1. Headline & intro
OpenAI has quietly admitted something the hype never did: turning ChatGPT into an Amazon-style shopping mall is far harder than it looked on keynote slides. By scaling back its Instant Checkout feature and refocusing on product discovery, the company is stepping away from the dream of frictionless one‑click shopping inside a chat window. That matters far beyond OpenAI. It’s an early reality check on a popular narrative: that AI agents will instantly eat e‑commerce. In this piece, we’ll unpack why this pivot happened, who it helps or hurts, and what it tells us about the next phase of AI‑driven shopping.
2. The news in brief
According to TechCrunch, OpenAI is de‑emphasising Instant Checkout, a feature launched last year that let users complete purchases directly inside ChatGPT. The tool allowed shoppers to chat about what they wanted, add products to a cart within the interface and pay, while fulfilment was handled by external merchants.
OpenAI now says the first version of Instant Checkout was too rigid and that merchants will increasingly use their own checkout flows. The company will still support checkout via apps inside ChatGPT, but it is no longer a core product priority.
Instead, OpenAI is leaning into discovery: richer product comparison views, more detailed information, and a centralised shopping research hub, all built on its Agentic Commerce Protocol (ACP), developed with Stripe. Previous reporting from The Information and CNBC, cited by TechCrunch, indicated usage of the direct checkout flow was low and e‑commerce sites saw limited revenue from ChatGPT‑referred traffic.
3. Why this matters
This is not just a product tweak; it’s a signal about what generative AI is actually good at in commerce right now.
OpenAI and many industry commentators assumed that if AI can recommend products, it should also own the transaction. That’s where margins and data live. But behaviour did not follow the slideware. Users chatted; they did not, at scale, hand over their card details and complete a purchase in a context that still feels like “search plus conversation.”
The winners in this pivot are, paradoxically, the incumbents AI was supposed to disrupt. Amazon, large retailers and established web shops keep control of checkout, returns, loyalty programmes and post‑purchase relationships. They get traffic and intent data without ceding the most valuable part of the funnel to OpenAI.
OpenAI, meanwhile, retreats to a more natural role for a horizontal AI assistant: the top and middle of the funnel. Helping people understand categories, compare specs and read condensed reviews is something models are already good at. Turning that into a trusted payments layer is another problem entirely, involving risk, regulation, fraud, refunds and consumer trust.
Small and mid‑size merchants may gain here. If ChatGPT becomes a serious discovery layer, ACP could give them a more level playing field against dominant marketplaces, provided integration remains open and affordable. But the strategic prize for OpenAI—being the new “Buy” button of the internet—moves further away, at least for now.
4. The bigger picture
OpenAI’s pivot sits inside a larger correction in AI commerce expectations.
We’ve seen a wave of announcements: Amazon’s AI assistant Rufus baked into its app, Google’s Shopping Graph and AI Overviews pushing product answers directly into search results, Microsoft’s Copilot suggesting purchases in Edge and Bing, Shopify experimenting with AI‑driven storefronts and support, and Klarna rolling out an AI shopping assistant that went as far as claiming it could replace customer service roles.
The promise everywhere has been similar: conversational agents will guide you from intention to transaction in one smooth flow. Yet, outside of China’s WeChat ecosystem—where mini‑programs, payments and messaging are tightly integrated—Western users haven’t fully embraced chat as a primary checkout surface.
Historically, this isn’t new. Facebook’s M assistant, early Messenger bots, and even voice shopping via Alexa all ran into the same wall: discovery and inspiration were interesting, but most users reverted to familiar web or app flows to actually pay. Habits, trust and the need to visually verify products matter more than executives like to admit.
OpenAI’s new focus on discovery acknowledges this. ChatGPT is better positioned as a meta‑layer above commerce than as a store itself. If ACP matures, OpenAI could become the comparison engine and intent router of the AI era—more akin to a super‑charged affiliate network or price comparison site than a direct Amazon rival.
That would still be enormously powerful. Whoever owns the user’s first query—and the ranking and filtering logic behind answers—owns outsize influence over which merchants win or lose. The pivot does not mean OpenAI is out of commerce; it means it is choosing to shape demand rather than process the payment, at least in this phase.
5. The European / regional angle
From a European perspective, OpenAI’s step back from embedded checkout may be a blessing in disguise.
The EU’s Digital Services Act (DSA) and Digital Markets Act (DMA) both push in one direction: limit the power of gatekeepers that sit between users and businesses, and clamp down on opaque recommender systems and dark patterns in e‑commerce. If ChatGPT had rapidly evolved into a closed shopping mall with its own payment layer, regulators in Brussels would almost certainly have taken an interest in ranking transparency, self‑preferencing and consumer protection.
By keeping checkout primarily on merchants’ own sites, OpenAI reduces its immediate exposure to those headaches. At the same time, the EU AI Act will still apply to any profiling and recommendation logic behind product suggestions, particularly when they rely on sensitive or behavioural data from users.
For European merchants—think Zalando, Otto, Allegro, Bol.com, or specialised national e‑shops—ACP and similar protocols could be an opportunity to get in front of global audiences without being swallowed by a US marketplace. But they will expect clarity: on data sharing, on ranking criteria, and on fees.
European consumers, often more privacy‑conscious than their US counterparts, are another wildcard. Will they be comfortable with an AI assistant that knows their chat history also shaping what they buy? Under GDPR, consent, data minimisation and purpose limitation are non‑negotiable. Any attempt to heavily personalise product suggestions must be explained and controllable.
In short, Europe is both a fertile testbed for AI‑driven product discovery—and the region most likely to push back if it turns into a dark, opaque gatekeeper layer.
6. Looking ahead
The most likely next step is that OpenAI doubles down on being the “universal product researcher” rather than the “universal checkout.” Expect several developments over the next 12–24 months:
- Richer comparison experiences. Side‑by‑side visuals, feature matrices, warranty and sustainability data, and real‑time price tracking across merchants. Think a conversational Idealo or Google Shopping.
- Clearer monetisation. Referral fees, sponsorships and promoted placements are almost inevitable. The open question is how transparently they will be labelled and how strongly they will influence rankings.
- Deeper merchant tools. If ACP succeeds, OpenAI will court retailers with dashboards, analytics and maybe even AI‑generated product content, in exchange for structured feeds and attribution data.
- Integration into devices. As AI assistants move into operating systems, cars and wearables, being the layer that answers “What should I buy?” becomes even more strategic. Commerce will follow attention.
There are also significant risks. If users perceive recommendations as biased or as thinly veiled ads, trust in ChatGPT’s answers could erode quickly. Regulators may decide that any dominant AI assistant is, by definition, a gatekeeper subject to DMA‑style obligations—forcing interoperability and strict transparency.
On the flip side, merchants that invest early in structuring their data and understanding how ACP‑like standards work could find themselves disproportionately visible in AI‑driven results. This is the new SEO: not just optimising for web search crawlers, but for AI agents that summarise, compare and choose.
7. The bottom line
OpenAI’s retreat from one‑click shopping inside ChatGPT is not a failure of AI, but a correction of misplaced ambition. For now, AI assistants are better at guiding choices than capturing payments. The real battle will be over who controls the discovery layer and on what terms. As ChatGPT, Amazon, Google and others race to own that layer, the question for users and regulators alike is simple: will the future of shopping feel like help, or like a new kind of walled garden?
