Transaction execution · OpenAI and Stripe · Apache 2.0, live in ChatGPT
Agentic Commerce Protocol
Defines how AI agents discover products, initiate checkouts, and complete purchases. The transaction execution protocol for ChatGPT and its commerce partners.
Abbreviation: ACPThe Agentic Commerce Protocol is the transaction execution protocol of the agentic commerce stack. Developed by OpenAI with Stripe as the payment infrastructure partner, ACP defines how AI agents discover products, initiate checkouts, and complete purchases inside conversational AI surfaces - primarily ChatGPT.
ACP is live in production. Instacart, DoorDash, Shopify, and Etsy are among the commerce partners with active ACP integrations. When a ChatGPT user asks about a product and the agent recommends a purchase, ACP is the protocol that handles the checkout flow from product recommendation through payment execution.
The protocol's architecture evolved significantly following the ChatGPT Instant Checkout failure in early 2026. OpenAI pivoted from a direct checkout model to a discovery-first architecture, where the agent recommends and the user confirms rather than the agent transacting autonomously. This pivot reflects a broader pattern in agentic commerce: the consumer trust ceiling constrains what is technically possible.
ACP is Stripe-native for payment execution. This means Stripe is not merely a supported payment option - it is the payment infrastructure layer. Merchants without Stripe integration require additional payment bridge configuration. Shopify merchants benefit from automatic ACP integration through Shopify Agentic Storefronts.
How does an agent initiate and complete a checkout inside a conversational AI surface?
What ACP provides
Product discovery layer: structured product data querying for agent evaluation
Checkout initiation: agent-native checkout flow without human-navigated steps
Payment execution: Stripe-native payment rails with tokenised agent authority
Partner ecosystem: Instacart, DoorDash, Shopify, Etsy live in production
Shopify Agentic Storefronts provide automatic ACP integration for merchants on the platform
Before you begin
The following must be in place before starting ACP integration. Missing prerequisites are the most common cause of delayed or failed protocol deployments.
Stripe account with API access (ACP is Stripe-native for payment execution)
OpenAI ChatGPT commerce partner application approved
Complete product data meeting ACP attribute requirements
Webhook infrastructure for order confirmation and fulfilment signals
What goes wrong
These are the failure modes the AXD Institute has observed across early ACP implementations. Each includes the root cause and the recommended mitigation.
Data quality below ACP threshold
ACP agents score products on attribute completeness before recommending them. Products scoring below the ACP recommendation threshold are excluded from ChatGPT shopping results even if technically integrated.
Webhook latency
ACP requires order confirmation within 10 seconds. Synchronous order creation pipelines that process sequentially often fail this requirement. Implement asynchronous order processing with an immediate confirmation response.
Price inconsistency
ACP checks prices against your catalog in real time. Price discrepancies between your ACP endpoint and your main checkout (caused by promotions or dynamic pricing not reflected in the ACP feed) trigger checkout failures.
Step-by-step walkthrough
A sequenced implementation guide for ACP. Each step includes the action, the rationale, and the validation criteria for completion.
Stripe account and ACP partner application
ACP requires a Stripe account with API access and an approved ChatGPT commerce partner application. These are prerequisites, not optional enhancements.
- •Verify your Stripe account has API access enabled and is in good standing.
- •Submit the OpenAI ChatGPT commerce partnership application. Approval timelines vary; submit early.
- •If on Shopify Plus, enable Shopify Agentic Storefronts instead - this provides automatic ACP integration without a separate partner application.
- •Configure webhook infrastructure for order confirmation and fulfilment signals. ACP requires order confirmation within 10 seconds.
Product data preparation for ACP scoring
ACP agents score products on attribute completeness before recommending them. Products below the ACP recommendation threshold are excluded from ChatGPT shopping results even if technically integrated.
- •Audit product data against ACP attribute requirements: name, SKU, real-time price, availability, description (50+ characters), product attributes (material, dimensions), and return policy.
- •Ensure price consistency between your ACP endpoint and your main checkout. Price discrepancies caused by promotions or dynamic pricing not reflected in the ACP feed trigger checkout failures.
- •Implement structured review aggregates (schema.org) - ACP agents weight review data in recommendation scoring.
- •Target 90%+ attribute completeness. Products in the 70-90% range appear intermittently; products below 70% are effectively invisible.
Build ACP product and checkout endpoints
Implement the product discovery layer and checkout initiation flow that ACP agents consume.
- •Product discovery endpoints: structured product data querying that returns complete product information in ACP-specified format.
- •Checkout initiation: agent-native checkout flow without human-navigated steps. The agent initiates checkout; the user confirms.
- •Payment execution: configure Stripe agent payment rails with tokenised agent authority.
- •Implement asynchronous order processing with immediate confirmation response. Synchronous pipelines that process sequentially often fail ACP's 10-second confirmation requirement.
Sandbox testing and validation
Complete ACP sandbox test scenarios before production activation. Focus on the failure modes that cause the most common ACP integration rejections.
- •Test product discovery: verify all products meeting the attribute threshold appear in agent queries.
- •Test checkout flow: complete end-to-end checkout from agent recommendation through Stripe payment execution.
- •Test price consistency: verify that prices returned by ACP endpoints match your live catalog in real time, including during active promotions.
- •Test webhook responsiveness: confirm order confirmation webhooks fire within the 10-second requirement under load.
- •Test error handling: verify graceful handling of out-of-stock items, payment failures, and session timeouts.
Production activation and AACR measurement
Activate ACP in production and begin measuring Agent-Assisted Conversion Rate from Day 1.
- •Activate ACP in production through the OpenAI commerce partner dashboard or Shopify Agentic Storefronts.
- •Implement AACR (Agent-Assisted Conversion Rate) tracking for ACP-attributed checkout sessions.
- •Set up surface-specific UTM parameters to distinguish ACP traffic from UCP and direct traffic.
- •Monitor the ACP recommendation threshold - products that were visible in sandbox may not meet the production threshold if data quality degrades.
- •Schedule a 30-day review comparing AACR across ACP and UCP surfaces to identify optimisation priorities.
Related metrics, frameworks, and essays
AXD Metrics Standard KPIs
AXD Practice Frameworks