Google's Universal Commerce Protocol is now live with a named enterprise retailer. The checkout-inside-the-chatbot model that failed for Walmart is back, with one critical difference: this time, the merchant agreed to it.

Three weeks ago, Walmart revealed that conversion rates for products sold through Instant Checkout inside ChatGPT were three times lower than products that required users to click out to Walmart's own website. The experiment was scrapped. OpenAI dropped Instant Checkout entirely and rebuilt ChatGPT as a discovery layer that sends shoppers to merchant sites to complete the purchase.

That was March 4. On March 24, Gap announced it would become the first major fashion brand to enable checkout directly inside Google's Gemini. Consumers can order from Gap, Old Navy, Banana Republic, and Athleta without leaving the AI platform. Google Pay handles the transaction. Gap manages fulfilment.

The same week that OpenAI conceded checkout belongs to the merchant, Google put checkout back inside the chatbot, with the merchant's blessing.

Gap's CTO, Sven Gerjets, framed the shift directly to CNBC: "It's not just keyword search anymore. It's conversations, and so we need to be relevant to that."

The integration runs on Google's Universal Commerce Protocol. Product data surfaced to shoppers inside Gemini is not crawled from Gap's website. Gap provides it to Gemini in advance, maintaining control over accuracy, pricing, inventory, and customer data collection. When a shopper decides to buy, the transaction flows through Google Pay. Gap handles shipping and logistics. The consumer never visits Gap's website, but Gap retains ownership of the data and the fulfilment relationship.

This is a fundamentally different architecture from what OpenAI attempted with Instant Checkout. In the OpenAI model, the Agentic Commerce Protocol handled payment delegation and order details through Stripe, but the merchant had limited control over the checkout experience itself. Tax compliance, fraud detection, and basket management all sat in a grey zone between OpenAI and the retailer. As we documented in our analysis of Walmart's Instant Checkout failure, OpenAI had not built systems for collecting and remitting state sales taxes as of February 2026.

Gap's model inverts that dynamic. The retailer controls the product data, the fulfilment, and the customer relationship. Google provides the conversational surface and the payment rails. The protocol layer connects them without abstracting away the merchant's operational infrastructure.

Gerjets told CNBC that Gemini's approach gives retailers better control over the shopping experience than ChatGPT's model offered.

The distinction between Google's Universal Commerce Protocol and OpenAI's Agentic Commerce Protocol is not cosmetic. It is structural.

UCP was announced at the National Retail Federation conference in January 2026, co-developed with Shopify, Etsy, Wayfair, and Target, and endorsed by more than 20 partners including Visa, Mastercard, Adyen, American Express, and Stripe. In March 2026, Google updated UCP with Cart, Catalog, and Identity Linking capabilities: agents can save multiple items to a shopping basket, retrieve real-time product details including variants and inventory, and carry loyalty and membership benefits across UCP-integrated platforms.

UCP also integrates with Google's Agent Payments Protocol, known as AP2, an open protocol developed with more than 60 organisations, including Mastercard, Visa, PayPal, Coinbase, and Worldpay, to securely initiate agent-led payments across platforms. AP2 sits beneath UCP the same way card network rails sit beneath a traditional checkout page: the consumer does not see it, but it handles authentication, tokenization, and settlement.

By contrast, ACP was built with Stripe and originally designed as a full-stack commerce layer. With Instant Checkout gone, ACP now serves a narrower role: feeding product catalogues, pricing, and availability into ChatGPT's discovery layer, and supporting embedded retailer apps. As we covered in our analysis of the agentic commerce standards race, ACP was intended to handle checkout and payment delegation. It still can, in theory. But the only major retailer that tested it in production pulled the plug.

The practical difference shows up in basket management. Walmart's Instant Checkout forced single-item purchases. Every product triggered its own transaction, its own shipping calculation, its own delivery. Consumers rejected it. UCP's Cart capability was designed specifically to solve this: multiple items, one basket, one checkout. Gap shoppers inside Gemini can browse across four brands, build a basket, and complete one transaction.

The comparison is unavoidable, but the differences matter.

Walmart's Instant Checkout failed because the merchant lost control. The checkout sat inside OpenAI's interface. Tax compliance was unresolved. Basket management was broken. The consumer experience was fragmented. Walmart's own data showed that shoppers who clicked out to Walmart.com converted at three times the rate.

Gap's integration keeps the merchant in the loop at every step. Gap controls the product data fed to Gemini. Gap manages fulfilment. Google Pay processes payment, but Gap retains the customer relationship. The product catalogue is not scraped. It is provided directly by Gap, with Gap setting the terms.

There is also a category difference. Walmart's top-performing Instant Checkout categories were supplements, automotive parts, and hardware: items expensive enough individually to clear shipping thresholds. Fashion is different. Clothing shoppers frequently buy multiple items, mix sizes, and build outfits. If UCP's basket management works as designed, fashion is a better test case for in-chatbot checkout than the utilitarian, single-item purchases that dominated Walmart's experiment.

Gap is also deploying Bold Metrics' Agentic Sizing Protocol alongside the Gemini integration. ASP provides personalised size recommendations directly within the conversational shopping flow, at the moment the shopper is ready to buy. Sizing uncertainty is the single largest driver of returns in online fashion. Solving it inside the AI interface, rather than after the purchase, addresses one of the structural friction points that checkout-inside-chatbot models have to overcome.

Zoom out, and the agentic commerce landscape is converging on a single architecture with two variants.

Variant one: discovery inside the AI, checkout on the merchant's site. This is where OpenAI landed after killing Instant Checkout. ChatGPT surfaces products with images, prices, ratings, and comparison tables. When the consumer is ready to buy, they click out to the retailer. Target, Sephora, Nordstrom, Lowe's, Best Buy, The Home Depot, and Wayfair have all integrated into ACP for discovery. Walmart is launching a dedicated in-ChatGPT app with Sparky, its own embedded agent, handling account linking and loyalty integration.

Variant two: discovery and checkout inside the AI, with merchant control. This is what Gap and Google are building with UCP. The consumer never leaves Gemini, but the merchant controls the product data, the fulfilment, and the customer relationship. Google provides the surface and the payment rails.

Both variants agree on the core principle: the merchant owns the transaction. The difference is where the transaction happens. OpenAI sends the consumer to the merchant. Google keeps the consumer in Gemini but gives the merchant the operational levers.

The payment infrastructure beneath both models is converging as well. As we explored in our analysis of how tokenization is making checkout disappear, Visa now has over 17.5 billion tokens globally. Visa's Trusted Agent Protocol and Mastercard's Agent Pay framework bind payment tokens to specific AI agents. Fiserv adopted both protocols, as we covered in our analysis of the processor layer powering agentic commerce. Google's AP2 protocol is designed to interoperate with these frameworks.

The settlement layer does not care whether the consumer clicked out to Gap.com or completed the purchase inside Gemini. The token resolves the same way. The difference is in the user experience, the data flow, and who controls the moment of conversion.

Gap is a test case. If checkout inside Gemini converts better than Walmart's checkout inside ChatGPT did, the UCP model gains credibility and more retailers follow Gap's lead. If it does not, the discovery-only model that OpenAI adopted becomes the default for the entire industry.

Three things will determine the outcome.

First, basket performance. Walmart's Instant Checkout collapsed because it could not handle multi-item purchases. UCP's Cart capability is designed to solve this. Gap's fashion catalogue, where consumers routinely buy multiple items in a single session, is the right stress test. If basket sizes inside Gemini are comparable to Gap's website, the model works.

Second, return rates. Fashion has the highest return rate in e-commerce. If Bold Metrics' Agentic Sizing Protocol reduces sizing uncertainty at the point of purchase, Gap could see lower returns from Gemini transactions than from its own website. That would be a powerful data point.

Third, the trust gap. Forrester's March 2026 ConsumerVoices survey found that completing a purchase inside an AI platform is the least-adopted use case among regular users. PYMNTS Intelligence research indicates 41 percent of consumers have used AI platforms for product discovery, with roughly one third completely replacing their previous search methods. The gap between discovery adoption and checkout adoption remains wide. Gap and Google need to close it.

The infrastructure for handling what goes wrong is also unresolved. As we documented in our analysis of the dispute crisis in agentic commerce, the systems for managing returns, chargebacks, and disputes after an AI-initiated purchase barely exist. Gap absorbs that risk by controlling fulfilment. But as more retailers follow, the dispute layer will need to scale with them.

Walmart proved that checkout inside a chatbot breaks when the merchant loses control. Gap is betting it works when the merchant keeps it. If Gap's conversion rates inside Gemini outperform Walmart's inside ChatGPT, does that vindicate Google's model, or does it just prove that the protocol matters more than the platform?