Google’s new AI cart can hand your money to a chatbot, and the company says it built a safeguard for exactly that fear

Google is moving forward with a major shift in how people shop online by introducing the Universal Cart, a feature designed to let an AI assistant handle the entire purchasing process. The tool was announced during the company’s I/O event in May 2026, and it is built to act as a personal shopping assistant, cart, coupon finder, and checkout lane all in one. As detailed by Android Police, the move positions Google as the primary layer between shoppers and the rest of the internet.

VP of Ads and Commerce Vidhya Srinivasan described the development as the foundation of agentic commerce. The core idea is that users can drop items into a Universal Cart while browsing the web, and the system will track price drops, price history, and restock alerts on their behalf. The cart follows shoppers across platforms including Search, Gemini, YouTube, and Gmail, so a product spotted in a YouTube review can be added to the cart and then cross referenced with Gemini for compatible accessories.

During the I/O keynote, Google demonstrated a user building a custom PC. The AI identified that a selected motherboard was incompatible with an AMD Ryzen processor and suggested a board with the correct AM5 socket instead. To make this work, Google developed the Universal Commerce Protocol, or UCP, an open source standard created with industry partners including Shopify, Etsy, Wayfair, Target, and Walmart.

UCP allows businesses to avoid building separate connections for every shopping platform

UCP acts as a shared rulebook that lets Google build a shopper’s cart on its own pages and then hand it off to the retailer if the retailer chooses to participate. By standardizing the commerce journey from discovery to final purchase, the protocol helps businesses avoid building bespoke connections for every individual shopping platform.

The system is designed to be vendor agnostic and to work with existing retail infrastructure, as detailed by Google’s developer blog, with businesses publishing their services and capabilities in a standard JSON manifest so agents can discover features, endpoints, and payment configurations without hard coded integrations. It is a structural shift not unlike the platform consolidation Google has pushed elsewhere, including its recent move to sunset a developer focused Gemini coding tool in favor of a single agent platform.

One of the biggest hurdles for AI driven shopping is security, since handing credit card information directly to a chatbot carries obvious risk. To address this, Google uses the Agent Payments Protocol, or AP2.

When a user authorizes a purchase, that permission is turned into a signed digital contract that cannot be tampered with, and because payments are processed through tokenized Google Pay, the user’s actual financial details never sit on the AI platform itself. This preauthorization allows the agent to assemble a cart and finish a purchase even if the user is not actively looking at the screen.

Retailers are already adjusting to the shift, and some industry concern centers on whether having Google handle discovery and comparison could mean fewer site visits and lost opportunities for upsells. Major brands including Nike, Sephora, Ulta Beauty, and Steve Madden are early adopters of the system.

The competitive pressure behind that decision is significant, since any retailer that allows Google’s agents to complete purchases in just a couple of taps puts pressure on competitors who stay out of the protocol. It arrives alongside a broader wave of AI commerce tools that Google has rolled out, including its earlier Gemini AI shopping expansion with major retail partners.

For developers and businesses interested in the technical side, Google has provided a reference implementation that includes a Python server and an SDK for testing. Businesses can set up a server, populate it with product data, and observe how an agent discovers capabilities and invokes a checkout session. The system supports transports including A2A, MCP, and standard APIs, giving businesses flexibility in how they integrate with the protocol.