OpenAI Just Blinked: Nobody Seems to Want to Shop Inside ChatGPT
OpenAI is scaling back ChatGPT’s Instant Checkout. Customers wouldn’t buy, merchants weren’t interested, and the operational complexity turned out to be staggering. Here’s why the commerce platform — not the chatbot — will own the transaction, and what this means for agentic shopping.
OpenAI just blinked.
After months of fanfare around “Instant Checkout” — the vision of buying anything directly inside ChatGPT — the company is scaling back its direct shopping plans. Purchases will no longer happen inside the chatbot. Instead, they’ll be routed through integrated retailer apps like Instacart, Target, and Expedia.
An OpenAI spokesperson confirmed the shift: Instant Checkout is moving to Apps, where purchases can happen more seamlessly.
Translation? The transaction is going back where it belongs — to the commerce platform.
We’ve been building marketplace infrastructure at Spree for over 15 years, and this news doesn’t surprise us one bit.
Here’s what happened, what it really means, and why this moment matters for anyone building the next generation of commerce.
What Happened
When OpenAI launched Instant Checkout in September 2025, the pitch was compelling: discover a product through a conversation with ChatGPT and buy it without ever leaving the chat.
The company partnered with Shopify and Etsy, co-developed the Agentic Commerce Protocol with Stripe, and positioned ChatGPT as the future storefront.
It didn’t work out that way.
Multiple outlets have now reported the details, and the picture is clear:
Users browsed but didn’t buy. ChatGPT generated plenty of product-related questions, but conversion rates were dismal. People used the chatbot for research, then went to the platforms they already trust to actually complete the purchase.
Merchants didn’t show up. Out of Shopify’s millions of merchants, roughly a dozen (~12) actually went live with ChatGPT checkouts. Shopify president Harley Finkelstein stated publicly that the bottleneck wasn’t on the merchant side — it was the AI firms that hadn’t opened the doors wide enough.
The infrastructure wasn’t there. As of February 2026, OpenAI still hadn’t built a system to collect and remit state sales taxes — a capability that Amazon, Shopify, and every serious commerce platform has handled for years. Real-time inventory synchronization across millions of retailers proved technically daunting. Fraud prevention added yet another layer of complexity.
The economics just don’t make sense. OpenAI planned to charge merchants a 4% fee on every ChatGPT checkout — on top of existing payment processing and platform fees. For context, competing AI shopping experiences from Google and Microsoft have so far not imposed comparable fees on merchants. When your conversion rates are already low and you’re asking merchants to pay a premium for the privilege, the math collapses fast.
The market noticed. When the news broke, travel stocks surged — Expedia jumped nearly 14% and Booking Holdings rose about 8% — as investors realized AI chatbots wouldn’t be disintermediating booking platforms anytime soon.
Are they right?
What Wasn’t Said (But We Already Knew)
Here’s the part the coverage is missing: the category economics were never going to work for most products. This isn’t just an OpenAI execution problem. It’s a structural one.
We’ve spent 15 years building marketplace infrastructure, and the one thing we know for sure: not every product can support a middleman. AI shopping is no different. Chatbot or marketplace, the category math is the same.
Here’s how it actually breaks down:
AI Takes All — High-Margin, Comparison-Driven Products
Insurance, mortgages, SaaS, travel, financial products, cars. These are categories built for AI-assisted shopping: high margins that can absorb a referral fee, complex comparison decisions where a conversational interface genuinely helps, and customers who value research over brand loyalty in their purchase journey.
Brokers, classifieds, and comparison sites in these verticals should be paying attention. Unless they join the movement, their days of owning the discovery layer are numbered.
AI Discovers — Discretionary, Moment-Driven Shopping
Gifts, fashion, home decor, beauty, specialty food. AI works brilliantly as a discovery engine when someone already knows they want to buy something — a birthday gift, a new outfit for an event, something to refresh the living room.
The intent is there; the AI just helps narrow the field. But the transaction? That still happens on the platform where the merchant controls inventory, fulfillment, and the customer relationship.
AI Arrives Too Late — Replenishment and Staples
Groceries, household supplies, pet food, toiletries. Amazon and Walmart already own this with data, logistics, and the simplicity of “just reorder my usual.”
By the time a customer needs more laundry detergent, the purchase is already automatic. There’s no discovery moment for AI to insert itself into.
AI Is Too Expensive — Low-Margin Commodities
Consumer electronics, appliances, commodity hardware, office supplies. When a product runs on razor-thin margins, a 4% AI tax kills the economics outright.
Not every product can afford a new middleman — and a chatbot charging commission on a $15 HDMI cable or a $200 microwave simply doesn’t work.
The real question was never “Is AI shopping the future?” It was always “Which products can afford it?“
And that question just got a very public answer.
Why the Commerce Platform Matters in the Era of Agentic Shopping
OpenAI’s retreat confirms something that should have been obvious from the start:
commerce is infrastructure, not a feature you bolt onto a chat interface.
Consider what a real transaction requires: real-time inventory data across potentially thousands of SKUs and multiple warehouses. Tax calculation and remittance across dozens of jurisdictions. Fraud detection. Payment processing with PCI compliance. Shipping rate calculation. Returns and refund management. Customer data protection. Loyalty programs. Promotions and pricing rules.
Commerce platforms have spent years — in some cases, decades — building this infrastructure. OpenAI tried to replicate it in months and hit a wall on the basics. They couldn’t even handle sales tax.
This is why Shopify’s Finkelstein was so clear about where the real value lies: shipping, payments, inventory, analytics — that’s what every merchant requires, and that’s where the platform shows up.
The emerging model isn’t AI replacing the storefront. It’s AI handling discovery and recommendation while the commerce platform handles everything that makes a transaction actually work.
The Agentic Commerce Protocol will continue to evolve, but its scope is now narrower: connecting a small pool of large, integrated retailers rather than trying to be a universal checkout for millions of merchants.
In this new model, the commerce platform isn’t just a backend — it’s the backbone.
Why Open-Source Commerce Infrastructure Matters Even More
If the future of agentic shopping depends on the commerce platform, then the question becomes: which platform?
And this is where the proprietary versus open-source distinction becomes critical.
The merchants who went live with ChatGPT’s Instant Checkout were a tiny fraction of Shopify’s base — and they were constrained by what Shopify chose to expose through its APIs. The integration required merchants to build and maintain structured product feeds, integrate with new payment and catalog APIs, and keep pricing, inventory, and shipping data synchronized in real time across an additional channel. Enterprise merchants with dedicated engineering teams could absorb that lift. Everyone else was stuck waiting.
This is exactly the problem open-source commerce infrastructure was built to solve.
With Spree, there’s no waiting for a vendor to expose the right API endpoints. There’s no hoping your platform decides to support the latest agentic protocol. There’s no 4% tax on top of your existing transaction fees because a chatbot decided to sit in the middle of your checkout.
You control the integration. When the next AI shopping protocol emerges — and there will be many — an open-source platform lets you implement it on your terms, at your pace, in a way that makes sense for your category economics.
You control the data. Your product catalog, your customer relationships, your transaction history. In a world where AI agents are going to become a major discovery channel, owning your data is the difference between being a participant and being a commodity.
You control the economics. No surprise fees. No middleman taking a cut that your margins can’t support. You decide which AI channels are worth the integration effort based on your own category math.
What We’re Building at Spree
This is what we’ve been building toward for 15 years: open-source, AI-ready marketplace infrastructure for the categories where the economics actually work.
The next wave of commerce won’t be chatbots replacing storefronts.
It’ll be AI-powered marketplaces built on flexible, open infrastructure — platforms that can plug into any discovery channel (ChatGPT, Gemini, Copilot, or whatever comes next) while keeping the transaction, the data, and the customer relationship exactly where they belong: with the merchant.
We’re building for the operators who understand that the real question isn’t whether AI will reshape commerce — it absolutely will. The question is which products and categories can afford it, and who controls the infrastructure that makes it all work.
If you’re running the margin math on your own categories, we’d love to hear what you’re seeing.
