Prepping your product feed for agentic checkout after Instant Checkout: the complete merchant guide.

The Agentic Commerce Protocol checkout flow has a precise four-endpoint structure. According to Svix’s integration guide, the sequence is: (1) ChatGPT calls “create checkout session” with cart contents and buyer context when a customer signals intent to buy; (2) “update session” is called as the customer modifies their selection; (3) “complete checkout” finalizes the order; (4) “cancel checkout” can be called at any point to abort. Each endpoint must return structured, predictable responses within tight latency windows.

SALT.agency’s technical analysis frames this well: “The Checkout endpoints are your agent-facing transactional APIs.” They are the equivalent of your storefront’s buy flow, rebuilt to speak directly to AI agents rather than browsers. Merchants on Shopify get partial scaffolding via Shopify’s native agentic storefront support, but custom implementations require building these endpoints against the published ACP spec. Pair this work with the webhooks layer below — the two are interdependent.

The ACP spec requires merchants to send webhooks for two order lifecycle events: order_created and order_updated. These keep ChatGPT — and by extension the buyer — up to date on order status after checkout completes. Without them, the agent has no reliable way to answer “where is my order?” queries, which breaks the conversational commerce experience.

Svix’s review of the ACP notes this is one of the more dynamic parts of the spec and is likely to expand as the protocol evolves. Invest in webhook delivery infrastructure that provides reliable retries and delivery guarantees rather than fire-and-forget HTTP calls — agents polling for status on behalf of buyers will expose any gaps in your delivery reliability quickly. See our overview of connecting AI tools to your commerce stack for related architectural guidance.

The ACP payments layer supports two paths: a delegated payment spec that keeps authorization within your existing payment infrastructure, or integration with a provider like Stripe that abstracts the agent-facing complexity. Stripe’s Agentic Commerce Suite lets merchants upload their product catalog directly to Stripe, which then distributes it across AI agents in the formats each platform requires — solving the multi-format problem where brands have had to reformat the same catalog six different ways to list across multiple agents.

Stripe’s technical documentation is explicit about the emerging maintenance burden: different agents want data via SFTP file drop, custom API integration, or their own feed specs entirely. A commerce provider that handles distribution removes that ongoing overhead. For merchants not on Stripe, DAC Group’s ACP guide confirms the delegated payment spec is open — you can implement it against your existing payment processor as long as it meets the real-time authorization requirements of the ACP checkout flow.

Stripe’s agentic commerce research surfaces a critical infrastructure edge case: while a human shopper browses one product at a time, an AI agent shopping on their behalf might query 50 products in under 2 seconds, checking availability, comparing prices, and reading specifications across your entire catalog nearly instantaneously. A standard Web Application Firewall may flag this as a DDoS attack and ban the IP entirely.

The recommended configuration: add a dedicated WAF rate-limit rule for known agent user-agents (e.g., GPTBot, Google-Extended) that returns a 429 Too Many Requests with a Retry-After header instead of a hard block. Cache read-heavy product detail endpoints aggressively to prevent timeout cascades. This signals to agents “you are allowed here, please pace yourself” rather than treating legitimate discovery traffic as an attack.

OpenAI’s updated ranking documentation states that merchants are ranked based on factors including availability, price, quality, whether they are the maker or primary seller of the item, and whether Instant Checkout is enabled. Crucially, product results are selected independently and are not ads — they are driven by merchant and product metadata from third-party providers or directly from merchants themselves.

Search Engine Land’s analysis of ChatGPT shopping readiness highlights a key question every merchant should audit: “Are key product descriptors present — especially those that influence customer decisions? In the conversational context of ChatGPT queries, these attributes carry high value.” Product pages that answer the specific questions AI users ask (“what are the dimensions,” “does it come in midnight blue,” “what is the return policy”) will outrank thin pages over time. This is where feed prep and autonomous page optimization tools like Ryze AI compound each other most powerfully.