Google I/O 2026: AI shopping cart to unify shopping across all Google services
At I/O 2026, Google introduced the "Universal Cart" – a cross-retailer shopping cart for Search, Gemini, YouTube, and Gmail.
(Image: non c/Shutterstock.com)
Google is increasingly integrating its services with online shopping. The "Universal Cart" presented at the Google I/O 2026 developer conference is intended to function as a central shopping cart, into which users can place products directly from Google Search, the AI assistant Gemini, YouTube, or Gmail – without having to switch between different apps or websites.
The technical foundation is to be Google's Shopping Graph, which Google describes as the most comprehensive product data directory worldwide. It currently comprises over 60 billion product listings that are continuously updated. Google wants to combine this Shopping Graph with its AI models to fundamentally redesign the shopping experience.
As soon as an item lands in the shopping cart, the system is to become active in the background: it will track price changes, display price development over time, and notify when an out-of-stock item is available again. Furthermore, based on Google's Gemini models, the shopping cart is intended to identify potential problems before they arise. As an example, Google cites the assembly of a PC from components from various retailers: the system is to proactively point out incompatibilities and suggest alternatives. Through integration into Google Wallet, payment advantages, loyalty points, and retailer offers are also to be automatically considered.
Checkout and Payment Protocol
For the payment process, Google relies on the Universal Commerce Protocol (UCP), which the company states it developed together with retail partners. A purchase can be completed either directly via Google Pay or by redirecting to the respective retailer's website. The retailer remains the "Merchant of Record." Announced retail partners include Nike, Sephora, Target, Ulta Beauty, Walmart, Wayfair, and Shopify merchants such as Fenty and Steve Madden.
According to Google, UCP is not designed as a purely Google protocol: Amazon, Meta, Microsoft, Salesforce, and Stripe are also to actively shape the open standard within the UCP Tech Council. The UCP checkout is to be rolled out to Canada, Australia, and later the UK in the coming months, as well as to areas such as hotel bookings and food deliveries.
Videos by heise
For fully automated purchases by AI agents, Google has developed the Agent Payments Protocol (AP2). Users are to be able to specify in advance which specific brands and products are permissible and how much can be spent at most. According to Google, a purchase will only be triggered if all defined conditions are met. The protocol is intended to establish a traceable connection between users, merchants, and payment service providers. Additionally, AP2 generates a permanent digital receipt: in the event of returns, users and merchants will always be able to access the same data set, which is intended to simplify return processes. The first AP2 functions are to be introduced in Gemini Spark – Google's new personal AI agent, which is intended to be able to perform tasks independently in the background. With Gemini Intelligence for Android, Google recently introduced agentic capabilities that are intended to, among other things, automatically transfer shopping lists to shopping carts.
The Universal Cart is scheduled to be available in the US in Google Search and the Gemini app in the summer of 2026. Integration into YouTube and Gmail is announced for a later date.
Google's announcement comes in an environment lacking mature standards. However, industry-wide protocols for agent communication are missing, as is a uniform semantic definition of terms – which can promote misinterpretations and thus erroneous purchases. The liability issue for autonomous purchases also remains legally unclear.
Furthermore, there are potential conflicts of interest: commission models or the preference for proprietary ecosystems could distort recommendations – to the detriment of users and smaller retailers who remain invisible to agents without structured data interfaces.
(vza)