Machine-readable product data is the part of a product page that software can read without guessing: a name, an image, a price with its currency, a stock status and an identifier, written as structured markup (usually JSON-LD) rather than as text on the page. If a fact is only visible in the design, an agent has to infer it. If it is in the markup, the agent can read it.

There is no single “AI-ready” standard. The vocabulary is schema.org, and the most precise public statement of which fields matter for selling is Google’s merchant listing documentation. This article uses both, as fetched on 2026-10-06.

What Google requires on a product you sell

Google separates two uses of Product markup: product snippets for pages where people can’t directly purchase the product, and merchant listings for pages where customers can. For merchant listings, the documentation lists these required properties:

  • name: the name of the product.
  • image: a crawlable, indexable product photo URL.
  • offers: a nested Offer. Merchant listings require an Offer, not an AggregateOffer, because the merchant has to be the seller.
  • price (or priceSpecification.price): the current, active price. Merchant listings require a price greater than zero.
  • priceCurrency: a three-letter ISO 4217 code, required whenever a price is given.

A price without a currency is therefore not a complete price, by Google’s own rule.

What Google recommends, and why it matters to a buyer

Several fields are optional for eligibility but carry the facts a purchase decision depends on:

  • availability: one value from the schema.org list (InStock, OutOfStock, PreOrder, BackOrder, SoldOut and others). Google says not to specify more than one.
  • Identifiers: the documentation says to include all applicable global identifiers and to use the most specific GTIN that applies, in numeric form, not as a URL.
  • brand.name, description, color, category and aggregateRating.
  • Shipping and returns: OfferShippingDetails and MerchantReturnPolicy. Google calls them optional, but they are where delivery time and return terms become readable facts.

schema.org’s Product type also defines sku, mpn, hasVariant and isVariantOf, which is how a parent product and its variants are tied together. Google notes that both product snippets and merchant listings support variants.

Where this goes wrong

Four patterns recur on store pages:

  1. Visible but not marked up. The price renders on screen but no Offer exists in the markup.
  2. Number without currency. price is present, priceCurrency is not.
  3. Markup that disagrees with the page. Google’s documentation says images must represent the marked up content; the same principle applies to prices and stock.
  4. Variants missing. The parent has a price range but no individual Offer per size or colour.

None of this is exotic. It is the same markup that makes a page eligible for Google’s shopping surfaces, and the same data an AI shopping agent reads before it adds anything to a cart. For the field-by-field audit view, see what shopping agents actually parse and the JSON-LD guide.

How to check yours

  1. Open a product page and view the page source, not the rendered DOM.
  2. Search for application/ld+json and confirm a Product with a nested Offer.
  3. Confirm price, priceCurrency and availability are all present and match the visible page.
  4. Run the page through Google’s Rich Results Test, which the documentation recommends for validating markup.

FAQ

What is machine-readable product data?

It is product information published as structured markup, typically schema.org JSON-LD, so software can read the name, image, price, currency, stock status and identifiers directly. It does not replace the visible page; it states the same facts in a form a program can parse without interpreting the layout.

Which product fields does Google require?

For merchant listings, Google requires name, image and a nested Offer containing a price greater than zero and a three-letter currency code. Availability, GTIN and other identifiers, brand, shipping details and return policy are recommended rather than required.

Is there an official standard for AI-ready product data?

Not a dedicated one. schema.org provides the vocabulary, and Google’s merchant listing documentation is the most detailed public statement of which properties matter for selling. Agent-specific protocols build on top of this data rather than replacing it.

Sources

An agent can only buy what the markup states, so the transaction starts with the product page’s own data.