OpenAI and ACP: what happened to in-chat checkout, and what to do now

If you tried in-chat checkout in ChatGPT last year and went looking for it again recently, you may have noticed it’s harder to find. That’s not your imagination. OpenAI scaled back the Instant Checkout pilot around March 2026, the same month it opened a different door: product discovery through ACP. Two changes, same month, easy to conflate. Here’s what actually happened, what’s open right now, and what a store owner should do about it today versus what to leave alone.

What ACP is

ACP stands for Agentic Commerce Protocol. OpenAI built it together with Stripe and open-sourced it under Apache-2.0. It’s the specification that lets an AI agent, in this case ChatGPT, read a merchant’s product catalog and, where enabled, complete a purchase on a shopper’s behalf. The spec has kept moving since launch; a stable dated version exists as of this writing.

The timeline, plainly

In 2025, OpenAI launched Instant Checkout inside ChatGPT. A shopper could find a product in conversation and buy it without leaving the chat window. It started with a small set of merchants, not a general rollout.

Around March 2026, two things happened close together. First, the in-chat Instant Checkout pilot was scaled back. Second, and separately, OpenAI extended ACP to cover product discovery, not just checkout. That second move is the one that matters most for most merchants right now, because it’s the door that’s actually open.

Checkout got smaller. Discovery got bigger. Those are two different stories, and most of the confusion out there comes from treating them as one.

What’s open today: discovery via feed

The actionable path right now is getting your catalog into ChatGPT’s product discovery results, not into its checkout flow. Here’s how that works.

A merchant applies at chatgpt.com/merchants with business details, a website, and catalog size. OpenAI verifies the business. Once approved, the merchant pushes a product feed in one of the common formats, and can refresh it frequently to keep prices and availability current.

Two limits matter enough that they’ll quietly gate a lot of stores out if you don’t check first:

• US merchants only, today. If your business isn’t US-based, there’s no application to make yet.
• Platform matters. Shopify and Etsy catalogs are auto-eligible, no separate application step on the catalog side. WooCommerce is not an integrated platform. If you’re running WooCommerce, or you’re outside the US, you’re effectively gated out of this program right now, regardless of how clean your catalog is.

That’s a real limit, not a formality to route around. Don’t build custom feed infrastructure for a program that won’t accept your platform yet.

What’s still gated: in-chat checkout

Instant Checkout, the part that actually completes a transaction inside ChatGPT, is still rolling and still gated to a limited set of merchants. Where it is live, OpenAI takes roughly 4% of the completed purchase. OpenAI has signaled a fully self-serve merchant portal for later in 2026, but that’s roadmap, not something shipped or datable yet. Treat it as a direction, not a plan you can build a Q3 launch around.

This is the honest split: discovery is open and actionable today. Full in-chat checkout is not something most stores can just turn on, and there’s no application queue to jump for it. Anyone telling you otherwise is either ahead of the public rollout or mistaken.

What to actually do now

If you’re a US-based Shopify or Etsy merchant, the move is straightforward: apply at chatgpt.com/merchants, get verified, and push a clean product feed. Your catalog is already auto-eligible on the platform side, so the application is the only gate between you and showing up in ChatGPT’s product discovery.

Before you apply, or while the application is in review, the highest-leverage thing you can do is make sure the feed itself is worth showing. A feed pulling from thin titles, empty descriptions, and missing attributes will get into the discovery pool and still lose to competitors with better data, because the agent surfacing results still has to decide what to recommend. This is exactly the gap Pollen closes: it enriches your catalog with structured facts and generates a compliant ACP product feed from data that’s actually usable, not just present.

If you’re on WooCommerce or outside the US, there’s nothing to apply for yet. Don’t spend engineering time building ACP feed infrastructure for a program that isn’t accepting your platform. Spend that time instead on the same thing every merchant should be doing regardless of protocol: clean product data, real attributes, descriptions an agent can actually parse. When the gate opens to your platform or region, and it likely will given how fast this space is moving, you want to be ready to flip a switch, not start from a blank catalog.

What to wait on

Don’t chase in-chat checkout access. There’s no visible application queue for it the way there is for the discovery feed, and it remains gated and rolling by OpenAI’s own choice, not a technical limitation you can work around. The self-serve portal that was signaled for later in 2026 is worth watching, but it hasn’t shipped, and building your roadmap around an undated promise is a good way to burn a quarter waiting.

The practical read: get discoverable now, because that path is open and it rewards good data immediately. Get checkout-ready in the sense of clean data and a working feed, but don’t expect to flip on in-chat transactions yourself anytime soon. When OpenAI widens that gate, the merchants who show up first will be the ones who already did the unglamorous work of making their catalog legible to an agent.