Product Data for AI Agents: What Your Catalog and Feed Need
Product data for AI agents decides what gets recommended. See the six fields that matter, common feed mistakes, an update rhythm and fixes sorted by effort.
Services
Page
Agentic Commerce Readiness Checklist: What to Fix Before AI Agents Shop Your Store
Want to discuss
your project?
your project?
Key takeaways
- Why it matters: an AI shopping agent cannot see your design. It reads fields, calls APIs and compares numbers. A store that fails at any layer gets skipped, and nobody sees the failure.
- Readiness has an order: catalog data, feed, checkout, payments, protocols, operations. Each layer depends on the one before it.
- The first fifteen checks are mostly data work. Most stores can fix them in days, without touching the platform.
- Checkout and payments are where real engineering sits. The answer there depends on your platform and your payment provider.
- Score every item as Ready, Partial or Missing, then fix the Missing items from the top of the list down.
Most "agentic commerce" advice starts with protocols. Merchants then spend a quarter reading about ACP and UCP and never find out that their feed lists half the variants without stock data. The protocol work matters, and it comes late in the list.
How to score yourself
Go through each check and mark it Ready, Partial or Missing. Partial counts as Missing when the gap affects price or stock. A store with six or more Missing items in the first two groups is not ready to spend time on checkout integration yet.
1. Catalog data (checks 1 to 8)
An agent matches, filters and compares products by their fields. This group decides whether your products can be understood at all. The details are in our guide to product data for AI agents.
1. Every product and variant has a unique, stable identifier that matches the feed.
2. Titles name the brand, the product type and the main attributes in plain words.
3. Descriptions state facts. Slogans and promotional phrases are removed.
4. GTIN, MPN and brand are filled in wherever they exist.
5. Each variant (size, color) is its own record with its own ID, price and stock level.
6. Attributes such as material, dimensions and compatibility are structured fields of their own, kept out of the description text.
7. The main image is clean, and every image has a text description.
8. Every product is mapped to a standard category taxonomy.
2. Feed and structured data (checks 9 to 15)
Agents do not query your live catalog. Protocols such as ACP expect you to push a feed in advance, so the feed is what the agent believes. OpenAI's merchant documentation describes a regularly refreshed feed with identifiers, pricing, inventory, media and fulfillment options.
9. The feed carries price, currency and availability for each variant.
10. The feed refreshes at least daily, and price and stock changes go out faster than that.
11. The price in the feed equals the price at checkout, including tax and shipping rules.
12. Product and Offer markup on product pages matches the feed field by field.
13. Google Merchant Center is connected, and its diagnostics show no critical errors.
14. Reviews and ratings are published as readable text.
15. Shipping and return policies are published as text and, where possible, in markup.
Stale price and stock data is a common failure here. An agent that quotes a price your checkout does not honor creates a complaint and an abandoned order.
3. Checkout (checks 16 to 21)
Both major protocols expect your store to expose checkout steps through an API: create a session, update it, complete it. Google's UCP guide for Merchant Center describes native checkout as three core endpoints for exactly that. Your platform may already provide them, so check before building anything.
16. A cart and checkout session can be created, updated and completed through an API.
17. Guest checkout works without an account.
18. Shipping options and taxes can be calculated from an address in a single call.
19. An out-of-stock item returns a clear error at cart time, not at payment.
20. Order confirmation and order status are available in machine-readable form.
21. Discount codes and promotions behave the same way on every call.
4. Payments and risk (checks 22 to 26)
In agentic checkout, the agent does not type a card number. It passes a token. OpenAI's documentation describes a Delegated Payment Spec, and Stripe's Shared Payment Token is its first compatible implementation. Under ACP you stay the merchant of record, so settlement, refunds and disputes remain yours.
22. Your payment provider supports delegated, tokenized payments for agent orders.
23. Refund and dispute handling for agent-initiated orders is defined and documented.
24. Fraud rules are reviewed for orders that arrive without a normal browser session or device fingerprint.
25. Requests are idempotent, so a retried call cannot create a second order.
26. Regulated and age-restricted products are flagged. UCP defines a consumer notice attribute group for them.
5. Protocols and discovery (checks 27 to 30)
27. You know which protocols your platform already supports, and which ones need work. Our explainers cover ACP and UCP, and ACP vs UCP shows how they differ.
28. If you use UCP, your profile at /.well-known/ucp declares exactly the capabilities you enabled in Merchant Center.
29. Products meant for agent checkout are marked for it in the feed. In UCP this is the native_commerce attribute.
30. You have decided which AI crawlers may read your site, and your robots rules say so.
Approval programs and availability differ by market and change often. Check each protocol's current documentation before you plan a launch date.
6. Operations (checks 31 to 34)
Readiness needs upkeep. Data drifts, and feeds break.
31. One named person owns feed quality.
32. You monitor feed errors, rejected products and failed agent sessions.
33. Once a month, someone asks the main assistants about your top products and records what comes back.
34. Support and finance know how to find and handle agent-placed orders.
What to fix first
Sort your Missing items by effort. Items 2, 3, 4, 14 and 15 are copy and data edits, and they take hours. Items 9, 10 and 11 usually need a feed or platform setting. Items 16 and 22 need your platform vendor and your payment provider, so start those conversations early, because they have the longest lead time.
If you want an outside read on the list, our agentic commerce readiness audit goes through these areas against your actual catalog, checkout and payment setup in one week at a fixed price.
FAQ
How many of the 34 checks do we need before we start?
Pass most of groups 1 and 2 first. Without clean data and a fresh feed, checkout and payment work has nothing to stand on.
╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌
Do we need to support both ACP and UCP?
Not on day one. Start with the protocol your platform and payment provider already support, and make sure the feed and structured data are good enough to serve either.
╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌
Who should own this inside the company?
Product data usually sits with the catalog or merchandising team, and checkout and payments with engineering. Name one person to track the whole list, or items fall between teams.
╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌
How long does it take?
Data fixes take days to weeks. Checkout and payment changes depend on your platform and provider, and program approvals can add weeks, so start those early.
╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌
Does this help with regular search too?
Yes. Clean identifiers, accurate prices and matching structured data also help your results in Google Shopping and in organic search.
Conclusions
Which protocol should your store prepare for first?
Need expert advice on your project?
Schedule a call with our team to discuss your needs and get expert guidance.
Related articles
What Is Agentic Commerce? How AI Agents Change Online Buying
What is agentic commerce? How AI agents search, compare and buy on a shopper’s behalf, which protocols make it work and what it changes for online stores.
What Is the Agentic Commerce Protocol (ACP)? How OpenAI’s Standard Works for Merchants
What is the Agentic Commerce Protocol? How ACP connects AI agents, merchants and payment providers, what changed in 2026 and what merchants must prepare.