No. OpenAI expanded it on the same day it dropped Instant Checkout. ACP now carries product feeds and promotions into ChatGPT, delivered directly or through providers including Salesforce and Stripe. The specification also kept shipping releases through the period, up to the 17 April 2026 version that added cart, feed, orders, authentication and MCP support.

Why did Instant Checkout fail?

OpenAI gives one reason in its own words: the initial version did not offer the level of flexibility it wanted, so it started letting merchants use their own checkout experiences. Reporting at the time, including CNBC on 24 March 2026, described the feature as having failed to take off. Treat any more specific explanation as unsourced unless it points at a primary document.

Can shoppers still buy through ChatGPT?

Yes, with the checkout on the merchant side or inside a ChatGPT app. Walmart launched such an app the same day, with account linking, loyalty and Walmart payments. Shopify merchants are represented through Shopify Catalog without per-merchant work.

What should a merchant build now?

Feed quality. Discovery is the layer that survived, and an agent that cannot parse your catalogue will never route a buyer to any checkout, yours or anyone else’s. See how AI agents find and read your products for the specifics.

Does this make UCP the safer bet?

It makes single-surface bets the risky ones. UCP covers discovery through post-purchase across Google surfaces and has broad retail backing, but the lesson of March 2026 is not to pick a different single winner. It is that the data layer outlives the surfaces, so support the protocols your buyers’ agents actually use and invest in the catalogue underneath.

ACP Outlived Instant Checkout

Short answer: On 24 March 2026, OpenAI ended the native in-chat purchase flow that had introduced ACP six months earlier, saying the first version did not offer the level of flexibility that we aspire to provide. It did not retire the protocol. It expanded ACP to carry product discovery instead, and told merchants to keep their own checkout. That split is the most useful thing that happened to agentic commerce in 2026: it separated the durability of a specification from the durability of the surface that launched it, and it showed which of the two layers merchants should actually be engineering for.

Six months, in dates

The timeline matters here more than the commentary, because the two things that happened are easy to conflate.

Date	What happened
29 Sep 2025	Instant Checkout launches in ChatGPT for US users, powered by ACP and built with Stripe. US Etsy sellers first, over a million Shopify merchants announced as coming soon. Merchant stays merchant of record.
12 Dec 2025	ACP specification release adds fulfillment.
16 Jan 2026	ACP specification release adds capability negotiation.
30 Jan 2026	ACP specification release adds extensions, discounts and payment handlers.
24 Mar 2026	OpenAI ships richer visual product discovery, extends ACP to power it, and lets merchants keep their own checkout. Instant Checkout as a native in-chat purchase flow ends.
17 Apr 2026	ACP specification release adds cart, feed, orders, authentication and MCP support.
17 Jun 2026	Shopify opens its agentic commerce layer to all developers, removes the approval requirement, exposes a public MCP endpoint and ships a UCP Skill.

Read that column downward and the shape is obvious. The product had a six-month life. The specification never stopped shipping, including three weeks after the product was pulled.

What OpenAI actually said

It is worth quoting rather than paraphrasing, because the framing has been widely distorted into OpenAI killed ACP, which is the opposite of what the announcement says.

On the checkout: We have found that the initial version of Instant Checkout did not offer the level of flexibility that we aspire to provide, so we are allowing merchants to use their own checkout experiences while we focus our efforts on product discovery.

On the protocol, in the same post: we are expanding the Agentic Commerce Protocol (ACP) to support product discovery, and ACP becomes the connective layer between merchants and users throughout discovery.

OpenAI named the retailers already integrated into ACP for discovery: Target, Sephora, Nordstrom, Lowe’s, Best Buy, The Home Depot and Wayfair. It said Shopify merchants are covered through Shopify Catalog with no additional work, quoting Mani Fazeli, VP of Product at Shopify, on millions of merchants being open for business in ChatGPT. And it announced Walmart shipping an in-ChatGPT app with account linking, loyalty and Walmart payments.

That is not a retreat from agentic commerce. It is a company deciding it would rather own the part where buyers decide than the part where they pay.

Three things this actually proves

1. Adopting a spec and adopting a surface are different bets

Merchants who integrated ACP in late 2025 to appear in Instant Checkout lost the surface. Merchants who integrated ACP as a data contract kept everything: the same feed work now powers discovery, which is where the traffic went. The integration did not become worthless, the storefront it pointed at did. When you evaluate a protocol, separate the question will this specification still exist in two years from will this particular consumer product still exist in two years. They have very different answers and very different owners.

2. Discovery is the more durable layer

Checkout is a solved problem with incumbents, regulation and switching costs. Discovery is not solved, and it is where an agent adds value the user cannot get elsewhere. When OpenAI had to choose which half to keep, it kept the half that is hard and that nobody else owns yet. Expect that to hold: the surfaces will keep churning at the payment step and consolidating at the discovery step.

3. Governance predicts durability better than momentum

ACP has two founding maintainers and is not yet under neutral governance. That is not a criticism of the specification, which kept improving. It is an observation about who can change direction unilaterally. Compare with AP2 moving to the FIDO Alliance in April 2026, and with MCP and A2A at the Linux Foundation. A protocol whose roadmap is a product decision at one company carries a different risk than one whose roadmap is a working group.

What to build instead

The practical answer is unglamorous and it has not changed since March: make the catalogue legible to machines, then connect it to whichever surfaces your buyers actually use.

Feed quality first. Complete, current, structured product data is the only input every agent surface consumes, whatever the protocol. Our breakdown of how agents read a product page is the checklist.
Support more than one protocol. ACP for ChatGPT discovery, UCP for Google surfaces. Both are open and neither excludes the other. See the ACP vs UCP comparison.
Keep checkout where you control it. That is now OpenAI’s own recommendation, not a defensive position.
Treat trust as a separate workstream. Agent verification and payment authorization are consolidating under the card networks and FIDO, not under the commerce protocols. See how AI agents pay.

For the full record of what happened to the checkout flow itself, see ChatGPT Instant Checkout: what happened.

The protocol outlived the product: ACP after Instant Checkout