Gemini Intelligence: agentic search & commerce

Google unveiled Gemini Intelligence at the Android Show on May 12, alongside a new laptop called the Googlebook. The company describes Gemini Intelligence as a layer beneath the Android operating system across laptops, phones, watches, and glasses. The Googlebook is built from the ground up around an AI agent that understands what is on your screen and acts on it: a date in an email becomes a meeting, furniture pieces in an app can be placed virtually in your living room.

When an operating system can complete tasks without users opening a webpage, search, discovery, and commerce change fundamentally. For SEO teams, ecommerce leaders, and GEO strategists, this marks a turning point: visibility alone is no longer enough when AI agents visit websites on users' behalf and complete transactions.

What the shift to an agentic operating system means

Until now, search followed a fixed pattern: users asked a question, typed it into a search engine, received a list of links, and chose one. The entire SEO industry was built around earning that click. Gemini Intelligence assumes something different: search intent remains, but an AI agent handles the middle steps — reading pages, filling out forms, and increasingly completing the entire task. Instead of people visiting websites, agents visit them on their behalf.

One example is Chrome Auto Browse, launched in January and built on Gemini 3. It handles multistep tasks like researching flights, filling out forms, scheduling appointments, and managing subscriptions, then pauses before making a purchase. For ecommerce, the move toward agentic AI is especially relevant because purchase decisions can happen without a classic site visit.

A 2025 preprint evaluated the declared-tools approach across online shopping, authentication, and content management. Pre-structured interaction data cut processing requirements by 67.6 percent and reduced costs by 34 to 63 percent compared with parsing full HTML documents. Task success was 97.9 percent versus 98.8 percent with the traditional method — a sign that machine-readable interfaces are not only more efficient but nearly as reliable.

For marketers, this means the classic funnel of query, click, and conversion is complemented by an agentic path. Goals are stated, agents monitor, compare, and act — often with user approval at the end. Teams that optimize only for organic clicks miss the layer where decisions are made before the first page visit.

The architecture behind Gemini Intelligence

AI agents prefer sites they can transact with cleanly because it is more efficient. Gemini Intelligence only works if agents can reliably perform tasks on websites. Two protocols form the technical backbone: WebMCP makes a site's actions callable, and the Universal Commerce Protocol (UCP) allows an agent to complete a sale. Together, they let agents finish the job without a human loading a page.

WebMCP

This API lets websites declare functions as structured tools an agent can call, such as searching inventory, starting checkout, or submitting a support request. In practice, you hand the agent a labeled menu instead of an unstructured interface. Google co-developed WebMCP with Microsoft. An origin trial is live in Chrome 149, Firefox has committed to the third quarter of 2026, and Safari is expected to follow in the fourth quarter.

Universal Commerce Protocol (UCP)

UCP gives AI agents a common language to discover products, build a cart, complete checkout, and handle orders without a user visiting the site. On the consumer side, Universal Cart collects items across Search, Gemini, YouTube, and Gmail. Google, Shopify, Walmart, Target, Etsy, Wayfair, PayPal, and Stripe co-developed UCP, which launched in January.

How to prepare for agentic AI

Websites are rapidly changing from destinations to backends — from places people visit to infrastructure agents use quietly in the background. The operating system is becoming the search layer. The central question is no longer whether you rank, but whether an agent can use your site.

• Audit actions: Review lead forms, booking flows, and checkout pages to see whether an agent could complete them independently.
• Measure agentic browsing: Use the Lighthouse Agentic Browsing score the way you check Core Web Vitals to see whether agents can use your site, not just read it.
• Check commerce protocols: Ecommerce providers should clarify whether checkout is reachable through UCP or ACP.
• Keep doing retrieval work: Agents must find and trust brands before they act — GEO, structured data, and reliable content remain the foundation.

At the same time, alignment between SEO, product, and engineering pays off. WebMCP implementations, merchant feeds, and API-ready checkout processes are not side projects but direct levers for visibility in agentic search and commerce surfaces. Teams that build machine-readable interfaces and trust signals today position themselves for a world where rankings are the entry point and executable web infrastructure delivers the competitive edge.