ChatGPT Apps Checkout: How Buying in ChatGPT Actually Works Now

Short answer: Buying inside ChatGPT no longer happens through a Buy button in the answer. In early March 2026 OpenAI deprioritized Instant Checkout and moved purchasing into merchant-branded apps that run inside ChatGPT, where the shopper logs into the merchant, the cart syncs, and checkout completes on the merchant's own stack. Target, Instacart, DoorDash and The Knot launched apps of this kind, and Walmart embedded its own assistant, Sparky. The Agentic Commerce Protocol underneath was not retired. What changed is where the checkout screen lives and who controls it.

What changed, precisely

Instant Checkout launched on September 29, 2025. A shopper asked ChatGPT for a product, saw results, tapped Buy, and completed the purchase on a checkout surface OpenAI rendered inside the conversation. The merchant stayed seller of record and charged the card through their own processor, but the buying experience belonged to OpenAI.

In early March 2026 OpenAI told reporters it was "evolving how we approach commerce in ChatGPT to better meet merchants and users where they are," and that it was "prioritizing making ChatGPT search and product discovery great, with ACP serving as the infrastructure that connects users to merchants across the full shopping journey. Instant Checkout is moving to Apps, where purchases can happen more seamlessly." On March 24, 2026 OpenAI followed with a rebuilt shopping experience weighted toward finding and comparing products rather than transacting on the spot. We keep the full timeline and the reasons behind it on our ChatGPT Instant Checkout page.

How ChatGPT apps checkout works

The mechanics are closer to a normal ecommerce session than to the old in-answer flow, which is exactly the point.

1. The shopper asks ChatGPT about a product or a task, and ChatGPT surfaces the relevant merchant app alongside its own results.
2. The shopper opens the app inside the conversation and signs into their existing account with that merchant.
3. The cart syncs between the app and the merchant's own site, so an item added in ChatGPT is the same item in the merchant's basket.
4. Checkout runs on the merchant's stack, with the merchant's payment methods, saved cards, loyalty program, subscriptions, delivery windows and returns policy all intact.
5. Order status and support flow back through the merchant's normal channels, not through OpenAI.

Walmart's implementation is the clearest illustration. Rather than exposing a catalog to a generic checkout, Walmart brought Sparky, its own shopping assistant, into ChatGPT. The shopper logs into Walmart, carts stay in sync across both surfaces, and the purchase completes inside Walmart's system exactly as it would on walmart.com.

Why merchants prefer it

Two numbers explain the shift better than any strategy narrative. Daniel Danker, Walmart's EVP of product and design, said purchases completed inside ChatGPT converted at roughly one third the rate of clicking through to walmart.com, and described the in-chat experience as "unsatisfying". And Forrester principal analyst Emily Pfeiffer counted only about 30 Shopify merchants actually participating as of February 2026, out of a base in the millions.

The reason sits in what a generic checkout cannot carry. Shopify president Harley Finkelstein made the point that transaction integrity is not only about payments, it is "checkout itself, subscriptions, inventory, shipping taxes, all the different merchandising options." A retailer's checkout is where loyalty points apply, where a subscription discount resolves, where a delivery window is chosen, and where a bundle is upsold. Strip that out and conversion falls, which is precisely what happened.

What a merchant actually builds

Less new work than the shift suggests, because the reusable asset was never the checkout. It is the product data. Every standard in this space consumes the same input: a regularly refreshed feed, as CSV or JSON, carrying accurate identifiers, descriptions, pricing, inventory, media and fulfillment options. That feed powers the Agentic Commerce Protocol, it powers Google's Universal Commerce Protocol, and it will power whatever arrives next. Merchants who already maintain a clean feed can light up a new agentic channel in days. Merchants who do not will spend a quarter on data cleanup no matter which protocol wins.

Beyond the feed, the practical list is short. Render product facts server side so agents can read them without executing your JavaScript. Decide whether an app is worth the build for your category, which for most non-enterprise merchants it currently is not. And fix attribution before the volume arrives, because orders from agent surfaces land with no session history and quietly break every acquisition model you run. We walk through the wider readiness checklist in agentic commerce for merchants.

What this means if your company is the buyer

Here is the part the coverage missed almost entirely. The retreat was reported as a merchant story, and on the buy side it made the visibility problem worse rather than better.

When checkout lived in ChatGPT, a purchase at least happened on a surface with a defined protocol and a scoped, single-use payment credential carrying a maximum amount and an expiry. Now the purchase completes inside a merchant account, using whatever card is saved to that account, in a session your systems never observe. Your shopping agents and your employees are still buying through a chat window. The charge simply arrives later, on a statement, under a merchant name nobody recognizes, with no record of what was asked for or who approved it.

None of the emerging standards close that gap, because none of them is scoped to. Each one authorizes a single transaction and none keeps a running total across agents, merchants or rails. That is true of ACP, of the mandates in Google's AP2, and of per-call pricing in x402. Twenty individually authorized purchases still add up to a number nobody approved.

The controls that survive a protocol change

The useful lesson from the last twelve months is to stop building controls that assume a specific checkout surface. In September 2025 the answer was Instant Checkout. By March 2026 it was apps. In a year it may be something else, and any control wired to one of those disappears with it.

What does not change is the credential. Issue a scoped virtual card per agent or per vendor rather than letting everything share one number, and the merchant, the amount and the frequency are constrained no matter which surface the purchase happens on. Fund each agent with a hard wallet limit so it is structurally incapable of exceeding its budget. Route anything above your threshold to human approval before money moves. And write every decision to a single audit trail that spans card spend, stablecoin settlement and metered API calls together, so the record does not fragment by rail.

The reconciliation side deserves attention early. Purchases made through agent surfaces arrive as vendor invoices and emailed receipts your accounting system has never seen, often as PDFs with no purchase order behind them. Pushing those through automated document data extraction before they reach the ledger is what keeps month end from absorbing the whole problem, and it pairs with the process we describe in reconciling agent spend at month end.

Frequently asked questions

Can you still buy things in ChatGPT?

Yes. The in-answer Buy button was deprioritized in March 2026, but purchasing continues inside merchant apps that run in ChatGPT. You sign into the merchant, your cart syncs, and checkout completes on the merchant's own systems rather than on a surface OpenAI renders.

Which merchants have apps in ChatGPT?

Target, Instacart, DoorDash and The Knot are among the retailers that launched apps, and Walmart embedded its own Sparky assistant. Expect the list to skew toward large retailers with the engineering capacity to build and maintain an app, rather than the long tail of smaller merchants the original Instant Checkout rollout targeted.

Is the Agentic Commerce Protocol still used?

Yes. ACP is Apache 2.0, co-developed by OpenAI and Stripe, and OpenAI named it directly as the infrastructure it continues to build on. As of August 2026 OpenAI's commerce documentation describes it as current with no deprecation notice. Instant Checkout was a product built on ACP, not ACP itself.

Should merchants build a ChatGPT app?

For most, not yet. The build is meaningful, the traffic is unproven outside a handful of large retailers, and the reusable value sits in the product feed rather than the app. Get the feed clean and server rendered first, since that is the work every future channel depends on, then evaluate an app once your category shows real assistant-driven demand.

Does app-based checkout give a finance team more visibility?

No, less. The purchase now completes inside a merchant account using a stored card, in a session outside your systems entirely. Visibility has to come from the credential layer instead: a card scoped to one agent and one merchant, a hard per-agent budget, and an audit trail that records what was bought and who allowed it.

The short version

Buying in ChatGPT did not go away, it moved into the merchant's own environment where conversion is better and control is theirs. Merchants should invest in product data rather than in any one checkout surface. Buyers should assume the surface will keep changing and put their controls where credentials are issued, which is the one layer that outlives whichever protocol is winning this quarter. Our ChatGPT Instant Checkout explainer tracks the full timeline, and how agentic checkout works covers the mechanics across both protocols.