ChatGPT wanted to become the world’s biggest shop. Here’s what actually happened.

It started with a grocery list. During the live demo that introduced the world to OpenAI’s Operator agent in January 2025, a user held up a handwritten sheet of paper in front of their camera. Within seconds, the AI had read it, opened Instacart, built a cart, and scheduled a delivery. No typing. No searching. The audience applauded.

It was a genuinely compelling vision: a world where the friction of online shopping – the tabs, the comparisons, the abandoned carts – simply disappears into a conversation. And for the next fourteen months, OpenAI pursued that vision with unusual aggression for a company that had, until then, stayed out of commerce entirely.

What followed was a sprint of partnerships with some of the biggest names in global retail. Walmart. Target. Shopify. Etsy. PayPal. Salesforce. And, just last week, Amazon – to the tune of $50 billion. The ambition was clear: make ChatGPT the place where people not only discover products, but buy them.

The reality, it turns out, was more complicated. This week, OpenAI confirmed it is stepping back from its most direct bet: native checkout inside ChatGPT. The reason, according to insiders, is painfully straightforward. People browse. They don’t buy.

To understand what OpenAI was trying to build, it helps to think about what traditional e-commerce actually looks like. You type a product name into a search bar. You get a list – sometimes hundreds of results, often cluttered with ads. You click around. You compare tabs. You add something to a cart, get distracted, and maybe come back later. Conversion rates across the industry hover somewhere between one and four percent.

OpenAI’s thesis was that AI could collapse that entire process into a conversation. Ask ChatGPT what running shoes to buy for a half marathon, and instead of a list of links, you get a recommendation, a reason, a price, and a button. One tap, done.

The commercial logic was equally neat. With 700 million weekly users and a growing share of product-related searches, ChatGPT had an audience. Merchants needed reach. The Agentic Commerce Protocol – an open standard co-developed with Stripe – would be the technical glue between them: a universal checkout layer that any retailer could plug into, regardless of their existing payment infrastructure.

When Walmart signed up in October 2025, it felt like a watershed. The world’s largest retailer (at the time), with 270 million customers a week and $681 billion in annual revenue, was betting that conversational AI was the next major commerce channel. Target followed weeks later. Instacart launched a full grocery experience in December. PayPal brought its 400 million wallets to the table. Salesforce connected the product catalogues of L’Oréal, Pandora, Saks, and hundreds of other enterprise brands.

On paper, OpenAI had assembled a retail ecosystem from scratch in under a year. The question was whether real shoppers would actually use it.

The problems emerged quietly, before any public announcement. OpenAI’s internal teams began noticing a pattern: ChatGPT users were asking plenty of product-related questions – comparing laptops, hunting for gift ideas, researching running gear – but they weren’t completing purchases inside the app. They were, in the language of e-commerce analytics, high-intent browsers with low conversion rates.

This wasn’t entirely surprising in retrospect. Shopping is a behaviour built on habits. People know how to buy on Amazon. They trust the checkout flow on their favourite brand’s website. They have saved cards, loyalty points, delivery preferences all embedded in platforms they’ve used for years. Asking them to complete a transaction inside a chatbot (however seamless) required rewiring a habit loop that commerce companies have spent decades reinforcing.

On the merchant side, the barriers were technical and structural. For ChatGPT to show a live price and confirm availability, inventory data needs to be synchronised in real time. For a large retailer with hundreds of thousands of SKUs, complex promotions, and regional price variations, that’s a non-trivial engineering problem. Add fraud prevention – how do you build safeguards against AI-initiated erroneous transactions at scale? – and the integration cost starts to look significant relative to an unproven revenue opportunity.

Harley Finkelstein, President of Shopify, confirmed the gap at an investor conference with unusual candour. Of the millions of merchants on Shopify’s platform, only about a dozen have integrated AI tools for sales. His diagnosis was direct: “The sole limiting factor is that we are just waiting for agent applications to open up further.” In other words, the infrastructure was built. The merchants were watching. The users weren’t converting.

There was also a compliance blind spot that had received little public attention. As of February 2026, OpenAI had not yet built a system for collecting and remitting state sales taxes in the US. For any transaction passing through ChatGPT’s checkout, that was a live regulatory problem and a sign that the underlying commercial infrastructure was less complete than the press releases suggested.

Etsy, notably, had agreed to cover OpenAI’s commission fees on behalf of its merchants to help jumpstart the ecosystem. The gesture underscored how much early partners wanted the model to work – and how much OpenAI still needed to prove.

This week, an OpenAI spokesperson confirmed the strategic shift in plain terms: “We are evolving our commerce strategy within ChatGPT to better meet merchants and users where they are. Instant checkout is transitioning to apps, where purchases can occur more seamlessly.”

Translated: OpenAI is no longer trying to be the place where you buy things. It is trying to be the place where you decide what to buy and then routing you to the apps and websites where you actually complete the transaction.

It’s a meaningful distinction, and not necessarily a failure. Discovery is still enormously valuable in e-commerce. The brands that get recommended by ChatGPT – ranked first when someone asks “what’s the best moisturiser for combination skin” or “running shoes for wide feet under €100” – will see real commercial benefit. OpenAI is, in effect, pivoting from competing with Stripe and checkout providers to competing with Google Shopping and affiliate networks.

The pivot landed days after Amazon announced a $50 billion investment in OpenAI. The deal includes customised AI models for Amazon’s consumer apps and makes AWS the exclusive cloud provider for OpenAI’s enterprise platform. What it doesn’t include, pointedly, is any agreement to sell Amazon products inside ChatGPT – Amazon had previously blocked ChatGPT from scraping its listings, and Andy Jassy has only said he’d collaborate “if the terms were attractive.” The biggest bet in AI history left the most commercially interesting question deliberately unanswered.

Meanwhile, Google has integrated shopping into Gemini, Perplexity has launched agentic purchasing, and Meta is testing commerce inside WhatsApp. OpenAI’s retreat from native checkout doesn’t mean it has conceded the race, it means it’s chosen discovery and influence over transaction and commission. Whether that’s the smarter bet depends on how much of commerce’s value sits in knowing what someone wants, versus processing the payment.

Four things worth acting on now.

First, visibility in AI is the new SEO – according to Adobe Analytics, traffic to retail sites from generative AI tools grew by 693% year-on-year over the 2025 holiday season. If ChatGPT doesn’t recommend your product, you don’t exist in that channel.

Second, the Agentic Commerce Protocol is worth watching: open-source, Stripe-backed, and likely to mature regardless of OpenAI’s own commercial trajectory.

Third, protect your first-party data – brands racing onto external AI platforms risk surrendering the customer relationship to the intermediary.

Fourth, product feed quality will determine AI ranking. Rich titles, real-time pricing, complete catalogue coverage: the fundamentals of good feed management are now a competitive advantage.

The technology worked. The partnerships were signed, the protocol was built, the checkout button was live. What OpenAI discovered is that building the infrastructure for a new commerce channel and changing consumer habits are two entirely different problems. Mobile shopping took eight years to dominate, not two. Voice commerce was “next” for half a decade. The behaviour change always lags the technology.

What’s different this time is the scale of surrounding investment – $50 billion from Amazon, an open protocol from Stripe, enterprise pipelines from Salesforce and Walmart. The foundations are being poured. The floor plan just needed to change.