What Is the Agentic Commerce Protocol (ACP)?
On September 29, 2025, Stripe announced Instant Checkout in ChatGPT and released the Agentic Commerce Protocol (ACP), an open standard co-developed with OpenAI to enable commerce flows between AI agents and merchants.
Why ACP exists
Traditional e-commerce was built around a merchant-controlled website. In agent-led commerce, a conversational agent sits between the buyer and the seller, helping the customer discover products and complete a purchase. Stripe and OpenAI designed ACP as a "shared language" so merchants do not need to build custom integrations for every agent interface.
In practical terms, ACP focuses on two connected needs:
- Discovery: the agent must understand what you sell (accurate products, pricing, availability).
- Checkout: the agent must complete a transaction without forcing the customer to leave the conversation.
ACP and product feeds: making products readable to ChatGPT
To make product discovery reliable, OpenAI introduced a Product Feed spec. The goal is straightforward: provide structured product data so ChatGPT can index and display items with up-to-date price and availability.
This is a shift away from "hope the model finds my site" toward "publish merchant-controlled catalog data." It also creates a new optimization surface. Your catalog quality, attribute completeness, and consistency start to directly impact how often an agent can confidently recommend your products during a conversation.
ACP and Instant Checkout: buying inside the chat window
Instant Checkout is the conversion layer. Stripe describes the flow as an inline checkout experience inside ChatGPT, where the user asks for recommendations and then completes the purchase without leaving the conversation.
Under the hood, ACP standardizes how the agent and merchant coordinate the checkout lifecycle. Stripe's documentation explains that a seller exposes a small set of endpoints (Create, Update, Complete, Cancel) that let the agent initiate a checkout, adjust it as the customer makes choices, and finish the payment. ACP can be implemented as a REST interface or an MCP server.
A key piece here is the Shared Payment Token (SPT). Stripe describes SPT as a payment primitive that allows an application like ChatGPT to initiate payment without exposing the buyer's payment credentials to the merchant or the agent. The token is scoped to a specific merchant and basket total, and the merchant uses it to process the transaction in their existing stack.
OpenAI's commerce guidance aligns with that lifecycle: ChatGPT collects buyer and fulfillment details, calls the merchant's ACP endpoints to create or update the checkout session, and then completes the purchase.
What merchants keep control of
A common concern is whether "the agent owns the customer relationship." Both Stripe and OpenAI position ACP so merchants retain core responsibilities as merchant of record, including order management, fulfillment, returns, and support, while the agent provides the conversational interface.
What this changes for product teams
ACP effectively introduces two new "AI-facing" surfaces you have to manage:
- Your product knowledge layer: feeds, attributes, taxonomy, policies, and brand materials that influence what an agent can recommend and how confidently.
- Your transactional interface: standardized checkout endpoints that allow a conversation to become an order.
In other words, it is not just "a new checkout." It is a new distribution channel where product data quality and semantic clarity become first-class requirements.
A practical recommendation before you go live
Stripe's release and docs make the point that agentic commerce requires merchants to expose products, pricing, and checkout in a way agents can reliably use.
In practice, that means you should test your catalog and content before you publish feeds or enable a production checkout. You want to validate how major models interpret your products, identify missing attributes, and catch ambiguity early. That is exactly what tools like ventic.ai are built for: an expert workflow to test, optimize, and serve AI-ready product knowledge through APIs before it becomes a live sales channel.