OpenAI Killed Its Shopping Feature.

So here's what happened. Last September, OpenAI rolled out native checkout inside ChatGPT. The pitch was elegant in that very Silicon Valley way where everything sounds inevitable right up until it isn't: people are already asking ChatGPT which air fryer to buy, so why not let them buy it right there? Research, recommendation, purchase — one seamless interface. The future of commerce, delivered.

Five months later, that future has been politely escorted off the premises.

According to The Information, OpenAI is pulling back from processing transactions directly and instead routing purchases through third-party apps like Instacart, Target, Expedia, and Booking.com. If you want to actually buy something ChatGPT recommends, you'll now be gently redirected to the retailer's own platform to complete the transaction there. The buy button era lasted roughly as long as my New Year's resolution to stop checking my phone before 8am.

The core problem is almost embarrassingly simple: people were using ChatGPT to figure out what to buy, but they weren't actually buying it there. Which, if you think about it, is the AI equivalent of that friend who spends forty-five minutes helping you pick a restaurant and then says they already ate. The discovery was working. The conversion was not.

And for OpenAI, this wasn't just a feature experiment — it was supposed to be a revenue stream built on sales commissions. Without people actually crossing the line from "ooh, that looks good" to "here's my card number," the whole business case collapses like a soufflé in an earthquake.

The Ghosts of Buy Buttons Past

Here's the thing that should make everyone in ecommerce sit up slightly straighter: OpenAI isn't even the first company to faceplant on this exact idea. Facebook poured resources into Facebook Shops and native checkout, then quietly retreated to driving traffic to external retailer sites. Google launched Buy with Google, gave it the full keynote treatment, and later wound it down. The pattern is so consistent it should probably have its own Wikipedia page.

The operational reality is brutal. OpenAI had to onboard each retailer individually — a process so labour-intensive they'd only managed about a dozen Shopify merchants by launch. Inventory synchronisation across thousands of different catalogues is a technical nightmare that, to date, only Google Shopping has managed at anything approaching scale (and even that comes with a generous definition of "managed"). As of February 2026, OpenAI hadn't even built systems to collect and remit state sales taxes. Which is one of those details that sounds boring until you remember that tax authorities have famously limited senses of humour.

Shopify's president Harley Finkelstein made an interesting observation at an investor conference: the bottleneck was on the AI platform side, not the merchant side. Retailers were willing to show up. The infrastructure to support them just wasn't there. It's like throwing a party, sending out all the invitations, and then realising you forgot to build the house.

The Behaviour Problem Nobody Wants to Talk About

But here's where it gets genuinely interesting, and it's the part that infrastructure spending alone can't fix.

People have deeply embedded habits around where they spend money online. Amazon has spent two decades — two decades — conditioning consumers to complete purchases without leaving its ecosystem. That's not a behaviour pattern you override by adding a checkout button to a chatbot people primarily associate with writing cover letters and settling pub arguments about whether a hot dog is a sandwich.

The contrast with China is instructive. Alibaba's Qwen and similar platforms have invested hundreds of millions in incentive programmes and merchant subsidies specifically designed to train shopping behaviour into AI interfaces. The logic being that behaviour change doesn't just happen because your product is technically capable — you have to actively, expensively, create the habit.

OpenAI's decision to retreat before making that investment suggests they either didn't have the appetite for that kind of patient capital deployment, or they did the maths and decided the near-term trade-off wasn't worth it. Either way, it's a revealing moment. Building the technology turns out to be the easy part. Getting humans to change how they've bought things for the last twenty years? That's the expensive bit.

The Amazon-Shaped Elephant in the Room

Now here's where the plot thickens in a way that should make every seller pay attention.

Amazon recently took a $15 billion stake in OpenAI, with the option to go up to $50 billion. And in a development that has received less attention than it probably deserves, Amazon — which had previously blocked AI applications like ChatGPT from accessing its product data — has signalled it may open that access up.

The commercial logic here isn't exactly subtle. If ChatGPT is going to function as a discovery engine that hands off purchases to external platforms, Amazon has every incentive to make sure its platform is the one catching those handoffs. Which means the "neutral AI commerce layer" vision that OpenAI originally pitched could quietly morph into something more like... another Amazon discovery channel.

For third-party Amazon sellers, this might actually be an unexpected tailwind. For competing platforms and DTC brands, it's the kind of thing worth watching with both eyes open.

The Money Problem (Because There's Always a Money Problem)

OpenAI's retreat from checkout can't be separated from the company's broader financial situation. They continue to spend significantly more than they earn, and only about five percent of ChatGPT users are paying subscribers. That's a lot of free usage being subsidised by a relatively thin slice of revenue.

The most likely near-term response is more advertising inside ChatGPT, though early evidence from Perplexity's experience suggests that chatbot advertising hasn't yet demonstrated the kind of returns that would substantially change the picture. Meanwhile, Anthropic is reportedly making real progress toward profitability in enterprise markets, which adds competitive pressure to an already pressurised situation.

With an IPO on the horizon, the gap between what OpenAI spends and what it earns is increasingly less "visionary investment" and more "strategic liability." The Agentic Commerce Protocol that OpenAI and Stripe are continuing to develop preserves some future optionality — but "optionality" is a word companies use when the present isn't working out as planned.

What This Actually Means If You Sell Things Online

For sellers and brand operators, the immediate practical takeaway is straightforward: ChatGPT is a discovery and referral channel, not a transactional one. Plan accordingly.

The strategic question isn't how to optimise for in-ChatGPT checkout (because that's no longer a thing). It's how to ensure that when ChatGPT surfaces your product in a recommendation, the path from that moment to an actual purchase on your own platform is as frictionless as humanly possible.

This is also a useful reminder of a distinction that keeps proving itself: product discovery and product purchase are fundamentally different behaviours, and AI assistants are reshaping the first one far faster than the second. The companies best positioned right now are the ones investing in answer engine optimisation — making sure their products are structured, described, and cited in ways that AI systems can parse and reference with confidence.

That work has a longer horizon than checkout integration, but it's also considerably more durable. Which is a polite way of saying it won't get quietly killed after five months.

The Bottom Line

The vision of AI agents completing purchases on your behalf — autonomously, across any retailer, with seamless inventory and tax reconciliation — remains plausible as a medium-term trajectory. But the infrastructure required to support it is still being built, and the companies most likely to win are the ones building for what the channel will become, rather than betting on what it was announced to be.

Or to put it less diplomatically: the future of AI commerce is real, it's just running on a slightly different timeline than the press releases suggested. Which, if you've been in ecommerce for more than fifteen minutes, should feel extremely familiar.

P.S. — The fact that OpenAI couldn't get people to buy things inside a chatbot but Amazon can get me to accidentally purchase a 48-pack of batteries at 1am with one thumb while half-asleep should tell you everything about who actually understands consumer behaviour.

P.P.S. — If you're an Amazon seller wondering how to optimise for AI discovery before everyone else figures it out, that's literally what this newsletter exists for. Stick around.

• Gemini Pro 3.1 UI gen looks scary good: one strong creative brief can now produce a cinematic, pixel-perfect React landing page with a full design system, micro-interactions, and pro-grade motion, basically turning prompts into production-ready frontend.

• WPP is merging agencies and cutting costs to become a simpler, AI-enabled business, targeting £500m annual savings by 2028 while reinvesting into AI transformation services as legacy ad models and staffing structures get squeezed.

• Amazon and Shopify now make up roughly half of US ecommerce, with Amazon dominating as the shopping destination and Shopify scaling as commerce infrastructure, leaving everyone else fighting harder for the remaining half of the market.

💰 Payback - GPT-4 Vision enabled expense tracker

🎙Outcast - AI content creation for podcasts

🗒️ PlusDocs - Create custom templates for Google Slides.

📹 Outset - An AI tool that interviews users/candidates with video.

📅 Agenda Hero Magic - Creates AI-generated calendar events

😍It was great, all I needed

🙂Good, not great

🙄It sucked