If you sell anything online, there's a good chance a customer has already tried to buy it inside a ChatGPT conversation without ever landing on your site. OpenAI's Instant Checkout lets people complete a purchase directly in the chat window, and it's built on an open standard called the Agentic Commerce Protocol that Stripe co-developed with OpenAI. This isn't a future feature to plan around eventually. It's live, it's expanding fast, and the merchants who understand how it actually works are the ones getting picked when it matters.
I want to walk through the mechanics first, because most of what's written about Instant Checkout treats it like a marketing bullet point instead of the piece of infrastructure it actually is. Then we'll get into the part that matters to your business: what you need in place to even qualify, and what seems to tip the scales when ChatGPT has to pick one merchant over another.
What Instant Checkout actually is
Instant Checkout is the feature that lets a ChatGPT user buy a product without leaving the chat. Someone asks for a gift idea, ChatGPT surfaces a few options, and instead of clicking out to a browser tab, the user can pay right there. Etsy and Shopify were the first launch partners in September 2025, and the merchant list has grown since, adding Walmart, Sam's Club, Salesforce-connected retailers, and grocery delivery through Instacart and DoorDash.
Two things are easy to conflate here, so let me separate them. ChatGPT Shopping is the discovery layer: the product cards, comparisons, and recommendations ChatGPT shows when someone asks a shopping question. Instant Checkout is the transaction layer sitting underneath a subset of those results. A product can show up in ChatGPT's shopping answers without being purchasable in-chat. Only merchants who've built the checkout integration get the buy button.
The Agentic Commerce Protocol, in plain terms
The whole thing runs on ACP, an open-source specification released under Apache 2.0 in September 2025 and maintained jointly by OpenAI and Stripe. The point of making it an open standard, rather than a proprietary OpenAI API, is that any compliant AI agent can eventually transact with any compliant merchant without a custom integration for each pairing. Whether that promise holds as more platforms adopt it is still an open question, but the architecture is designed for it.
ACP breaks the purchase flow into a handful of building blocks. Here's what each one does:
- Product feed: merchants publish a structured feed, in CSV, JSON, TSV or XML, listing SKUs, prices, inventory, images, and fulfillment options. Stripe's documentation specifies this feed should refresh frequently, since ChatGPT is quoting live price and stock data, not a cached snapshot from last week.
- Agentic checkout: this is the transaction API itself. ChatGPT calls a `POST /checkout_sessions` endpoint on the merchant's server with the cart contents and buyer context, then updates that session as the shopper adjusts quantities, shipping, or applies a discount.
- Delegated payment: a payment token gets passed securely from the buyer to ChatGPT to the merchant, so OpenAI never stores raw card data and the merchant still processes the charge through its existing payment stack, most commonly Stripe.
- Delegated authentication: OAuth 2.0 handles the handoff that lets an agent act on a buyer's behalf without the buyer re-entering credentials at every step.
- Order webhooks: merchants publish order status updates back to ChatGPT so the conversation reflects fulfillment-grade truth, not just "order placed and forgotten."
Who actually processes the money
This is the part brands get wrong most often. OpenAI is not the merchant of record and it does not touch your payment processing. You keep charging the customer, remitting sales tax, handling fulfillment, and owning returns exactly as you do today. ChatGPT is the discovery and negotiation layer; your existing commerce stack does the actual selling. Stripe built ACP so payments can route through Stripe's existing infrastructure, but merchants on other processors can implement the protocol too, since ACP is a spec, not a Stripe-exclusive product.
Practically, that means the checklist to get live looks a lot like standing up any API integration: a developer needs to build or expose the five ACP endpoints, connect them to your product catalog and order management system, and pass OpenAI's verification before the buy button goes live in chat. OpenAI's developer documentation has the current spec version and onboarding requirements, and it's worth reading in full rather than relying on a summary, since the requirements have shifted more than once since the September 2025 launch.
How ChatGPT picks which merchant fulfills an order
OpenAI has been explicit that these placements are not paid media. The company has stated plainly that shopping results "are not ads, they are not sponsored," which is a meaningfully different model than Google Shopping or Amazon's sponsored listings. That said, OpenAI hasn't published a full ranking algorithm, and I'd treat any article claiming to have cracked it with some skepticism. What we do know, from OpenAI's own statements and how the system behaves in practice, is a set of factors that clearly carry weight:
- Relevance to the actual query, not keyword matching. A search for "a quiet vacuum for a small apartment" favors listings whose structured data actually mentions noise level and dimensions over ones that just repeat the word "quiet" in the title.
- Feed data quality, meaning complete, accurate, and current Schema.org Product markup with pricing, availability, and reviews. A thin or stale feed is a fast way to get excluded from consideration entirely, regardless of how good the product is.
- Whether you're the primary seller or maker of the item, which OpenAI has referenced as a factor when multiple listings compete for the same product.
- Price and availability, checked against your live feed rather than a static catalog dump, since Instant Checkout only works if the item is actually in stock when the transaction happens.
- Brand trust signals, including review volume and rating quality, which line up with how ChatGPT already weighs sources for any factual answer, not just shopping ones.
The honest summary: this behaves less like a search engine you can game with backlinks and more like a procurement system that wants clean, verifiable data. If your feed is inconsistent with what's actually on your site, or your reviews are thin, you're giving the model reasons to route the sale somewhere else, even if your product is objectively better.
What brands need in place to be eligible
Instant Checkout access is currently rolled out to approved partners rather than opened up self-serve to every merchant, so the first requirement is simply getting into that pipeline. Beyond approval, here's what actually needs to exist before you're in a position to compete for the buy button:
- A structured, machine-readable product feed that matches your live site. If your feed says a product is $42 and in stock, and your actual checkout charges $48 or shows it backordered, that mismatch either kills the sale mid-transaction or gets your account flagged.
- Working ACP endpoints, meaning your engineering team (or your ecommerce platform, if it's built support in for you, as Shopify has done) can handle the checkout session lifecycle described above without manual intervention.
- A payment processor that supports delegated payment tokens. Stripe is the reference implementation, but the spec is processor-agnostic in principle.
- Sales tax and compliance handled on your end, since you remain the merchant of record for every transaction that closes through ACP.
- Genuine review and rating data, ideally aggregated through Schema.org markup, since this appears to function as a trust signal the model weighs when multiple sellers offer a comparable item.
- Accurate fulfillment metadata, including realistic shipping windows and return policy details, since ChatGPT is representing those terms to the buyer before the sale closes, and a bad experience there reflects on both you and the platform.
What this changes for your team
If you're running ecommerce for a mid-size brand, this isn't a marketing project you hand to whoever owns your Instagram account. It's an integration project that touches engineering, ops, and merchandising at once. Your product data needs to be clean enough to survive being read by a model instead of a human shopper scanning a page. Your customer service team needs a plan for orders that originate from a conversation instead of a cart abandonment email. And whoever owns pricing needs to understand that a stale feed doesn't just look bad, it can actively block you from being selected.
I think the brands that get ahead of this in the next year are going to be the ones already comfortable treating their product feed as a product in its own right, not an afterthought exported from their PIM once a quarter. That's true whether or not you ever flip on Instant Checkout specifically, because the same structured data discipline is what determines whether you show up in ChatGPT's shopping answers at all.
Instant Checkout vs. showing up in ChatGPT Shopping results
| Layer | What it does | What you need |
|---|---|---|
| ChatGPT Shopping (discovery) | Surfaces product cards and comparisons in chat answers | Clean structured data, strong reviews, relevant content ChatGPT can crawl and cite |
| Instant Checkout (transaction) | Lets the user complete the purchase in-chat | ACP-compliant endpoints, live product feed, payment processor support, OpenAI approval |
| Neither | Product mentioned in text but user must click through | Baseline: a site that's crawlable and a brand ChatGPT recognizes as relevant |
It's entirely possible to be strong in the first layer and absent from the second, and for most brands right now that's the realistic starting point. Getting cited and recommended in ChatGPT's shopping answers at all is table stakes, and it depends on the same structured-data and trust-signal fundamentals that also decide whether ChatGPT can even find you when someone asks it a fan-out style question. If you haven't looked at how the model breaks a shopping query into sub-searches, our piece on fan-out queries from ChatGPT is worth a read before you assume you know why you're missing from results. Instant Checkout is the next step for brands who've already got that foundation solid and want to remove the last click between recommendation and sale.
A word on timing and volatility
Worth being straight about this: OpenAI's shopping and checkout rollout has not been a smooth straight line. Reporting from outlets like CNBC has covered early stumbles in the agentic shopping push, and the spec itself has changed versions multiple times since its September 2025 release. If you're building an ACP integration, budget engineering time for the fact that endpoint requirements may shift again before this settles into something stable. That's normal for infrastructure this new, not a reason to wait on the sidelines, but it is a reason to build your integration in a way you can update quickly rather than hardcoding assumptions about today's spec.
Where to start
If Instant Checkout access isn't open to you yet, the highest-leverage work is still the same: get your product feed accurate and complete, make sure your reviews and ratings are structured with proper markup, and check that what ChatGPT can see about your pricing and availability matches reality. Every piece of that groundwork pays off in ChatGPT Shopping's discovery layer immediately, and it's exactly what you'll need the day your Instant Checkout application gets approved.
This is also, not coincidentally, the same groundwork that improves your visibility across every AI answer engine, not just ChatGPT's shopping surface. If you want a clearer read on where your brand currently shows up (or doesn't) across ChatGPT, Perplexity, and Google's AI answers, AI Peekaboo tracks exactly that, with real-time visibility monitoring built for ecommerce and agency teams managing more than one brand. Our roundup of the best ChatGPT visibility trackers for 2026 is a good starting point if you're comparing options, and our review of the best AI visibility tools for ecommerce brands covers tools built specifically for teams in your position.
If you want a second pair of eyes on whether your product feed and structured data are actually ready for agentic checkout, email me at filipe@aipeekaboo.com or grab time on my calendar. Happy to look at what you've got and tell you honestly whether you're close or still have real gaps.
