Product Data an Agent Understands

A human reads a product page and fills the gaps from experience: that “in stock” probably means it ships this week, that the price excludes shipping, that the sizing runs small. An agent does none of that. It takes what is marked up and skips the rest. Whoever wants to be compared has to be comparable.

The fields it hangs on

Four pieces of information decide whether a product appears in a machine comparison at all:

• Price with currency, and specifically the one that applies at the end.
• Availability as a status, not as a sentence – “in stock” inside prose is not a state to a machine.
• Shipping cost and delivery time, because an agent compares total cost, not item prices.
• Identifiers (GTIN, brand, manufacturer part number). They are the key by which your product and the same product elsewhere are recognised as the same.

Without an identifier you are not sorted incorrectly – you are not sorted at all. That is the difference between ranked badly and absent.

Why prose does not rescue this

Descriptions are good for humans and weak for comparisons. “Runs small” is useful information and not a field. The rule is not “less text” but: anything that is a value belongs in a field – dimensions, material, quantity, compatibility. The text then explains what the fields cannot say.

Returns and terms are part of it

What an agent needs for a purchase decision does not end at the price. Return window, warranty period and payment methods are part of the comparison, and in many stores they exist only in the terms and conditions – a document nobody reads in a structured way. Making those machine-readable is the step most stores still have ahead of them.

Two things not to do

Putting promotional claims in titles and fields. “Best price”, “top quality”, emoji: for matching, that is noise, and it degrades recognition of the actual product.

Maintaining several versions of the truth. If your feed, your product page and your structured data state different prices, the deciding value is neither the highest nor the lowest but whichever the machine finds first – and your store gets classified as unreliable. One source, all outputs derived from it.

Where to check what you have

Whether the markup holds up technically is something an outside look will tell you: our Structured Data Check tests whether the JSON-LD is valid and which merchant fields are missing, and the Agent-Ready Check tests the standards of the agentic web around it. Both are free and give a fix guide per finding – so the next step after this lesson is to run your own store through them.