OpenAI Just Killed Native Checkout in ChatGPT. The Retailers Who Saw This Coming Are Already Ahead.
OpenAI killed Instant Checkout in ChatGPT this week. The Information broke the story on March 6: purchases are moving out of the chat interface and into merchant apps and websites. An OpenAI spokesperson confirmed it, saying they're "evolving our commerce strategy within ChatGPT to better meet merchants and users where they are."
Translation: building a checkout inside a chatbot turned out to be way harder than demoing it on stage.
If you've been in ecommerce long enough, this feels familiar. A platform announces a revolutionary new shopping experience, retailers scramble to integrate, and then the platform quietly walks it back when the operational reality hits. I watched this exact pattern play out with social commerce on Facebook circa 2015. The difference now is that the stakes for product data readiness are permanent, even if the checkout venue keeps shifting.
Why did OpenAI pull back from ChatGPT checkout?
OpenAI's Instant Checkout feature, which allowed users to complete purchases directly within the ChatGPT interface, was shut down in March 2026. The company cited a shift toward routing purchases to merchant apps instead. Only approximately 30 Shopify merchants were ever live on the platform, and core infrastructure for inventory management was never built. The feature faced fundamental technical barriers around real-time product data synchronization.
Forrester published a sharp analysis on March 9 titled "What It Means That The Leader In 'Agentic Commerce' Just Pulled Back." The core finding: completing a purchase inside an answer engine is the least-adopted use case among consumers who actually use ChatGPT for shopping. According to Forrester's December 2025 Consumer Pulse Survey, only 35% of Gen Z, 32% of Millennials, and 23% of Gen X had even used ChatGPT for product search. The people who did use it mostly browsed. Very few bought.
The scale problem was obvious from day one. Only about 30 Shopify merchants were live with ChatGPT checkout, a number Shopify confirmed directly to Forrester. Harley Finkelstein, Shopify's president, said at a recent investor conference that only about a dozen merchants among Shopify's millions were using AI tools to sell. Shopify's ChatGPT-specific landing page now redirects to their homepage. That tells you everything about where this integration ended up on their priority list.
What went wrong with AI-native checkout technically?
AI-native checkout systems require real-time synchronization of pricing, inventory, shipping options, and availability data from millions of retailers simultaneously. OpenAI's Instant Checkout lacked inventory management infrastructure, which Forrester described as "disastrously absent from the plan." Without accurate live data, order failures and customer frustration were inevitable, making the feature unscalable beyond a handful of pilot merchants.
Think about what ChatGPT would need to actually process a transaction reliably. Live pricing from millions of SKUs. Real-time inventory counts. Accurate shipping estimates based on the buyer's location. Tax calculations. Return policies. Every one of those data points changes constantly, and a stale price or an out-of-stock item doesn't just create a bad experience - it creates a chargeback, a support ticket, and a customer who never comes back.
Forrester nailed it when they called inventory management "disastrously absent from the plan." You can build the slickest checkout UI imaginable, but if the product the customer just bought is out of stock, you've created a worse experience than if you'd never offered checkout at all. Amazon spent two decades building the infrastructure to handle this. OpenAI thought they could skip that part.
Should retailers still optimize product data for AI discovery?
Yes. OpenAI's checkout retreat does not reduce the importance of AI-readable product data. ChatGPT, Google AI Mode, Perplexity, Amazon Rufus, and Meta AI all surface product information to consumers through conversational interfaces. The checkout location may shift between platforms, but product discovery increasingly happens through AI. Retailers with structured, accurate, and richly described catalogs will appear in these results regardless of where the transaction completes.
The retailers who spent the last six months waiting for OpenAI to build them a buy button wasted that time. The retailers who spent those same months cleaning up their product data, enriching descriptions, structuring attributes, and making their catalogs machine-readable - those retailers are ahead regardless of where the checkout happens. AI visibility was always the real battleground, not checkout.
Checkout was never the hard problem. Any Shopify store already has checkout. Any retailer with a website already has checkout. What they don't have is product data good enough for an AI to confidently recommend their products over a competitor's. When ChatGPT or Google or Perplexity surfaces a product recommendation, the quality of the underlying data determines who shows up. Bad titles, missing attributes, inconsistent pricing across channels - that stuff was always going to matter more than whether the buy button lived inside ChatGPT or on the merchant's site.
What does Forrester's agentic commerce research say about the future?
Forrester is publishing a formal agentic commerce glossary this month, signaling that the analyst community sees this category solidifying despite OpenAI's retreat. Their research distinguishes between AI-assisted product discovery (growing) and AI-native transaction completion (premature). The data shows consumer adoption of AI shopping tools is real but shallow - most users research through AI and purchase through traditional channels, a pattern that favors infrastructure-level approaches over platform-native checkout.
The pattern emerging from this week's news is clear. OpenAI tried to own the full transaction - discovery, recommendation, and checkout - inside one interface. That approach failed because the operational complexity of commerce at scale is genuinely enormous. Meanwhile, the companies building protocol-level infrastructure (Spreedly, Google's UCP, the card networks) are gaining traction precisely because they don't try to replace the merchant's existing checkout. They just make it accessible to AI agents.
I built Gigya on the premise that identity should be a layer, not a destination. The same logic applies to commerce. When Instacart launched on ChatGPT, they succeeded because they brought their own infrastructure rather than relying on OpenAI to build it for them. Checkout should be a capability that AI agents can invoke, not a feature that every AI platform has to rebuild from scratch. OpenAI learned this the hard way. The payment networks and infrastructure companies figured it out faster.
What should ecommerce teams do right now about AI commerce?
Ecommerce teams should prioritize three things: catalog data quality (structured attributes, accurate inventory, consistent pricing), presence across multiple AI discovery surfaces (not just one platform), and integration with emerging payment protocols like Google's Universal Checkout Protocol. Waiting for any single AI platform to "win" the commerce race is a losing strategy, as the landscape is fragmenting, not consolidating.
If your product data is a mess, fix that first. It doesn't matter whether the next big AI shopping surface is ChatGPT, Google AI Mode, Perplexity, or something that doesn't exist yet. Every one of them will need clean, structured, rich product information to work with. The retailers who treat their catalog as a first-class product - investing in data quality the way they invest in their storefront - will win on every AI surface that emerges.
The checkout button was always a distraction. The real competitive advantage is being the product that AI recommends. As we covered in our analysis of zero-click AI shopping, 93% of AI shopping searches end without a click to a retailer's site. The discovery layer is where the battle is won or lost.
