AI Shopping Agents Are Coming for Your Checkout. Prepare the Data.

The question for a store is whether an agent can understand what you sell without a human to interpret the page.

What an agent needs from a product page

A person looks at a photo and a price and fills in the rest. An agent needs the facts stated explicitly, in a place it can parse without rendering a JavaScript app.

The minimum, per product:

• Name, brand, and a stable identifier (GTIN, MPN, or SKU)
• Price, currency, and whether that price is current
• Availability, stated plainly: in stock, backordered, discontinued
• Variants with their own prices and stock (size, color, capacity)
• Shipping cost and delivery window for the buyer's region
• Return policy, in a sentence
• Specifications as structured attributes, not a paragraph

Product schema carries most of this. In our testing, stores with complete Product and Offer markup tend to be described accurately by agents, while stores with a photo, a price, and a lifestyle paragraph tend to be skipped or misdescribed.

Where the data usually breaks

The common failures are the same gaps that break Google Shopping feeds, now with higher stakes.

Price lives only in JavaScript. Many storefront themes render price client-side, so the HTML an agent fetches shows a placeholder. Run the page through a JavaScript visibility checker and see what a non-rendering fetch returns.

Variants collapse into one URL. If a medium and an extra-large share a page and the only way to see the price difference is a dropdown, the agent sees one price and may quote the wrong one.

Availability is a badge, not a field. A red "Sold out" image tells a human everything and a model nothing. Use the availability property in Offer schema and say it in text too.

Shipping is hidden behind checkout. An agent asked for total cost cannot answer if shipping only appears after an address is entered. Publish a shipping policy page and mark it up with shippingDetails.

Specs are in a PDF or an image. Spec sheets as images are invisible. Specs as a table in HTML are gold.

Agents also read off-site

An agent deciding between your jacket and a competitor's does not stop at your page. It tends to pull from review aggregators, Reddit threads, YouTube comparisons, and "best of" lists. If your product is absent from those sources, the agent has only your word, and engines weigh independent sources more heavily than first-party claims.

This is the same work as earning citations for any query: get into the lists buyers trust, answer questions where buyers ask them, and make sure the facts there match your site. A product that is $179 on your site and $199 on a comparison site creates doubt the agent resolves by picking something else.

Agent checkout protocols are still settling and differ by platform. Do not wait for them. The data layer is the prerequisite, and it is entirely in your control today.

A practical order of operations

1. Fetch five product pages without JavaScript and confirm price, stock, and variants appear in the raw HTML.
2. Add or repair Product and Offer schema, including availability, priceCurrency, and shippingDetails.
3. Put specs in an HTML table with consistent attribute names across the catalog.
4. Publish plain-text shipping and returns pages, linked from every product.
5. Check how agents currently describe your top ten products and fix the mismatches.

At AEO GEO Labs this is the ecommerce version of the standard visibility audit: ask the engines about your products, record what they say and which sources they used, then fix the data and the sources in that order. Share of answer for product queries is measured weekly.

Start with one product

Pick your best seller. Ask three engines to recommend a product in its category with your key constraints and see whether you appear, and whether the facts are right. For the full picture across your catalog, AEO GEO Labs runs that check as part of a visibility audit, and the free schema markup generator will get your Product markup to a sane baseline in an afternoon.