Technical

The Agentic Commerce Protocol Explained: What ChatGPT's Instant Checkout Means for AEO

The Agentic Commerce Protocol (ACP) is the open standard behind ChatGPT's Instant Checkout, and it changes the optimisation target from 'get quoted' to 'get transacted with'. Here is what ACP actually requires, how a product feed differs from Product schema, and how to prepare even if you never turn checkout on.

Neil Walsh·July 2026·7 min read

The Agentic Commerce Protocol (ACP) is an open standard, co-developed by OpenAI and Stripe, that lets an AI agent complete a purchase on a user's behalf inside the chat interface itself, without redirecting to the merchant's website. Launched publicly in February 2026 as the mechanism behind ChatGPT's 'Buy it in ChatGPT' Instant Checkout feature, ACP is the first widely adopted protocol that moves AI commerce from citation to transaction: instead of merely recommending a product and linking out, ChatGPT can now take payment, confirm the order, and hand fulfilment to the merchant, all within the conversation.

For AEO practitioners this matters because it introduces a second, parallel optimisation surface alongside classic answer-engine citation. Getting a product mentioned in a ChatGPT response is a page-schema and content-structure problem. Getting a product into Instant Checkout is a data-feed and API problem, governed by a completely different specification with its own required fields, refresh cadence, and eligibility rules.

How Instant Checkout actually works

When a ChatGPT user asks a shopping question, the model can now do more than cite a product page in prose. If the merchant has published a compliant product feed and enabled ACP, ChatGPT can surface the item as purchasable, walk the user through size or variant selection inside the chat, and complete payment through Stripe's shared payment infrastructure. The order is then passed to the merchant's existing systems for fulfilment exactly as any other order would be.

  • The user asks ChatGPT a shopping question or browses a product it surfaces
  • ChatGPT checks whether the product is available through a merchant's ACP-compliant feed
  • If eligible, the user can complete the purchase inside the chat via Instant Checkout
  • Payment is processed through the shared Stripe-based payment layer defined by the protocol
  • The order and payment details are handed to the merchant's systems for fulfilment as normal

Merchants pay a small fee on completed purchases; the feature is free for shoppers, does not change the merchant's listed prices, and OpenAI has stated it does not influence which products ChatGPT recommends. That last point matters for AEO strategy: enabling ACP is not a ranking lever in the way schema completeness is, it is an eligibility gate for a specific purchase flow.

The product feed: a different artefact from Product schema

The core technical requirement of ACP is a structured product feed, submitted directly to OpenAI as CSV or JSON, rather than JSON-LD embedded in a page. This is the detail that catches teams out: years of AEO advice has focused on Product and Offer schema markup rendered into the HTML of each product page, and that work is not wasted, but it is not the same artefact ACP consumes. A feed is a standing, centrally hosted export of your entire catalogue that OpenAI polls on a schedule you control, independent of whether any individual page happens to be crawled that day.

Required feed fields

The specification defines a baseline set of fields needed simply to display price and availability correctly, and a broader set of recommended fields that improve ranking, relevance, and buyer trust once a product is eligible.

  • A stable product identifier and title
  • Description, brand, and category
  • Price, currency, and current availability status
  • Primary product image and any relevant variant media
  • Fulfilment and shipping details relevant to checkout

Recommended fields sit on top of that baseline: aggregate rating and review counts, richer media, and performance signals such as return rate or delivery speed. These do not gate eligibility the way the required fields do, but the guidance from OpenAI is consistent with what AEO practitioners already know from schema work generally: the more complete and specific the structured data, the more a model favours the listing it describes.

Feeds are expected to refresh frequently, in some implementations as often as every 15 minutes, so price and stock levels shown inside ChatGPT stay accurate. Treat this as an operational integration with your inventory system, not a one-off export.

Why this matters even if you never enable checkout

Instant Checkout is currently US-only and opt-in, so most merchants reading this will not turn it on this quarter. The reason to understand ACP now regardless is directional: it is the clearest signal yet that answer engines are moving from a citation model to a transaction model for commerce queries, and the underlying discipline it demands (a single, machine-readable, continuously accurate source of truth for every product) is the same discipline that already improves plain AI citation for shopping queries answered without checkout.

A merchant with clean Product and AggregateRating schema on every page, consistent pricing across every surface, and an inventory system that can export a live feed is already most of the way to ACP-ready. A merchant relying on stale, manually maintained product descriptions is not, and that gap will show up in ordinary AI citation long before it shows up in a checkout eligibility check.

Treat ACP as the leading edge of a broader shift covered in CiteRank's Agentic Search Optimization guide: AI systems increasingly need to verify and act on your data, not just read it. A product feed is one instance of a pattern that will keep recurring across other agentic surfaces.

Getting ready: a preparation checklist

  1. Audit whether your catalogue already has complete Product, Offer, and AggregateRating schema on every live page, since the underlying data is the same data a feed needs
  2. Confirm your inventory and pricing systems can export a structured feed on a short refresh cycle rather than a nightly or weekly batch
  3. Standardise product identifiers (SKU, GTIN, or MPN) across your site, schema, and any existing shopping feeds such as Google Merchant Center
  4. Review OpenAI's current developer documentation at developers.openai.com/commerce for the live field-level specification before building an integration, since the spec is still evolving
  5. Decide, separately from any technical readiness, whether Instant Checkout's fee structure and US-only availability make enabling it worthwhile for your business today

Who else is involved, and what's next

ACP is published as an open standard rather than a proprietary OpenAI format, and Stripe's involvement as co-developer signals an intent for other platforms to adopt the same payment rails rather than each AI vendor building an incompatible checkout flow. Whether Google, Perplexity, and Anthropic converge on ACP itself or ship comparable but distinct protocols is still unresolved as of mid-2026, which is itself a reason to build the underlying data discipline (accurate, structured, frequently refreshed product data available outside your own HTML) rather than integrating narrowly against a single vendor's current specification.

Frequently asked questions

What is the Agentic Commerce Protocol?

An open standard, co-developed by OpenAI and Stripe, that lets an AI agent complete a purchase on a user's behalf inside a chat interface. It defines how merchants submit structured product feeds and how payment and order handoff work, and it is the protocol behind ChatGPT's Instant Checkout feature.

Is ChatGPT Instant Checkout available outside the US?

At launch in February 2026 it was limited to US ChatGPT Plus, Pro, and Free users. International availability has not been confirmed, so merchants outside the US should still prepare the underlying product-data discipline without expecting to enable checkout imminently.

Does joining the Agentic Commerce Protocol cost merchants anything?

Merchants pay a small fee on completed purchases made through Instant Checkout. The feature itself is free for shoppers, does not change a merchant's listed prices, and OpenAI states it does not influence which products ChatGPT recommends or cites.

How is an ACP product feed different from Product schema markup?

Product schema is JSON-LD embedded in an individual page's HTML and read when that page is crawled. An ACP feed is a separate, centrally hosted CSV or JSON export of your whole catalogue submitted directly to OpenAI and polled on a schedule, independent of page crawling. Both should describe the same underlying data, but they are different artefacts with different delivery mechanisms.

How often does an ACP product feed need to update?

The specification expects a frequently refreshed feed so that price and availability stay accurate, with some implementations updating as often as every 15 minutes. This should be treated as a live integration with your inventory system rather than a periodic manual export.

Do I need to enable Instant Checkout to benefit from ACP-style optimisation?

No. The data discipline ACP requires, a single accurate, structured, frequently updated source of product truth, also improves ordinary AI citation for shopping queries that never reach a checkout flow. Building towards that discipline pays off even for merchants who never enable Instant Checkout itself.

Will other AI platforms adopt the Agentic Commerce Protocol?

ACP was published as an open standard rather than a proprietary format, and Stripe's role as co-developer suggests an intent for broader adoption. As of mid-2026 it is unconfirmed whether Google, Perplexity, or Anthropic will adopt ACP directly or ship a comparable but separate protocol.

Does enabling ACP change a product's ranking or citation rate in ChatGPT?

OpenAI states that participating in Instant Checkout does not influence which products ChatGPT recommends or cites; it is described as an eligibility gate for the checkout flow rather than a ranking signal. Ordinary AEO fundamentals, complete schema, review signals, and accurate content, remain what drives citation and recommendation.

Free tool

See your AEO score in seconds

Paste your URL and get a full audit across all 9 AEO signals - schema, crawlers, E-E-A-T, and more.

Audit my site - it's free

Related reading

Technical

Why Schema.org markup is the single biggest lever for AI citation

May 2026
Technical

Is your robots.txt accidentally blocking ChatGPT and Claude?

May 2026