AI Product Feeds: What Ecommerce Brands Need to Give Shopping Agents
What AI shopping agents need in a product feed: IDs, variants, price, availability, shipping, policies and checkout eligibility, plus validation and freshness.
Quick answer
AI shopping agents recommend and buy products based on structured data, not on how a product page looks. An AI product feed gives them that data: stable product IDs, SKUs and variants, factual titles and descriptions, price, availability, media, shipping and fulfillment options, return and seller policies and, where supported, checkout eligibility.
Two things matter more than in a classic shopping feed. IDs must match your checkout API, because an agent that selects a product must be able to buy that exact item. And freshness is critical, because an agent that acts on stale price or stock produces failed checkouts, not just a bad ad click.
Why agents need structured product information
A shopper scanning a product page can infer that 'ships in 2–3 days' applies to their country or that the blue option is the one in the photo. An agent cannot safely infer either. It compares products across merchants on explicit attributes, checks constraints the user gave it (size, budget, delivery date), and may create a checkout session for a specific variant. Every gap becomes a wrong recommendation or a failed purchase. Our guide to product data for AI search covers content quality; this article covers the feed agents act on.
The fields that matter
| Field group | What to send | Why agents need it |
|---|---|---|
| Product ID | Stable ID that matches your checkout and order systems | The selected item must be purchasable exactly |
| SKU and identifiers | SKU, GTIN, MPN, brand | Matching across merchants; avoiding wrong-item purchases |
| Title and description | Product type, key attributes, what is included; factual | Matching user intent; comparisons |
| Variants | Group ID plus per-variant ID, options (size, colour), price and availability | Choosing the right variant, not the parent |
| Price | Current price, sale price and sale window, currency | Budget constraints; checkout totals must match |
| Inventory and availability | in_stock / out_of_stock / pre_order / backorder, per variant | Avoiding orders that cannot be fulfilled |
| Media | Accurate images per variant | User review before purchase |
| Shipping and fulfillment | Regions, costs, speeds, pickup options | Delivery-date constraints; total cost |
| Policies | Return window and conditions, seller terms and privacy policy | Agents and users weigh risk; some platforms require them for checkout |
| Eligibility flags | Whether the item may appear in search and in agentic checkout | Controls where it can be bought |
| Freshness metadata | Last-updated timestamps | Agents and platforms judge reliability |
What current platform specifications ask for
Specifications differ by platform and change often, so treat these as examples checked in October 2026, not a permanent list. OpenAI's product feed specification requires item_id, title, description, url, brand, seller_name, image_url, availability and price in its own format, and also accepts a Google-compatible format. Optional fields cover variants (group_id), identifiers (gtin, mpn), shipping, returns (return_deadline_in_days, return_policy), reviews, and eligibility flags for search and checkout; checkout eligibility requires search eligibility, a separately enabled checkout integration and seller terms and privacy policy links (OpenAI product feed spec).
For UCP-powered checkout on Google, eligibility is set per product with a checkout-eligibility attribute in Merchant Center, and Google's documentation notes that the product ID in the feed must match the product ID your checkout API expects (Google UCP Merchant Center guide). Shopify merchants supply AI channels through Shopify Catalog rather than separate feeds; see Shopify agentic commerce.
AI product feed architecture
PIM / commerce platform (master product data)
│ products · variants · media · attributes
├───────────── inventory system (stock per location)
├───────────── pricing engine (price, sales, currency)
├───────────── shipping rules · return policy · seller terms
▼
Feed builder ── map fields per channel · validate · version
│
├─▶ OpenAI feed (full snapshot + updates)
├─▶ Google Merchant Center (incl. checkout eligibility)
├─▶ Shopify Catalog (via platform)
└─▶ other agents / marketplaces
│
▼ agent selects variant ID
Checkout API ── same IDs · re-checks price and stock liveValidation before you publish
- Every item has a stable ID that resolves in your checkout API
- Variants carry their own ID, options, price, availability and image
- Prices in the feed equal prices at checkout, including currency and tax treatment
- Availability values are from the allowed list and reflect sellable stock
- Shipping and return data are present for every country you sell to
- Required policy links resolve and are current
- Titles and descriptions are factual; no promotional claims the page does not support
- No duplicates or orphaned variants; group IDs are consistent
- Platform diagnostics show no errors; warnings are triaged
Update frequency and freshness
Separate slow-changing content (titles, descriptions, media) from fast-changing commercial data (price, sale windows, availability). Send full snapshots on a regular schedule, commonly daily, and push updates whenever price, sale status or stock changes. OpenAI's specification notes that date fields such as sale windows and expiration do not by themselves change price or stock, so current values must be sent. Most importantly, your checkout API must re-check price and stock at checkout time: the feed is for discovery, the checkout session is authoritative. For setting freshness requirements per field, see data freshness for AI.
Key takeaway
The feed gets you selected; the checkout session decides whether the order succeeds. Keep them on the same IDs and the same prices, and treat mismatches as incidents.
How this differs from feed SEO
Classic feed optimization focuses on titles, categories and attributes that win clicks in shopping ads and listings. That still matters; see ecommerce product feeds. AI feeds add operational correctness: purchasable IDs, per-variant availability, shipping and return facts, seller policies and eligibility for checkout. An optimized title cannot rescue an item whose variant ID fails at checkout.
Monitoring
Track feed errors and warnings per platform, price and availability mismatches between feed and checkout, checkout failures by reason (out of stock, price changed, item not found), and the age of the last successful update. Review AI-channel orders for wrong-variant purchases and returns caused by inaccurate data. See tracking AI-referred ecommerce sales.
Getting product data ready for AI shopping?
ZSpace Labs builds product data pipelines, feeds and checkout integrations for Shopify and custom commerce. See Shopify development.
Conclusion
AI product feeds are operational data, not just marketing data. Send complete, factual, per-variant information with IDs that match checkout, keep price and availability fresh, include shipping, returns and seller policies, validate before publishing and re-check everything at checkout. For how feeds fit the rest of the system, see the agentic commerce stack.
Common questions.
A structured file or API that sends a merchant's products to AI shopping platforms, including identifiers, descriptions, variants, prices, availability, media, shipping, return policy and seller details, so agents can recommend products accurately and, where enabled, start a checkout.