Is your product data ready for a named shopping channel? A layer-by-layer check
Name one shopping channel, take one real product record and check it layer by layer: identity, identifiers, title, description, category, attributes, variants and offers, media, and policy. A feed moves data; it does not repair it. The check gives notes, not a score.
Published by Youzu · Editorial policy
- Published
- Reading time
- 5 min read
Key points
- Check one product record layer by layer against the channel you name, not channels in general.
- Keep in mind that a product feed moves data; it does not repair missing or wrong facts.
- Review the notes per layer; the 9-layer check is a checklist, not a score.
In this article
"Is our product data ready for AI shopping assistants?" is too broad to answer. Ready for which one? Each channel that takes product data publishes what it needs, and those needs differ. The useful question names a channel, picks one real record, and walks it through the layers that channel depends on. This article does that with a single worked record and a 9-layer checklist.
Name the channel first
Pick one destination: a shopping feed, an assistant that renders products and checkout, or your own site search. Then read its current documentation on the day you check. Google's Merchant Center specification, for example, lists required attributes such as id, title, description, link, image link, price and availability, and treats identifiers such as GTIN and brand as conditional or required depending on the product.[1] OpenAI's commerce documentation describes ChatGPT checkout as starting from a shared, structured product feed, with required fields that ensure correct display of price and availability.[2] Specifications change, so write the access date next to every requirement you copy.
The 9 layers
- Identity: a stable product id that does not change between sends.
- Identifiers: GTIN, brand or manufacturer part number where they exist.
- Title: says what the product is, within the channel's length limit.
- Description: plain text in the length the channel accepts.
- Category: mapped to the channel's category tree, not only your own.
- Attributes: typed values for the facts shoppers filter on, such as size, material and colour.
- Variants and offers: one product, its variants, and the price and stock of each offer kept apart.
- Media: images at or above the channel's minimum size, with a link that resolves.
- Policy and format: the channel's rules on claims, restricted items and file format.
One record, layer by layer
Take a single record: a 2-seat sofa in grey fabric. We check it against a feed specification that wants an id, title, description, link, image link, price, availability and a brand. The notes below are the output; there is no score.
| Layer | What the record has | Note |
|---|---|---|
| Identity | id SOFA-2041 | Present and stable. |
| Identifiers | No GTIN, brand set | Brand present; confirm the maker has no GTIN before declaring none. |
| Title | Sofa | Too thin: add type, seat count and fabric. |
| Description | 2 lines | Present; check it repeats nothing from the title. |
| Category | Furniture | Too broad for the channel's tree; map to a sofa category. |
| Attributes | Colour: grey | Width, depth and material are missing; leave empty until sourced. |
| Variants and offers | One price, in stock | Fine for one variant; a second fabric needs its own variant. |
| Media | One image, 400 by 400 | Below a 500 by 500 minimum where the channel states one. |
| Policy and format | Plain text | No claims to check; format matches the feed. |
4 layers need work and none of them is fixed by sending the record faster. That is the point of checking by layer: the feed carries the record to the channel exactly as it is.
How to read the notes
Sort the notes into 3 groups. The first is values that exist but are thin, like the one-word title; a person can fix those from what the product already shows. The second is values that are missing, like the sofa's width and material; these need a source, such as the supplier file or the product label, and should stay empty until one is found. The third is values the channel will not accept as they are, like an image below the stated minimum; these need a new asset or a different field. Treating the 3 groups separately keeps the work honest, because only the first group can be done at a desk without new information.
Common mistakes are easy to spot once the layers are written down. Teams fill the description with marketing text to make a field look complete. They copy one title pattern across every category. They send a single price for items that exist in several variants, so that the offer for one variant is shown for all. And they check the record once, then never again, although the channel's specification and their own catalog both keep changing.
What a feed does not do
Protocols and feed formats define how product data travels and which fields are required. They do not supply a missing width, correct a wrong category or decide which of 2 similar listings is the product. Those decisions happen upstream, in the catalog. A record that passes a format check can still be thin, and a thin record gives any search engine or assistant little to work with. This checklist promises no ranking.
Run it as a routine
- Check 10 records, not one: include the best, the typical and the worst.
- Group the notes by layer; the layer with most notes is where to start.
- Leave a value empty when the source does not support it, rather than guessing to fill a field.
- Repeat the check when the channel updates its specification, and date each pass.
Questions buyers ask
No. A feed and a protocol move data to a channel; they do not repair it. A missing attribute stays missing until someone sources it from the supplier or the product itself, and a wrong value stays wrong until a person corrects it. Check the record before it is sent, not after the channel rejects it.
No. Agent Readiness is an explainer and checklist for whether product and offer information suits a named machine-facing channel; it gives no score and human approval stays. Use the layer notes to decide what to fix first, and check again after each change to the data.
Clear, consistent product facts give search engines and shopping assistants better material to work with. No ranking is promised. What changes is the material a channel has to read, so the first benefit is a record that is complete, consistent and easy to check by hand.
Sources
Product data specification, Google Merchant Center Help. support.google.com/merchants/answer/7052112. Accessed . Back
Agentic Commerce: key concepts, OpenAI developer documentation. developers.openai.com/commerce/guides/key-concepts. Accessed . Back
Statements about other vendors describe what each vendor documents on the access date shown.
Related
Keep reading
Product
Agent Readiness: is your product data ready for AI shopping channels?
An explainer and checklist for whether product and offer data suits a named machine-facing channel: identity, attributes, offers, evidence, policy.
Guide
GTIN and GS1 identifiers for product matching
What a GTIN is, how EAN and UPC relate, how marketplaces use identifiers to match products and what to do when a listing has none.
Product
AI product data enrichment from the text and images you already have
Youzu fills the attributes your categories define from text and images, localizes titles and descriptions and leaves a field empty when evidence is thin.