Product feeds for AI agents: build a catalog buyers can rely on

A useful product feed tells a shopping platform exactly what you sell, which version is available and what it costs. Build it from your working catalog, match it to your product pages, and check that buyers reach the same offer at checkout.

Reviewed October 1, 2026 · 8 min read

A product feed is a structured catalog submitted to a shopping platform. Each record describes an item or variant using fields such as its identifier, title, price, availability and landing page. For AI shopping, those details help a system compare products against a customer’s requirements.

Your feed is one source among several. A platform may also use your product pages, structured data, account settings and supported commerce interfaces. These sources should describe the same offer, even though they serve different purposes and may use different formats.

This guide explains which data to prioritize, how to prevent mismatches and what a feed can and cannot do. A complete catalog improves the information available to shopping systems; it does not guarantee that a platform will list or recommend every item.

What product data helps an assistant compare?#

Consider a request for waterproof hiking boots in a wide size 10, under $180. A useful catalog identifies the exact variant, its fit, price and current stock. A title that says only Adventure Boot leaves the system with more work and fewer facts it can verify.

Put the important attributes in their designated fields as well as readable copy. A clear description helps explain fit and use; a size field lets a platform distinguish one variant from another. Avoid stuffing every search phrase into the title.

Start with the platform’s current specification. Google’s product data requirements vary by product and destination. Required fields establish eligibility; recommended fields can add useful detail. Neither category is a promise of a particular ranking.

Feeds, page markup and commerce interfaces#

Keep these three parts distinct when planning the integration:

For Google integrations, Merchant Center holds product information while UCP can support eligible commerce interactions. UCP is not simply another universal product-feed format. Follow the platform’s onboarding and capability requirements for the experience you want to support.

Your own page markup comes from you, so it is not independent corroboration of your feed. Its value is consistency and clarity: Product and Offer JSON-LD can express the same facts visitors see, while shipping and return markup add commercial terms where supported.

Keep the feed, landing page and checkout consistent#

Google documents price mismatches as a reason for product disapproval. Check the submitted offer against the selected variant on the landing page and the amount shown at checkout. Apply the platform’s rules for taxes, shipping and other charges rather than treating every price field as interchangeable.

Generate exports and page markup from the same catalog wherever possible. Then monitor the exports: a shared source reduces manual drift, but caches, failed jobs and delayed platform ingestion can still leave stale data in circulation.

The product attributes to prioritize#

Audit these fields first, then add the attributes that answer questions specific to your products:

• Stable identifiers and variants. Keep item IDs consistent; include accurate GTINs or other required identifiers when available, and group related variants correctly.
• Clear titles and descriptions. Name the product and its distinguishing attributes. Explain materials, dimensions, fit, compatibility and limitations using facts you can support.
• Price, currency and availability. Include applicable sale details and effective dates. Make sure the selected variant can actually be purchased on the stated terms.
• Useful images. Show the correct product and variant clearly, following the destination’s image requirements.
• Shipping and returns. State supported destinations, costs, delivery estimates and return conditions in the appropriate feed or account settings.
• Working landing pages. Link to the right offer, preserve variant selection and make the information accessible without unnecessary interaction.

Google has also announced conversational product attributes for questions, compatible accessories and substitutes. Check which fields your account and data source currently support. Add useful answers without inventing compatibility or treating an announcement as universal availability.

Choose the integration your platform supports#

Start with your existing commerce system and the destination where customers are likely to find you:

1. On a hosted commerce platform: inspect the installed sales channel, catalog export and account eligibility. Platform support does not mean every merchant or product is automatically enabled.
2. With an existing Merchant Center account: resolve diagnostics, check variant links and improve the data already being submitted before adding another integration.
3. On another commerce platform: use a maintained connector where it covers your requirements. Verify how it handles stock changes, deletions and export failures.
4. With a custom catalog: build an export from your source data to the destination’s current specification, with logs and alerts for rejected records.
5. For ChatGPT product discovery: check OpenAI’s current merchant application and feed requirements. Its checkout approach changed in 2026, so separate discovery work from the purchase integration.

What if you sell services instead of products?#

A retail product feed may be a poor fit for an appointment or a custom quote. Check the destination’s supported categories rather than forcing every service into a shippable-product model. A consultation needs duration, location, price basis and booking conditions more than a size or shipping weight.

Publish named service pages with clear terms and an accessible booking or inquiry path. Where useful, a supported API or MCP server can expose availability and actions, but it is not the only route. The service booking guide covers ordinary booking links, handoffs and tool integrations.

A practical product-feed audit#

Choose a sample that includes bestsellers, variants, sale items and recently out-of-stock products. Then work through these checks:

1. Compare each feed record with its landing page, JSON-LD and checkout. Record the exact field that disagrees.
2. Check identifiers and variant grouping. Confirm that links open the item and configuration the feed describes.
3. Review titles and descriptions against real customer questions. Add missing specifications before adding more adjectives.
4. Verify shipping, returns and promotions for the intended country and currency.
5. Change a test item and measure how long each downstream source takes to update. Set a refresh target appropriate to your stock turnover.
6. Monitor rejected records, stale exports and failed jobs. Assign someone to investigate alerts and confirm the fix reached the destination.

A useful feed makes comparison easier and reduces surprises after the click. Judge the work through accepted records, accurate listings and completed customer journeys, alongside impressions and referrals. More attributes are valuable when they help someone choose the right product.