When OpenAI let people buy products without leaving ChatGPT, the retail press treated it as the moment the storefront died. The demos looked like the future in a way that was easy to screenshot. A shopper asks for a product, the assistant shows it with a buy button, a few taps later the order is placed, and the merchant's website never loads. It was clean, it was legible, and it convinced a lot of people that the whole funnel was about to collapse into a chat window. Then reality arrived, and the more useful story turned out to be the opposite of the one everyone rushed to tell.

The rollout did not scale the way the announcement implied. Instant Checkout launched broadly for United States shoppers in February 2026, powered by the Agentic Commerce Protocol that OpenAI built with Stripe, and the plan on paper reached over a million Shopify merchants. In practice only a tiny handful went live. Shopify's own leadership acknowledged that the bottleneck was not merchant willingness but the onboarding process on the AI side, which simply did not scale to millions of stores. By early spring OpenAI was visibly recalibrating, shifting away from buying directly inside chat toward a model that leaned on retailer apps and, more importantly, on product discovery as the durable value.

The number that should stop any store owner mid-sentence came out of a large retailer's test. Walmart tried the in-chat checkout and found it converted roughly three times worse than simply sending shoppers back to Walmart's own site. A senior Walmart product leader described the experience as unsatisfying, and OpenAI's own follow-up conceded that the first version lacked the flexibility it wanted, then pivoted toward letting merchants use their own checkout while the assistant focused on helping people find things. This was not a small stumble. It was a signal that the specific act of completing a purchase inside a chat window was solving a problem most shoppers did not actually have.

It helps to sit with why that happened, because the reason is not a temporary bug. Shoppers do not arrive at a purchase as blank slates. They already have carts, saved payment methods, loyalty accounts, delivery expectations, substitution preferences, and return relationships with the merchants they use. Buying a single item through a chat interface does not replace any of that. It adds friction to a checkout that was already working, while stripping away the account context that makes repeat buying smooth. The chat window is a superb place to decide what to buy. It is a mediocre place to complete the mechanics of buying, at least for now, because the mechanics were never the painful part.

This is where the reframe becomes genuinely good news for stores, and especially for smaller ones. The value of the assistant was never mainly in owning the final tap. It was in owning the moment of decision, the point where a shortlist forms and one product wins. Discovery is the part that moved decisively and durably. When someone asks an assistant for the best option in a category, the assistant reads structured product data, evaluates it, and recommends specific items. That recommendation is worth enormous amounts of high-intent demand, and it happens whether or not anyone ever completes a purchase inside the chat. The checkout hype deflated. The discovery shift did not, and discovery is the layer where your catalog either shows up or does not.

Consider what this does to a merchant's priorities. If the prize were in-chat checkout, the work would be a race to integrate payment endpoints and hope shoppers actually used them, a race that early evidence suggests would end in disappointing conversion. Because the prize is discovery, the work is different and far more valuable. It is making your products legible and recommendable to the assistants people are already asking, so that when the shortlist forms, you are on it. You do not need the shopper to check out inside ChatGPT. You need them to be shown your product, decide it is right, and arrive at your own well-tuned checkout ready to buy, where your own account context, loyalty, and conversion optimization take over. That is a better outcome than an awkward chat purchase, and it plays to strengths a good store already has.

There is a competitive angle here that the checkout obsession obscured. Because discovery runs on structured data rather than on ad spend, and because OpenAI has stated that paying its transaction fee does not influence product ranking, the levers that decide visibility are your product data, your reviews, and your structured content, not your marketing budget. That is a rare situation in modern retail. It means a well-organized small brand with genuinely clean, complete catalog data can appear in the same answer as a giant, evaluated on the quality of its information rather than the size of its media plan. The retailers who understood the checkout story as the whole story missed that the real opening was upstream, in the part of the funnel where being legible beats being loud.

Skepticism is warranted about anyone still selling in-chat checkout as the imminent center of commerce, and that skepticism is precisely what points you to the right work. The industry did not conclude that agentic commerce was overblown. It concluded that the first attempt to collapse the entire funnel into a single chat purchase was premature, and it redirected toward the part that actually delivers value today. Discovery. Being present, being understood, being recommended. The checkout mechanics will keep maturing, and some categories and some large integrated retailers will make in-chat or app-based purchase work well over time. But a store that waits for checkout to be solved before caring about assistant visibility has the sequence backwards, because visibility is the thing that pays off right now and checkout is the thing still finding its footing.

The practical consequence is liberating rather than daunting. You do not have to bet your quarter on a checkout integration whose conversion is unproven. You have to do the more grounded work of ensuring an assistant can see what you sell, understand it precisely, and have a reason to prefer it. That means complete attributes, accurate stock and pricing, descriptions rich in the specific facts a query might hinge on, and consistency between what your site says and what your feed says, so an agent checking one against the other trusts you. This is not speculative future-proofing. It is the work that determines whether you appear in answers being generated today, for shoppers who will then buy from you through whatever checkout they prefer.

It is also worth naming the trap that the checkout narrative set, because avoiding it is half the advantage. The trap was to treat agentic commerce as a payments story, to wait for the payment rails and the in-chat purchase flows to be perfect, and to do nothing about visibility in the meantime. Stores that fell into that trap are invisible in the answers happening right now, and they will stay invisible until they do the data work they postponed. Stores that saw through it are quietly making their catalogs legible while the payment layer sorts itself out, accumulating a discovery position that will only become more valuable as more shoppers start their journeys by asking rather than searching.

So the stall of Buy in ChatGPT is not a reason to relax. It is a clarifying event that tells you exactly where to point your effort. The assistants are not going to stop recommending products. They are going to keep getting better at it, more people are going to keep starting there, and the recommendation is going to keep being decided by whose data is cleanest and richest. The checkout may or may not eventually live in the chat. The decision already does. Win the decision by being legible and recommendable, send the ready-to-buy shopper to your own checkout, and let the payment protocols mature on their own schedule. The brands that read the stall correctly are not the ones mourning a checkout that underdelivered. They are the ones seizing the discovery it accidentally proved was the real prize.

It is fair to expect that in-chat and app-based purchasing will keep improving, particularly for large integrated retailers who can justify building dedicated experiences, and some categories will make it work well over time. That does not change the near-term instruction for most stores, and it may even reinforce it. The retailers who eventually win at agent-driven checkout will be the ones who were already winning at agent-driven discovery, because you cannot complete a purchase for a product the assistant never surfaced. Discovery is upstream of checkout no matter how good checkout becomes, which means the visibility work pays off first and keeps paying off as the purchase mechanics mature. A store that builds its discovery position now is not betting against in-chat checkout. It is building the foundation that any future checkout, wherever it happens, will stand on. The sequence is discovery first and everything else after, and the stall of the checkout hype simply made that sequence impossible to ignore.