Google is going all in on AI-driven shopping even as some competitors back off.

At Google I/O, the company unveiled the latest iteration of its AI commerce tools: a “Universal Cart” that works across different retailers and Google products like Gemini — and eventually YouTube and Gmail, too. Users can add products to Google’s universal cart as they browse Search and chat with Gemini and then check out through Google. The cart will also track prices, provide in-stock notifications, suggest potential discounts, and alert shoppers to potential issues with their selections.

Despite the transformative changes AI has brought to the workplace, business, and culture, tech companies are still trying to make the case to the average person that AI can improve their lives or make tedious, unpleasant tasks easier. One place Google thinks that could be is shopping. In November, the company introduced a way for shoppers to dispatch an AI voice to call brick-and-mortar stores to ask about inventory; it also began rolling out a semi-automatic way for shoppers to have AI agents purchase items online on their behalf.

The Universal Cart attempts to corral people’s shopping habits into one place. People shop over the course of days, across different devices and accounts, says Vidhya Srinivasan, vice president and general manager of ads and commerce at Google.

“A lot of the ways I capture this is by having many, many, many tabs open and by syncing profiles and things like that. And it kind of works,” Srinivasan said in an exclusive briefing. “What the shopping cart does from a current problem perspective is it brings all of this together … It is a cart that’s going to be available wherever I am across Google properties.” A cart icon will be displayed next to a user’s profile picture.

Srinivasan envisions the cart almost like a personal shopper working in the background. The Universal Cart works across different retailers, including Sephora, Target, Wayfair, and Walmart, and eventually users will be able to add items to their cart from YouTube or when they see products in Gmail. Once a product is in the cart, users can get price-drop alerts, view price history, and be notified when an out-of-stock item is available again. Srinivasan says the cart — which runs on Gemini — can also alert a user to potential issues with their planned purchases. She gives the example of someone building their first PC choosing a motherboard and processor with incompatible sockets without realizing it; the cart would flag the discrepancy and warn the shopper of potential problems. Shoppers can also connect retailer loyalty programs and credit cards through Google Pay, and the Universal Cart will suggest payment methods and potential ways to save money. If a shopper wants to build a cart but doesn’t want to check out through Google, they can also transfer the contents of their cart to a retailer’s website and finish checkout there.

“The retailer might have other things they want to show the person when they land over there, and they can go deeper in other ways potentially,” Srinivasan says.

Agentic shopping is only possible — and helpful — with the buy-in from a variety of actors: search engines, retailers, payment processors, and so on. Participation from retailers is especially important, considering widespread adoption of agentic shopping could mean customers have little reason to actually visit a store’s website at all (we’ve been calling this “the Doordash problem” at The Verge). Amazon sued AI company Perplexity in November for allowing users to buy products through its Comet AI browser. OpenAI’s efforts at checkout features within ChatGPT were disappointing. As more shoppers use AI chatbots to research products to buy or get recommendations, getting surfaced in AI search platforms is becoming more and more urgent for retailers and brands, which are already tweaking their online presence to try to get chatbots to mention them.

Google seems to know it is getting between merchants and their customers; Srinivasan says the company is “very focused” on the value exchange between all parties. “[Consumers] benefit, but also merchants benefit, because [in the] long run, that’s the only way it works,” she says.

Having billions of products available for purchase within Gemini is great for Google, but retailers need something in return. Srinivasan says Google does not currently take a cut of sales or a commission for products purchased. I asked Srinivasan whether she’s heard concerns from retailers about the idea that Google could become the portal through which shoppers buy things online. She describes Google’s place in the interaction as a “matchmaker.”

“We really want to facilitate lots of consumers talking to lots of merchants,” Srinivasan says. “We don’t want to be the merchant of record.”

On the infrastructure level, there are signs that the retail industry is coalescing around Google. In January, the company announced the Universal Commerce Protocol (UCP), a new open standard it developed with major retailers like Walmart, Shopify, and Target that makes the entire AI shopping journey possible: researching items, putting them in a cart, buying them, paying for them, and getting post-purchase customer service. (OpenAI has its own competing version.) In April, Amazon, Meta, Microsoft, Salesforce, and Stripe joined the committee that governs UCP.

Google previously introduced a way for shoppers to purchase products directly within AI Mode in Search and in the Gemini app, and it’s now expanding into hotel and local food delivery categories. Using Gemini Spark, a new “24/7 personal AI agent” announced by Google, users will also be able to give AI agents more specific guidelines for purchases, like the brands they like, items they’re looking for, and budget. The shopping agent can then make purchases on the shopper’s behalf, provided all criteria are met. A shopper could specify the exact model of a pair of boots they want, for example, set a price limit, and have the AI agent purchase the item when it finds it. The purchases use a technology called Agent Payments Protocol (AP2), essentially a digital paper trail and approval process for having an AI agent carry out a task like buying something.

Shopping is complicated — what happens, for example, when the robot buys something under the price a user specifies, but with tax and shipping it ends up being much more expensive than another option? Would most shoppers trust AI enough to spend their money on their behalf? (Srinivasan tells me Google is currently working through all of this, but that in general, a shopper would go to the actual retailer, not Google, to resolve problems after purchase. Which raises another question: Will retailers introduce policies around purchases made with AI?) Buying things is also emotional: If a rare item on my wishlist pops up for $4 more than I told the chatbot was my limit, I might pull the trigger even if the robot couldn’t. It is hard to imagine a world where shoppers immediately outsource their shopping to a machine: It would be a radical reshaping of what it means to buy things. Most of all, adoption would require an enormous amount of consumer trust — and that is still an uphill battle.