Google’s New Shopping Bet: Letting AI Spend Your Money for You
On Tuesday, Google quietly took a step toward turning its AI into a full-fledged shopper.
The company introduced the Universal Commerce Protocol, or UCP, an “open standard” developed with retailers including Shopify, Etsy, Wayfair, Target, and Walmart. The idea is to let AI agents discover products, negotiate terms, and complete purchases directly inside Google Search and its Gemini app, with Google Pay as the first payment option.
In practice, it means a chatbot could soon go from “find me a carry-on” to “your suitcase arrives Friday” without ever sending you to a retailer’s website.
“Google is trying to own the moment when intent turns into a transaction,” said one retail tech consultant in New York, who asked not to be named because he works with several of the launch partners. “If this works, checkout becomes an API call, not a web page.”
An AI Agent With a Wallet
In Google’s upcoming experience, a shopper might type into AI Mode in Search or the Gemini app: “I need a lightweight suitcase under $200 for a four-day trip.”
Instead of the usual page of links, UCP-enabled merchants would surface products inside a conversational interface. If the shopper chooses one, a native checkout appears in the same chat, already filled in with payment and shipping details from Google Wallet.
No new account. No redirect. No long checkout form.
“Every extra click is a chance to lose the customer,” said a former product manager at a major U.S. retailer. “If the AI can close the sale inside the conversation, that is a direct shot at cart abandonment.”
UCP’s first implementation leans on Google Pay, but Google is stressing that the protocol itself is meant to be neutral and vendor-agnostic. Other payment services, from PayPal to competing digital wallets, could plug in as well. The pitch is that the standard—not the wallet—is what really matters.
Inside the Protocol: A Shared Language for Commerce
Under the hood, UCP is less a new checkout button and more a shared language for how AI agents talk to commerce systems.
Today, retailers and platforms tend to build one-off integrations for every bot, app, or marketplace. UCP tries to standardize that. It defines how an agent and a merchant find each other, describe what they’re capable of, and then actually place an order.
When an AI agent starts a transaction, it sends a profile to the merchant. The merchant responds by checking where their capabilities match: what kinds of orders it can accept, what extensions it supports, which payment handlers it can work with. Only after that “capability negotiation” do they move on to a purchase.
“It’s similar to how your browser and a website agree on what format to use,” said a developer at a UCP partner, who described it as “content negotiation, but for shopping.”
Technically, UCP is designed to run over more than one transport. It can sit on top of traditional REST APIs or Google’s Model Context Protocol, which plugs tools directly into AI systems. Adapters can bridge it to other messaging frameworks. On the payments side, UCP doesn’t replace card networks or processors; it relies on tokenization and existing PCI-compliant handlers so card numbers never touch the AI layer.
Google is also keen to say this is not just “Buy with Google Pay” in disguise. On paper, UCP can operate on any surface—other apps, other agents, and other platforms—whether or not Google controls them.
Why Retailers Signed On
Google says more than 20 partners across the retail ecosystem weighed in on UCP’s design. Shopify, Etsy, Wayfair, Target, and Walmart are among the names Google is already highlighting.
For merchants, the sales pitch is speed and control. Connect to one shared standard, declare what you’re willing to support, and let compatible agents find you rather than rebuilding integrations over and over.
“Retailers learned the hard way that letting one platform define everything is a risk,” said a commerce strategist in San Francisco. “An open standard gives them at least a theoretical say in how agentic commerce works.”
Shopify’s engineers have started explaining how merchants will be able to “declare and define” their capabilities: for example, switching on single-item checkout now and multi-item carts later, once those features land on the protocol roadmap.
The strategic backdrop is less subtle. As more shoppers begin their product searches on Amazon, Google needs a way to pull high-intent traffic back into Search and keep more of the transaction on its own surfaces.
“Whoever owns discovery plus checkout owns the customer,” the strategist said. “This is Google’s way of stitching those together again.”
Security Promises, Unanswered Questions
Letting software agents spend on someone’s behalf is a serious shift, and Google appears to recognize that.
The company says UCP is built on familiar security standards such as OAuth 2.0 for account linking and specialized payment protocols for mandates and verifiable credentials. The goal is to create a clear trail showing who did what at each step of a transaction, from merchant to credential provider to payment service.
So far, though, the public documentation leaves gaps on issues that will matter to regulators and consumers.
Key questions remain:
- How are spending limits defined and enforced for AI agents?
- What explicit permissions must a user grant before an agent can transact on their behalf?
- How easy is it to review, challenge, or reverse what an agent has done with your money?
“There is a big difference between a user clicking ‘buy’ and a system doing it for them,” said a cybersecurity researcher at a Midwestern university. “We haven’t seen clear answers yet about consent, logging, and dispute resolution.”
So far there has been little visible response from consumer advocates or financial regulators in Washington, Brussels, or other major capitals. That quiet may end once the first real-world disputes land over unauthorized or mistaken AI-triggered purchases.
The Competitive Fault Lines
For all the talk of openness, UCP is also an escalation in Big Tech’s ongoing fight over commerce.
Amazon has spent years turning its marketplace into the default search engine for shopping. Apple continues to push Wallet and Apple Pay deeper into apps and Safari. PayPal is courting AI developers with its own agent-focused payment tools. Meta is still trying to make shopping work inside its messaging platforms.
None of them has endorsed Google’s new standard.
“Calling it ‘universal’ doesn’t make it so,” joked one payments executive, who said his company is “watching closely but not committing.”
In his view, the real measure will be whether independent AI agents—ones Google doesn’t control—adopt UCP and start routing serious volume through it.
There’s also the issue of who actually wins if UCP takes off. It’s described as surface-agnostic and suitable for businesses of any size, but large retailers with in-house engineering teams are clearly better positioned to implement a new protocol, validate edge cases, and tune the models that sit on top of it.
“Standards tend to favor whoever has the resources to adopt them first,” said the New York retail consultant. “If you’re a small shop, you might just end up as data.”
A Familiar Gamble, A New Form
Google has tried to reshape online shopping before, with mixed results. Google Wallet has been reworked multiple times. Google Shopping and Google Express both aimed to counter Amazon’s grip on e-commerce. Neither rewired consumer behavior at scale.
This time, instead of building a new destination, Google is trying to erase the destination entirely by folding the store into the AI layer.
The bet is that in a world where people increasingly ask an assistant to “just handle it,” the real power lies in defining the rules that every assistant follows when it reaches for a product, a price, and a payment method.
Whether UCP lives up to its “universal” name or settles into yet another walled garden will depend on how retailers, rivals, developers, and regulators respond from here.
For now, Google’s AI is learning how to shop. The harder question is how, and when, it will learn to stop.
