OpenAI is pulling native Instant Checkout out of ChatGPT and moving it to its Apps platform, after only a few dozen merchants went live on the feature it launched in September 2025. The durable lesson for brands is not that AI shopping failed. It is that product discovery inside AI keeps growing while the checkout rail consolidates. The practical response: optimise to be discovered and recommended inside AI answers, and convert that intent on your own site. This is part 13 of the agentic-commerce series.
What actually changed with ChatGPT's Instant Checkout?
OpenAI launched Instant Checkout. A native, in-chat purchase flow built on the Agentic Commerce Protocol. In September 2025. By early 2026 it had reversed course. The Information reported the scale-back in March 2026, and CNBC reported on 24 March 2026 that OpenAI was revamping the ChatGPT shopping experience after struggling with the offering.
An OpenAI spokesperson framed the shift plainly, per Search Engine Land and Modern Retail:
- "Instant Checkout is moving to Apps, where purchases can happen more seamlessly."
- In practice, checkout moves from a single native in-chat flow to individual retailer apps: names like Instacart, Target, Expedia, and Booking.com, that plug into ChatGPT.
So the change is architectural, not a retreat from commerce: OpenAI is stepping back from being the checkout and toward routing to where checkout already works. Discovery and recommendation inside ChatGPT continue; the native payment surface is what consolidated.
Why did in-chat checkout stall?
Two reasons, both company-reported and best read as directional early-to-mid-2026 figures.
| Signal | What was reported | Source |
|---|---|---|
| Merchant adoption | ~30 Shopify merchants live by Feb 2026 (Rye counted ~12) | Forrester (Emily Pfeiffer); Rye |
| Conversion | In-chat checkout converted ~3× worse than click-through to walmart.com | Walmart |
| New-customer rate | ChatGPT drove ~2× the new-customer rate Walmart sees from search | Walmart |
The pattern reconciles cleanly: AI drove valuable demand, but the native in-chat checkout under-converted it. Shopify president Harley Finkelstein, speaking at an investor conference, located the bottleneck on the AI-firm side rather than with merchants. The plumbing to complete a purchase in-chat, not merchant willingness. Attribute the specifics to those sources and hedge accordingly; the robust finding is the shape of the problem, not the exact multiples.
The demand AI created was real; the in-chat checkout that was supposed to capture it was not ready. Discovery outran conversion, so the smart move is to win the discovery and take the conversion home.
Does this mean AI shopping is fading?
No, and conflating the two surfaces is the mistake to avoid. As covered in what agentic commerce is, there are two separate things happening: discovery (an AI engine surfacing and recommending your product) and checkout (the payment rail that completes the sale). Only the second one stalled.
- Discovery is still growing. Buyers increasingly start product research inside ChatGPT, Gemini, and Perplexity. That is exactly why where ChatGPT product recommendations come from is a strategic question.
- Checkout is consolidating toward big integrated retailers' apps and first-party sites, where conversion is higher and the experience is controlled.
This is consistent with the rest of the series: a shopper agent is a channel, not a visibility lever, and payment rails are settlement, not selection. Whether an engine recommends you is decided upstream of any checkout button.
What does "discover in AI, buy on your site" mean for brands?
It means splitting your effort to match where each stage actually happens:
- Win discovery inside AI. Be the product an engine surfaces and recommends. That depends on the factors AI uses to recommend products. Accurate, complete catalog data, structured data, entity strength, and third-party corroboration, not on any checkout integration.
- Prepare your catalog for agents. Machine-readable product data (prepare your catalog for AI agents) keeps you eligible as engines route more buyers to more surfaces.
- Convert on ground you control. Send that AI-created intent to your own site or first-party app, where conversion is stronger and you own the relationship. Don't stake your revenue on any single engine's native checkout being ready.
- Stay rail-agnostic. Support the emerging agentic checkout standards as they mature, but treat them as distribution plumbing, not the thing that decides whether you are recommended.
The through-line: answer-engine optimisation determines whether you are discovered; your own storefront determines whether you convert. OpenAI's pullback just made the division of labour explicit.
Discovery inside AI is now a measurable surface of its own. One that keeps growing even as the checkout layer reshuffles. Tracking whether AI engines actually surface and recommend your products, across ChatGPT, Gemini, Claude, and Perplexity, sampled over time, is exactly what Buffy Intel does.