OpenAI just pulled native checkout out of ChatGPT.
Six months ago they launched Instant Checkout. The pitch was simple: find a product in ChatGPT, buy it without leaving the chat. Walmart, Target, and over a million Shopify merchants were supposed to be on board.
About a dozen actually integrated. Twelve. Out of millions.
Users asked plenty of product questions. 50 million shopping queries a day across ChatGPT. But almost nobody clicked “Buy.” As of February, OpenAI still hadn’t even built a system for collecting state sales taxes. Expedia and Tripadvisor stocks jumped 8% and 13% on the news that OpenAI was backing off.
So what happened?
Half the internet is saying agentic commerce is dead. The other half is saying this means nothing. I think both takes are lazy, because both are focused on the technology. The technology worked fine. The problem is something nobody in AI wants to talk about.
Habits.
People have been buying on Amazon for 20 years. Their credit card is saved. They have Prime. They know what returns look like. They don’t think about any of it. When you go to buy something on ChatGPT, you’re suddenly asking yourself questions you haven’t thought about in a decade. Do I need an account? Is my card saved? What if I need to return this? Where does my confirmation go?
Nobody’s sitting there running a trust calculation on OpenAI. They just go to Amazon because it’s muscle memory.
We ran a consumer survey. 63% of consumers are now using AI to help them shop, up from 49% last year. But only 4% actually completed their most recent purchase on an AI platform. Another 8% clicked through a link in the AI experience to finish the purchase somewhere else. Everyone else went to Amazon or the brand’s site on their own.
63% using AI. 4% buying through it. That gap tells you everything.
Discovery and checkout are two completely different problems. OpenAI won the discovery side. 50 million shopping queries a day. AI-driven traffic to e-commerce sites was up 693% over the 2025 holiday season. That traffic converts at 8x the rate of social. People love using AI to figure out what to buy.
They just won’t buy it there.
Browsing somewhere new is easy. Buying somewhere new means changing a habit. That’s a completely different ask. And OpenAI learned that the hard way.
Go back to any checkout innovation of the last 25 years. Amazon 1-Click. Apple Pay. Shop Pay. Google Pay. Cart abandonment has been stuck at 70% since 2010. Baymard has been tracking it forever. None of those moved the number. Because the problem was never the checkout. It was everything before checkout. People browsing without intent. Sticker shock at shipping. Not being ready to commit.
Instant checkout shaves 2 minutes off a process where someone already spent 45 minutes deciding.
Amazon isn’t winning because their checkout is faster. Amazon is winning because their entire purchasing context is familiar. Saved addresses. Order history. Delivery estimates you actually believe. A returns process you’ve used before. You know exactly what’s going to happen when you click “Buy.” That’s not a UX advantage. That’s 20 years of behavioral conditioning.
People will compare this to voice commerce. Amazon put billions into Alexa. $10 billion loss in 2022. Shopping by voice never got past paper towel reorders. But that’s actually a different failure. Alexa didn’t fail because of habits. Alexa failed because the technology couldn’t handle the job. You can’t browse visually over voice. You can’t compare three products read aloud to you one at a time. The interface was just wrong for shopping. Reordering a known product worked. Anything that required evaluation or comparison fell apart.
OpenAI’s problem is the opposite. ChatGPT is actually good at the hard parts. It can research, compare, explain tradeoffs. 50 million shopping queries a day prove people find that useful. The technology works for discovery. It just couldn’t get people to change where they buy. That’s a habit problem, not a capability problem. Which is actually better news, because habits shift. Bad technology doesn’t fix itself the same way.
And the shift is already being built. Mastercard dropped an open-source authorization standard called “Verifiable Intent” on March 5. Visa has its Trusted Agent Protocol. Google launched the Universal Commerce Protocol at NRF in January with 20+ endorsers including Visa, Stripe, Home Depot, and Macy’s. These aren’t new places to buy things. They’re agent capabilities layered onto your existing Visa card, your PayPal, your Apple Pay.
That’s how habits actually change at scale. You don’t ask people to switch. You make the new thing work inside the old thing. That’s how contactless payments happened. Nobody “adopted” tap-to-pay. The card you already had just started doing it. One day you tapped instead of swiping. That was it.
Agentic commerce will work the same way. Not “buy it on ChatGPT” but “your Visa card now has an agent that watches prices and buys at the right time.” The agent works on top of the habit, not against it.
So what should brands actually be doing right now?
About 60% of e-commerce catalogs have broken data. Missing product identifiers, inconsistent attributes, stale inventory. AI agents need pricing and inventory accurate within a 15-minute window. Most merchants aren’t even close.
When agent capabilities do arrive inside existing payment rails, and they will, the merchants with clean, structured, real-time product data will be the ones agents can actually transact with. Everyone else gets skipped.
The work right now isn’t “get your products into ChatGPT.” It’s clean your data. Structure your catalog so agents can read it. Get your inventory sync under 15 minutes. The merchants who do that over the next 12 months will capture most of the agent-driven demand when purchasing habits finally shift.
OpenAI’s checkout had to fail. You can’t ask people to change how they buy overnight. But the underlying shift is real. AI is already changing how people decide what to buy. How they actually purchase it will catch up. It just has to happen inside the habits people already have, not outside them.