Product schema for AI shopping agents: what ChatGPT and Google actually read

There is a lot of noise about "optimizing for AI". Most of it is vague. The product data layer is not vague at all. It is a small set of fields with public specs, and AI shopping tools and Google both rely on it. This post covers what to publish, where, and what is live today.

Two ways a machine reads your catalog

An AI assistant or a search engine learns about your products in two ways:

1. Your product pages, through structured data in the HTML. This is schema.org markup, usually as a JSON-LD block.
2. A product feed you send on purpose: Google Merchant Center, or the feed programs that OpenAI, Perplexity and others run for merchants.

You want both. Pages give context and proof. Feeds give fresh price and stock at scale. When the two disagree, you lose trust.

The fields that matter on the page

Here is a minimal Product block that covers what most systems need:

<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "Product",
  "name": "Trail Runner 2 Waterproof Shoe",
  "image": ["https://yourstore.com/img/trail-runner-2.jpg"],
  "description": "Waterproof trail shoe with a 6 mm drop and a grippy outsole.",
  "sku": "TR2-WP-42",
  "gtin13": "0123456789012",
  "brand": { "@type": "Brand", "name": "YourBrand" },
  "offers": {
    "@type": "Offer",
    "url": "https://yourstore.com/products/trail-runner-2",
    "price": "89.00",
    "priceCurrency": "USD",
    "availability": "https://schema.org/InStock",
    "itemCondition": "https://schema.org/NewCondition"
  }
}
</script>

Product

name, image and description are the basics. Write the description as a fact sheet, not a slogan. An assistant needs to know what the product is, who it fits and what makes it different.

brand, sku and a GTIN (barcode) help a machine match your product with the same item elsewhere. That matters when an assistant compares stores for the same shoe.

Offer

This is where most stores fail:

• price: a plain number with a dot as the decimal mark, like 89.00. No currency symbol, no "from", no range in a single Offer.
• priceCurrency: the ISO code, like USD or GBP. Without it, a price is just a number.
• availability: InStock, OutOfStock, PreOrder and so on. An assistant will not recommend what it thinks it cannot buy.
• url: the canonical page of that offer.

If you sell variants, give each variant its own Offer, or use AggregateOffer with lowPrice and highPrice for the parent. Keep the visible price on the page identical to the markup. Google treats a mismatch as a problem, and so will any agent that checks.

Trust fields

Shipping details and return policy (shippingDetails and hasMerchantReturnPolicy) are optional in schema.org. They answer the questions shoppers ask assistants: "Do they ship to the UK?", "Can I return it?". Add an Organization block on your homepage too, with your name, logo, contact and social profiles.

Feeds: the official door

Page markup is read when a crawler visits. A feed is pushed on your schedule and carries every product. For shopping, feeds are the channel the big players formally support.

Google Merchant Center

Free product listings in Google Shopping come from Merchant Center feeds. It is free to join. On Shopify, the Google and YouTube app connects it. On WooCommerce, the Google for WooCommerce plugin does the same. If you only do one thing beyond page markup, do this one.

OpenAI and ACP

OpenAI and Stripe published the Agentic Commerce Protocol (ACP), which includes a product feed spec. Be clear about the timeline. OpenAI launched Instant Checkout in ChatGPT in September 2025, then reworked shopping in March 2026. Discovery now runs on merchant feeds, and the purchase happens on your own site. That program is US only today.

Google, Shopify and UCP

Google launched the Universal Commerce Protocol (UCP) in January 2026 with Shopify, Etsy, Walmart and others. Stores expose it at /.well-known/ucp. It runs in the US and Australia. Shopify turned on Catalog and UCP by default for eligible stores in June 2026. In September 2026, Shopify opened checkout to browser based AI agents through WebMCP.

Perplexity

Perplexity runs a free merchant program fed by product feeds, with its shopping assistant live in the US.

What is live, and what is not

Honest summary for October 2026:

• US: AI assistants can find, compare and recommend products from feeds and pages. Agent checkout exists in specific programs: Shopify through UCP and WebMCP for eligible stores, Microsoft Copilot Checkout, and Perplexity with PayPal.
• UK and the rest: assistants read and recommend, and send the shopper to your site to buy. Agent checkout has not opened there yet.

So the goal today is simple: let AI read you, understand you and recommend you, and be ready when checkout arrives in your market. Clean product data serves both goals.

Platform notes

Shopify

Most current themes output Product JSON-LD. Check that your theme still includes it after customization, and that every variant has its own price and availability. Shopify also serves a native llms.txt since May 2026, so you do not need an app for that.

WooCommerce

WooCommerce core adds Product structured data to product pages. SEO plugins like Yoast WooCommerce SEO or Rank Math extend it with brand, GTIN and more. The common bug is two plugins outputting two Product blocks that disagree. Keep one.

How to test your markup

• Run a product URL through Google's Rich Results Test. Fix errors first, then warnings.
• Check that price and currency in the markup match the visible price.
• Run the free AISaysYes scan. It checks product data, prices, robots.txt rules for AI bots and AI usage headers, and gives you a 0 to 100 score.

The full guide walks through each fix by impact. You can compare your score with other stores on the ranking. If you also want to tell agents what they can do on your store, the agents.md generator is free.

Stores that already passed the scan