Google has given online shoppers a new place to leave things they might buy: a cart that follows them between Search and Gemini. It can hold products from several retailers, watch for price changes and, with participating stores, take the shopper through checkout. The US rollout began on 15 October.
OpenAI is taking a different route. In March, it acknowledged that its first version of Instant Checkout had not given merchants enough flexibility. ChatGPT's shopping update put more emphasis on finding and comparing products, with purchases completed through merchants' own checkout experiences.
That is the useful distinction in Google Universal Cart vs ChatGPT Shopping. Both help people choose products, but Google is building a place to keep the decision alive between visits. For a retailer, more of the buying journey could happen before the customer arrives.
Google's cart remembers the unfinished purchase
Imagine someone looking for running shoes during lunch. They save a pair through Google Search, then open Gemini that evening to compare a jacket from another retailer. The items can sit in one Universal Cart, organised by store. Google says it can alert shoppers when prices drop or an item returns to stock.
The feature is currently available in the United States, across Search, Shopping, AI Mode and Gemini. Availability elsewhere should not be assumed from Google's longer-term announcements.
At checkout there are two routes. A shopper can use Google Pay with an eligible merchant, or move to the retailer's site with the cart prepared for them. In either case, Google says the retailer remains the seller of record and handles customer support.
Google has not become the warehouse or returns desk. It has become a more persistent part of the shopping trip. A customer may return to Google's cart for a price alert instead of visiting the shop's product page again.
That possibility is an inference from the design, not a measured change in traffic. There are no public conversion figures in the launch announcement that tell us how often people actually use the new cart.
ChatGPT is competing for the shortlist
ChatGPT Shopping starts earlier, while a shopper is still trying to decide. A request can contain a budget, a preferred style or an awkward combination of requirements. ChatGPT can then show products side by side and help narrow the options.
OpenAI originally introduced Instant Checkout as a way to complete some purchases inside ChatGPT. Its March 2026 update changed the emphasis. OpenAI said the first checkout experience did not provide the flexibility it wanted, and that merchants could use their own checkout systems while ChatGPT focused on discovery.
OpenAI is still investing in commerce infrastructure. It has expanded the Agentic Commerce Protocol, or ACP, to carry product feeds and promotional information into ChatGPT. Shopify merchant data is integrated through Shopify Catalog, while other retailers can connect through supported partners.
Consider a small homeware shop selling a particular kind of desk lamp. A customer might ask ChatGPT for a lamp that fits a narrow shelf, works with a certain bulb and stays under a budget. If the shop appears in the shortlist, it has reached a buyer who may already have ruled out several alternatives. If it does not appear, the store's beautifully designed homepage may never enter the conversation.
Neither company publishes a formula that guarantees a place in the recommendations.
UCP and ACP solve connections, not visibility
The names sound similar, which makes the difference easy to miss.
Google's Universal Commerce Protocol (UCP) is designed to let shopping services and merchants communicate across parts of a purchase. In the Universal Cart flow, Google's Cart API documentation describes a one-way handoff that can create a prepared cart on a retailer's site. The current initial implementation does not provide continuous two-way cart synchronisation. Native checkout is a separate capability.
OpenAI's Agentic Commerce Protocol (ACP) connects merchant information with ChatGPT's shopping experience, including discovery feeds and checkout-related integrations. OpenAI's present emphasis is on product discovery, even though the protocol has a broader commerce ambition.
For developers and commerce platforms, these standards can reduce the number of custom connections they need to maintain.
Supporting a protocol does not mean a product will be recommended. A merchant can make its catalogue available and still miss a shortlist because the product is a poor fit for the shopper's request. Technical eligibility is no evidence of sales.
The retailer still gets the order. What does it lose?
The merchant remains responsible for the product, fulfilment, returns and support. Those are substantial parts of a sale. Introducing the customer does not relieve the retailer of delivering what was promised.
Product discovery has value of its own. A shop once had room to explain why a fabric lasts or how a chair fits a small room. Now those questions may be answered before the visit.
Clear dimensions, compatibility, availability, shipping terms and returns policies therefore matter wherever the comparison happens. Google has announced additional Merchant Center attributes for conversational shopping, including common product questions and compatible accessories.
We do not yet have independent figures showing whether these experiences raise or lower merchants' conversion rates.
What a smaller store can check now
There is no need to rebuild a storefront around a protocol announcement. Start with a few checks:
- Inspect your product data. Prices, stock, variants and specifications should agree with the live store. Conflicting feeds make recommendations less dependable.
- Check the customer journey. Try representative buying questions in the shopping experiences available to you. Record what appears, where the links lead and whether the final price matches. One test is a snapshot, not a ranking study.
- Follow the handoff. In analytics, identify visits from shopping assistants when referral information is available. Compare them with orders, without assuming every AI-assisted sale is identifiable.
- Keep product pages useful. Customers still need accurate sizing, delivery details and after-sales support. Those pages also give you a reliable source to maintain as feeds change.
For independent operators, this is another version of a familiar problem: important parts of a business run through systems the owner does not fully control. ArsKhan's guide to the infrastructure beneath an AI agent examines the connection and identity layers behind that wider shift.
The next contest happens before the buy button
Google has a working US cart that can retain products across its shopping surfaces. OpenAI has chosen to strengthen product comparison while allowing merchants to retain their checkout experiences. Both approaches leave retailers with real responsibilities, and both place an influential part of the purchasing decision on someone else's platform.
Merchants will need to check whether their products are represented accurately when customers begin shopping elsewhere. The companies have shown us the interfaces and the technical connections. The sales evidence is still catching up.