OpenAI scrapped Instant Checkout, rebuilt ChatGPT as a visual shopping layer, and handed transactions back to retailers. Gap launched checkout inside Google Gemini the same day. The agentic commerce model is converging fast: AI owns discovery, merchants own checkout. We called it three weeks ago.

On March 4, OpenAI quietly killed Instant Checkout, the feature that let consumers buy products without leaving ChatGPT. Three weeks later, the company replaced it with something fundamentally different: a rich visual shopping experience powered by a version of GPT-5 mini, trained specifically for product discovery. Images. Prices. Ratings. Side-by-side comparison tables. Photo uploads to find similar items. Conversational refinement until the user lands on the right product.

Then the handoff. When the consumer is ready to buy, ChatGPT sends them to the retailer's own store. Target, Sephora, Nordstrom, Lowe's, Best Buy, The Home Depot, and Wayfair have all integrated into the Agentic Commerce Protocol for discovery. Shopify merchants connect automatically through the Shopify Catalog. Walmart is launching a dedicated in-ChatGPT app with account linking and loyalty integration.

On the same day, Gap announced it would become the first major fashion brand to enable checkout inside Google's Gemini, using Google's Universal Commerce Protocol. Consumers can order from Gap, Old Navy, Banana Republic, and Athleta without leaving Gemini, with Google Pay handling the transaction and Gap managing fulfilment.

Two of the largest AI platforms in the world arrived at the same structural conclusion in the same week. The chatbot is the discovery layer. The merchant is the checkout layer. The only question left is who controls the seam between them.

The Prediction That Landed

Three weeks ago, we published a detailed analysis of Walmart's Instant Checkout failure. The data was unambiguous. 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. Shoppers feared that interacting with five product recommendations would produce five separate shipments. The single-item purchase model collapsed the moment a consumer needed two items in one basket.

We wrote at the time that checkout is not a commodity step at the end of a transaction. It is the point where loyalty programmes, bundle logic, upsell mechanics, returns policy, tax compliance, fraud detection, and customer identity converge. Abstracting it away broke the experience.

OpenAI has now confirmed that thesis with its own actions. The company told The Decoder that Instant Checkout "didn't offer the flexibility retailers needed." The new model positions OpenAI as an "upstream discovery layer that sends users to the retailers' existing stores."

That is precisely what the Walmart data predicted. Discovery is where AI adds immediate, defensible value. Checkout requires an entirely different kind of infrastructure, and the cost of getting it wrong is regulatory exposure, chargebacks, and broken customer trust.

Gap and Google: The Same Pattern, Different Protocol

The Gap announcement is significant because it validates the convergence from a different direction. Where OpenAI retreated from checkout and handed it back to merchants, Google built its Universal Commerce Protocol to let merchants control the checkout experience from inside Gemini.

Gap's CTO framed the shift directly: "It's not just keyword search anymore. It's conversations, and so we need to be relevant to that." The product data surfaced to shoppers is not crawled from Gap's website. Gap provides it to Gemini in advance, maintaining control over accuracy, customer data collection, and the overall experience.

The checkout flow runs through Google Pay. Gap handles shipping and logistics. The consumer never visits Gap's website, but Gap retains ownership of the transaction, the data, and the customer relationship.

This is the same structural outcome as Walmart's Sparky model inside ChatGPT, arrived at through Google's protocol stack instead of OpenAI's. Discovery is conversational and AI-mediated. Checkout is merchant-controlled. The protocol layer connects them.

PYMNTS Intelligence research indicates 41 percent of consumers have used dedicated AI platforms for product discovery, with roughly one third completely replacing their previous search methods with AI-based discovery. The demand side is real. The question was always about the transaction layer, and the industry is now converging on the answer.

The Protocol Layer: ACP, UCP, and the Settlement Stack

Two protocols are emerging as the connective tissue between AI discovery and merchant checkout.

OpenAI's Agentic Commerce Protocol, built with Stripe, was originally designed to handle checkout and payment delegation inside ChatGPT. As we covered in our analysis of the agentic commerce standards race, ACP was intended to be the full-stack commerce layer. With Instant Checkout gone, ACP now serves a narrower but potentially more durable role: feeding product catalogues, pricing, and availability into ChatGPT's discovery layer, and supporting embedded retailer apps rather than native checkout.

Google's Universal Commerce Protocol takes a different architectural approach. UCP was designed from the start to give merchants control over the shopping experience. Gap is the first major fashion brand to deploy it, but the protocol is built for scale across Google's surfaces, including Gemini, AI Mode in Search, and the broader Google ecosystem.

Beneath both protocols sits the settlement layer. Stripe powers the payment rails for ACP integrations. For micropayments and machine-to-machine commerce, protocols like Stripe's Machine Payments Protocol and Coinbase's x402 are building the infrastructure for transactions that do not involve a human at all. The discovery-to-checkout split we are seeing in consumer commerce is the first wave. The next wave is agent-to-agent settlement, where the checkout concept disappears entirely.

The protocol layer is where the agentic commerce model becomes infrastructure. ACP and UCP are not competing for the same role. They are converging on the same architecture: AI surfaces the product, the protocol transmits the intent, the merchant closes the sale.

Why Merchants Won This Round

The original vision for AI commerce was that chatbots would become storefronts. The AI platform would own discovery, checkout, and payment. Merchants would become fulfilment providers, shipping products that someone else sold.

That vision failed for three reasons the Walmart data made explicit.

First, checkout is operationally dense. Tax compliance, fraud detection, returns processing, loyalty integration, basket management, and shipping optimisation are not features you bolt on. They are decades of infrastructure investment. OpenAI had not built systems for collecting and remitting state sales taxes as of February 2026, a gap we documented in the Walmart analysis.

Second, consumers trust their existing checkout flows. Forrester's March 2026 ConsumerVoices survey found that completing a purchase inside an answer engine is the least-adopted use case among regular users. Asking questions and researching products ranked first and second. The trust gap in agentic commerce remains real.

Third, the economics favour the merchant. Walmart disclosed that Sparky users have an average order value approximately 35 percent higher than non-Sparky customers. ChatGPT brings Walmart roughly twice the rate of new customers compared to search engines. The value of AI as a discovery channel is enormous, but only if the merchant retains the transaction.

The merchants that prepared their own infrastructure won. Walmart built Sparky. Gap integrated with Google's UCP. Shopify connected its entire merchant catalogue to ACP. The retailers that treated AI as a checkout replacement lost. The retailers that treated AI as a discovery channel and retained checkout ownership gained a new, high-value customer acquisition layer.

The Invisible Checkout Underneath

The discovery-checkout split does not mean checkout stays visible forever. It means the merchant controls when and how checkout becomes invisible.

The infrastructure for invisible checkout already exists. Visa now has over 17.5 billion tokens globally, more than triple its physical card count. Guest checkout has dropped from 44 percent of transactions in 2019 to 16 percent today. Among top merchants, 96 percent of transactions require only a click or biometric confirmation. We mapped this trajectory in detail in our analysis of how tokenization is making checkout disappear, and how the credential layer becomes the commerce layer.

Visa's Trusted Agent Protocol and Mastercard's Agent Pay framework bind payment tokens to specific AI agents. A payment can only be initiated by the right agent, for the right purpose, at the right moment. Fiserv adopted both protocols, as we covered in our analysis of the processor layer powering agentic commerce.

The pattern is clear. OpenAI and Google are building the discovery layer. Merchants retain checkout. The card networks and processors are building the invisible settlement layer underneath. Over time, the merchant's checkout compresses into a single token resolution, invisible to the consumer but fully controlled by the merchant's infrastructure.

The checkout page is not coming back. But who removes it matters. When OpenAI tried to remove it, the experience broke. When merchants remove it on their own terms, backed by tokenized credentials and processor infrastructure, checkout becomes invisible without losing any of the operational layers that make a transaction safe and complete.

What Comes Next

The agentic commerce model has converged faster than anyone expected. Discovery belongs to the AI platform. Checkout belongs to the merchant. The protocol layer connects them. The settlement layer makes it invisible.

Three things to watch.

First, the transparency question. OpenAI has not clarified what criteria determine which products appear in ChatGPT's shopping recommendations, or whether integrated retailers receive preferential placement. As shopping volume scales, the line between discovery and advertising will blur. The merchants paying for ACP integration will expect visibility. The consumers trusting ChatGPT for unbiased recommendations will expect neutrality. Those expectations are incompatible.

Second, the infrastructure gap. Most retailers are not Walmart or Gap. They do not have embedded AI agents, protocol integrations, or basket sync across surfaces. The gap between wanting to participate in agentic commerce and having the infrastructure to do so on your own terms is enormous. The winners in the next 12 months will be the platforms that make this infrastructure accessible to mid-market merchants.

Third, the dispute layer. As we documented in our analysis of the dispute crisis in agentic commerce, the systems for handling what goes wrong after an AI-initiated purchase barely exist. Discovery may belong to the chatbot. Checkout may belong to the merchant. But when something breaks between them, nobody has built the infrastructure to resolve it. That is the next crisis waiting to happen.

Sources

If both OpenAI and Google have conceded that checkout belongs to the merchant, what happens to the retailers that do not have the infrastructure to accept that handoff?

Reply

Avatar

or to participate

Keep Reading