an intelligent synthesized update with the ability to take action."

The practical scope of this is substantial. A user looking for a specific apartment can describe requirements - location, price, layout, proximity to a metro stop - and the agent scans listings continuously, notifying the user when a match appears. A user tracking a product release receives an alert when the event is confirmed. Neither case requires the user to return to Search. The agent handles the monitoring.

Information agents will initially be available only to subscribers of Google AI Pro and Ultra tiers, launching this summer in the United States.

Gemini Spark

The concept of Gemini Spark, announced alongside the search agent changes, extends the background agent framework further. Spark is a personalized assistant that runs on Gemini 3.5 Flash and integrates with Google Workspace - Gmail, Docs, Calendar, Slides, and more. According to the keynote materials, Spark continues running even when a device is closed, using the Antigravity infrastructure.

Haynes described what Spark represents in practical terms: "Spark represents a big shift for Gemini, transforming it from an assistant that can answer your questions into an active partner that does the real work on your behalf and under your direction." She also noted that Spark "will run with workspace tools, meaning it can run across your Gmail, your docs, your slides, your calendar, and more" and runs "in the background - even when you close your laptop or lock your phone."

Spark can connect to external tools through the Model Context Protocol. Initial MCP integrations include Canva, OpenTable, and Instacart, currently available in the United States only. According to the I/O materials, Spark will ask for user confirmation before taking high-stakes actions - spending money or sending emails - but those confirmations are an explicit acknowledgement that eventually Spark will be able to complete transactions autonomously.

Haynes identified the recurring tasks and learning capabilities as particularly relevant for search professionals: "I will have probably ongoing checks to see if competitors of my clients have created new content or what type of strategies they're taking and then use Gemini to analyze those strategies and create complete workflows."

Google on May 19 also announced that Antigravity's agentic coding capabilities are now integrated into Search results. According to the announcement, Search can build a response "in the right format for your question completely on the fly" - referred to as custom generative UI. Rather than returning a static list of links or a summary paragraph, Search can now generate an interactive interface tailored to the specific query.

The examples demonstrated include custom dashboards and trackers. A query about building a balanced workout and meal plan can produce a structured interactive app rather than a set of articles. As Haynes described it: "Google will make a customized app for you. And this connects using personalized intelligence and also using the features of Gemini Spark to actually converse with you." She noted that such an interface could integrate with services like Instacart to allow a user to order missing ingredients directly.

This builds on a longer-running development. Google launched Gemini 3 with generative UI capabilities in November 2025, establishing the architectural foundation. The I/O 2026 announcement integrates that capability directly into the main Search product and ties it to Gemini 3.5 Flash's speed advantage. Faster inference is what makes on-the-fly interface generation practical at scale.

Alongside generative UI, Google announced that Lighthouse - the web performance and auditing tool used by developers and SEOs - now includes agent monitoring capabilities. One signal it checks for is the presence of an LLMs.txt file. Haynes addressed this directly: "LLMs.txt helps agents understand how to use your website, which eventually will be search, but at this point, it's not something that you need unless you specifically have agents that are using your website." She added context on why Google recently created markdown versions of its developer documentation: "This was not for search reasons. This was primarily to use it in MCPs."

Personal Intelligence expands to 200 countries

The fifth major area from the May 19 announcement is the expansion of Personal Intelligence to nearly 200 countries. Personal Intelligence is Google's framework for incorporating a user's own data - Gmail, Google Photos, YouTube viewing history, and Search history - into AI-generated responses. Two users submitting the same query may receive substantially different answers.

Haynes described the implications for how search works: "Our search results will actually pull from our Gmail. It'll pull from our photos even and that's going to change how search works and what type of results people see. The results that I see are going to be totally different from the results that you see for a lot of queries."

Personal Intelligence had previously been available as a beta for Google AI Pro and Ultra subscribers in the United States only, launching January 14, 2026. The expansion to nearly 200 countries on May 19 makes it broadly accessible without a subscription requirement for most users. PPC Land documented the feature's initial US launch and its implications for personalized signals in AI Mode. The geographic expansion compounds those implications for anyone relying on aggregate keyword tracking.

Universal Cart and the commerce layer

The commerce infrastructure announced at I/O 2026 adds another dimension to the May 19 changes. Universal Cart is a cross-merchant shopping cart spanning Google Search, Gmail, YouTube, and the Gemini app. A user can add products from multiple merchants to a single cart across those surfaces. According to Google, the moment an item enters the cart, it begins checking in the background for price drops, deals, and stock availability.

Haynes described the competitive dynamic this creates: "Somebody might be in search and do a search for one of your products, put it in their universal cart from your store. The cart in the background might say, 'Oh, actually, this competitor has it for cheaper.'" That background comparison happens without additional user action.

At checkout, users can either visit individual merchants or pay across all items using Google Pay. The underlying infrastructure is the Universal Commerce Protocol, which Google launched at the National Retail Federation conference in January 2026 in partnership with Shopify, Etsy, Wayfair, Target, and Walmart. Universal Cart is rolling out in the United States this summer for Search and the Gemini app, with YouTube and Gmail to follow.

The Agent Payments Protocol - AP2 - sits under this as well. Haynes noted: "Google has something called the AP2, the agent payments protocol that is a very secure way for you to pay with agents." With AP2, users effectively pre-authorise an agent to complete purchases on their behalf without producing a credit card at the moment of each transaction.

What this means for search and marketing professionals

The scale of these changes matters for anyone whose work depends on how people find information online. As PPC Land has tracked since the SISTRIX March 2026 data showed click-through rates at position one collapsing from 27% to 11%, AI features in Search have already materially altered the relationship between a ranked position and the traffic it generates.

Background agents introduce a further structural shift. When an agent monitors the web on behalf of a user and surfaces an alert, the user may convert through that alert rather than through any search results page. Haynes framed the problem precisely: "Currently, SEO is optimized for a single snapshot in time - a user typing a query right now. With background agents, you're optimizing for continuous tracking."

The generative UI announcement introduces a comparable challenge. A custom interface generated on the fly for a fitness or planning query is not a ranked result in any conventional sense. Whether a brand or publisher appears inside that generated interface will depend on factors that remain undisclosed.

Google's new guide for optimizing content for generative AI features in Search, published on May 15, 2026, addressed some of these questions. Google Marketing Live on May 20 made clear the advertiser-facing layer is shifting just as rapidly, with AI Mode now carrying ads, new commerce surfaces opening, and Gemini-based tools being pushed deeper into campaign management.

The May 19 announcements do not replace what came before. AI Mode existed before I/O 2026. Generative UI existed before I/O 2026. Agentic booking existed before I/O 2026. What changed on May 19 is that all of these capabilities were tied together under a faster model, connected to background monitoring agents that require no further user input, and extended to a global audience numbering close to 200 countries. Haynes put the underlying direction plainly: "What we've been building for the last 25 years in search is the basis of this AI assistant."

Timeline

Summary

Who: Google announced the changes at I/O 2026 on May 19, 2026. Elizabeth Reid, VP of Search, and Nick Fox, SVP for Search and Maps, presented the search-specific announcements. SEO consultant Marie Haynes analysed the announcements in detail on May 20, 2026 at https://www.youtube.com/watch?v=wjxjWqQxiGk.

What: Google replaced the default AI model in AI Mode with Gemini 3.5 Flash, redesigned the search box for the first time in more than 25 years, introduced information agents that monitor the web in the background without user prompts, launched Gemini Spark as a personalized background assistant integrating with Workspace tools and MCP, integrated Antigravity's agentic coding capabilities into Search results to enable on-the-fly generative UI, expanded agentic booking to home repair, beauty, and pet care services, and rolled out Personal Intelligence to nearly 200 countries.

When: The announcements were made on May 19, 2026, at the Google I/O keynote. Information agents and expanded booking will roll out in the United States this summer. Gemini Spark's initial MCP integrations - Canva, OpenTable, and Instacart - are available initially in the US only.

Where: Most features rolled out globally on May 19 where AI Mode is currently available. Information agents and Gemini Spark will launch first for US-based Google AI Pro and Ultra subscribers. Personal Intelligence expanded to nearly 200 countries simultaneously with the announcement.

Why: The changes reflect a strategic shift in how Google is positioning Search - moving from a tool that responds to individual queries toward a platform that monitors, synthesizes, and acts on behalf of users continuously. For marketing professionals, the concrete implications are: background agents change attribution patterns, generative UI displaces conventional ranked results, Universal Cart introduces real-time price comparison that disadvantages higher-priced sellers, and Personal Intelligence means aggregate keyword tracking no longer reflects what any specific user sees.

Share this article
The link has been copied!