OpenAI turns ChatGPT shopping into product discovery

OpenAI’s March 24 shopping update reads like an upgrade if you only skim the product demo. ChatGPT can now show richer product cards, better side-by-side comparisons, image-led inspiration, and fresher merchant data. But the sentence that matters sits lower in the company’s own post: OpenAI says it found that the initial version of Instant Checkout did not offer the flexibility it wanted, so merchants will use their own checkout experiences while ChatGPT focuses on product discovery.

That is not a small product tweak. It is OpenAI backing away from the idea that ChatGPT itself should be the main place where a shopping conversation ends. The company is still very much in the commerce game. It is just choosing a different job. Instead of trying to replace checkout, ChatGPT shopping is being recast as a discovery and comparison layer that hands high-intent users into merchant-controlled environments.

What changed

When OpenAI launched Instant Checkout last year, the pitch was straightforward: discover products in ChatGPT, hit buy, confirm your order, and complete the purchase without leaving the chat. The launch leaned hard on the idea that agentic commerce could collapse research and transaction into one smooth flow, powered by the Agentic Commerce Protocol and Stripe. TechCrunch noted that OpenAI rolled the feature out in September and framed it as a way to make ChatGPT more like an e-commerce hub.

The new product discovery update tells a different story. OpenAI is now talking about product discovery, not native checkout, as the place where ChatGPT adds the most value. The company says users can browse visually, compare options side by side, upload inspiration images, and see more up-to-date pricing, reviews, and product details. Under the hood, ACP is being expanded to support discovery feeds and promotions so merchants can supply better catalog data without handing the entire conversion moment to OpenAI.

That shift matters because it changes what success looks like. The old version tried to turn ChatGPT into a place where transactions close. The new version is happy to be the place where decisions get narrower, faster, and more informed.

The handoff paths also matter. OpenAI says Shopify merchants already flow into ChatGPT through Shopify Catalog, with purchases completing on the merchant’s own online store through an in-app browser. On the same day, OpenAI introduced Walmart’s new in-ChatGPT app, which keeps the user inside a tailored Walmart environment with account linking, loyalty, and Walmart payments. Those are not the same flow, and the distinction is important: the common thread is merchant control, not one universal checkout box owned by ChatGPT.

Why OpenAI backed off native checkout

OpenAI’s official explanation is that Instant Checkout did not offer the flexibility it wanted. In retail language, that word does a lot of work. Flexibility means loyalty programs, account identity, payment preferences, fraud controls, local inventory, fulfillment choices, returns, support, merchandising, and all the other details that make commerce feel less like a demo and more like a real business.

Stripe’s own ACP write-up quietly explains why that is hard. It says agentic commerce has to deal with trust, fragmentation, and support for more complex flows than a single-item purchase. That lines up with what OpenAI appears to have learned in production. Native chatbot checkout is elegant right up until merchants ask for the parts of commerce they actually care about.

The strongest outside signal came from Walmart. Search Engine Land reported that Walmart EVP Daniel Danker said purchases completed directly inside ChatGPT converted at one-third the rate of click-out traffic to Walmart’s own site. That figure is second-hand reporting and should be treated that way. Still, if it is even directionally right, it explains the pivot better than any product copy could. Retailers can tolerate a lot of experimentation. They get much less philosophical when the supposedly frictionless path converts worse.

This is why the March 24 update feels less like a step forward for native checkout and more like a retreat into the part of the shopping journey where AI is already useful. ChatGPT is good at helping people describe what they want, refine tradeoffs, and compare options without opening twelve tabs. That is valuable. It is also a very different claim from saying the bot should own the final checkout surface.

Why it matters

The bigger implication is that OpenAI may have found the more durable business position in AI commerce. Discovery is a powerful layer even if checkout happens somewhere else. If ChatGPT becomes the place where users decide which laptop, rug, serum, or shirt actually fits their needs, OpenAI still sits close to the most valuable part of the journey: the moment before intent hardens into a purchase.

That pattern fits a broader OpenAI habit. We have already seen versions of it in OpenAI’s agents platform shift and in the company’s attempt to shape developer workflow surfaces with Astral. Own the interface. Own the protocol. Let partners carry more of the operational mess. Commerce, unsurprisingly, is trying to force the same lesson.

It also lines up with where the market is heading outside OpenAI. In January, Google launched its own Universal Commerce Protocol and retailer-facing agent tools. The common idea is not that storefronts disappear. It is that AI becomes the layer that mediates discovery while merchants fight to keep the branded environment, customer relationship, and payment stack.

That is a much more believable future than “everything becomes checkout inside a chatbot.” The chatbot can influence demand without becoming the merchant of record in spirit, even if protocols and payments make the handoff smoother underneath.

What to watch next

There are still moving parts here. OpenAI says the updated shopping features are rolling out to Free, Go, Plus, and Pro users this week, while Walmart’s in-ChatGPT app is live on the web first with iOS and Android to follow. Merchant participation will keep shifting. ACP discovery integrations will change. And OpenAI’s shopping flows will not look identical across Shopify merchants, Walmart, and any future ChatGPT apps.

The next real test is simple: does ChatGPT become a durable product discovery surface that merchants actually want to feed, or does this settle into another flashy AI layer that sends inconsistent traffic and weak buying signals? If Walmart’s reported conversion gap turns out to be representative, OpenAI’s retreat from native checkout will look less like a stumble and more like the company learning the right lesson early.

The first version of ChatGPT shopping tried to be the place where the sale closed. The new version looks more plausible because it stops trying to do everything at once.