Keep the product and the offer distinct
A product describes what something is. An offer describes how a seller makes a particular variant available: price, currency, stock, delivery and terms. Mixing these layers makes it harder to diagnose whether a recommendation fails on relevance or on the ability to fulfil the purchase.
Use consistent identifiers where applicable and preserve the distinction between a product family and a purchasable variant. A capacity, size or processor difference can change suitability even when the marketing name is almost identical.
Audit five connected layers
The practical goal is agreement across the systems and surfaces you control, with an owner for each discrepancy. Do not add an attribute merely because a field exists; verify it first.
| Layer | What to inspect |
|---|---|
| Identity | Brand, model, variant, GTIN or other valid identifiers. |
| Suitability | Verified specifications, compatibility, use cases and limitations. |
| Representation | Agreement between visible page, structured data and feed. |
| Offer | Seller, price, currency, availability, delivery and returns. |
| Operations | Update frequency, validation, monitoring and correction ownership. |
Markup is a representation of the facts
Google documents Product and merchant-listing structured data for eligible search experiences. Its requirements differ by use case. OpenAI also documents a commerce product-feed format. These are specific interfaces; satisfying one format does not establish that every AI system will understand, recommend or transact on a product.
When page copy and structured data disagree, investigate the source of truth before changing either. A technically valid field can still contain the wrong fact. Validation needs both a format check and a content check.
Make freshness an operational responsibility
Choose a small sample of important variants and follow each one from source system to page, feed and offer. Record when each surface last changed and how errors are resolved. Give time-sensitive fields explicit update and escalation rules; a one-off audit cannot maintain live inventory.
- ↗Keep a register of discrepancies, source evidence and accountable owners.
- ↗Recheck a changed product across all relevant representations.
- ↗Track correction completion separately from AI recommendation outcomes.
Original documentation and published evidence. The working approach and priorities on this page are editorial recommendations, not platform guarantees. Reviewed 7 November 2026.
Google Search Central · Product structured data↗OpenAI · Product feed specification↗