Agent readiness for marketplaces: many sellers, one standard

When thousands of sellers write their own titles, set their own shipping and handle their own returns, an AI agent comparing offers meets a different store on every page. A marketplace is only as readable as its weakest listing. Here's how to set one standard that agents can rely on, and how to check it holds across your sellers.

One question, many sellers

On a marketplace the shopper rarely asks just "find X". They ask "find X from someone I can trust, at the best total price, that arrives in time". That breaks into several jobs:

• Search and filter. "Handmade stoneware mugs under $30 with free shipping."
• Compare offers for the same item. New or used, price plus shipping, delivery date, which seller.
• Judge the seller. Where they ship from, their rating, how they handle returns.
• Read reviews of the product, and separately of the seller.
• Buy from more than one seller at once, and know what the split cart will cost.

1. Every match All listings for the search
2. Same product Identified reliably
3. Can buy In stock, ships to the shopper
4. Best total Price plus shipping, arrives in time
5. Trusted seller One the agent can vouch for

Where listings let agents down

• Uneven listings. One product listed many times under different titles, some with brand and part number, some with neither, so an agent can't tell they're the same thing.
• Sellers who are just a username. No location, returns policy or rating in the HTML, and no ordinary link to a seller page.
• Shipping revealed at checkout. Cost and delivery time per seller appear only once the cart is split, so the "cheapest" offer often isn't.
• Returns that depend on the seller, with nothing on the product page saying whose policy applies.
• Search and filters without addresses. Filters that only work through JavaScript, so an assistant can't open "mugs under $30, free shipping" directly.
• Reviews mixed together, or loaded only on scroll, so an agent can't tell product quality from seller service.

What AgentScore samples

AgentScore treats a marketplace like any store. It finds a product page through your home page's links, your sitemap or product cards, and checks its Product data for price, currency and availability, reading every offer, so a page with several sellers' offers counts. It checks that the add-to-cart button is a real button that nothing covers, and if that product is sold out, it tries up to two others.

The catch is the sample. AgentScore checks one product page, and on a marketplace that page may belong to your best seller or your worst. What your platform controls (templates, structured data, pop-ups, checkout) shows up reliably. What sellers control doesn't. Read the score as a verdict on the template, and use the sampling in the last section for the listings.

Fixes for the platform and the sellers

1. Set a minimum listing standard, and enforce it at upload

Require the fields agents compare on: brand, a product identifier where one exists, condition, the key attributes for the category (size, material, capacity) and a title that follows a pattern. Reject or flag listings that miss them at upload, rather than cleaning up later. See product identifiers.

2. Generate structured data from the platform, not the seller

Build Product data in your template from the fields you validated, with one Offer per seller. Each Offer can carry its own seller, itemCondition, shipping details and return policy. Where a page lists many offers, an AggregateOffer can sum them up with a low price, a high price and an offer count.

"offers": [{
  "@type": "Offer",
  "price": "24.00",
  "priceCurrency": "USD",
  "availability": "https://schema.org/InStock",
  "itemCondition": "https://schema.org/NewCondition",
  "seller": {
    "@type": "Organization",
    "name": "Ridgeline Pottery",
    "url": "https://fernhill.example/sellers/ridgeline-pottery"
  }
}]

See product data AI shopping agents can read and, if you send listings to shopping platforms, product feeds and merchant listings.

3. Give every seller a real page

A public page with the seller's name, where they ship from, how long they've sold with you, their rating, their returns policy and how to reach them. Link it from every offer with an ordinary link. That's how an agent answers "can I trust this seller?" without guessing.

4. Show shipping per seller on the product page

Beside each offer, say in text what shipping costs and when it should arrive, for the shopper's region or a stated default. Then an agent can compare total prices before the cart. Say which seller's returns policy applies, in a sentence the agent can quote. See shipping, returns and FAQ pages AI assistants can quote.

fernhill.example/products/stoneware-mug-12oz

Stoneware mug, 12 oz, speckled white · Brand: Ridgeline Pottery

$24.00 · New · Sold by Ridgeline Pottery · Ships from Bend, OR Seller linked

Shipping to Denver: $5.00, arrives in 3 to 5 days Shipping per offer

Returns: Ridgeline Pottery accepts returns within 30 days Whose policy applies

Add to cart

5. Give search and filters their own addresses

Make every search and filter combination a URL that loads with results in the HTML, use real controls for filters, and paginate with links rather than only infinite scroll. See site search and filters AI agents can use.

6. Keep product reviews and seller reviews apart

Show both in the HTML, clearly labelled: product reviews on the product page, with the product's rating in its structured data, and seller reviews on the seller's page. See reviews and ratings AI agents can read.

7. Explain split carts in text

When a cart holds items from several sellers, show each seller's shipping and delivery estimate as text, with a clear total. See carts AI agents can manage.

Test by sampling

One good page proves the template. It doesn't prove the catalog.

1. Run AgentScore on your home page to check the template: access, product data, the add-to-cart button, pop-ups and checkout.
2. Pick a handful of listings from different kinds of seller: a large one, a new one, one with many variants. Use View Source on each and look for the seller's name, the price, shipping, the return policy and the product's identifiers.
3. Ask an assistant a comparison question, such as "Who sells this mug on fernhill.example, and which offer is cheapest with shipping to Denver?", and check the answer against the page.
4. Set up Ghost Agent tests for a search-to-cart journey and a two-seller cart, passing when the agent reaches the payment step. The replay shows which listing or step confused it.