Agent readiness for fashion retailers: sizes, fit and returns

A jacket isn't one product to an AI agent. It's a style in a dozen sizes and colors, each with its own stock, and the shopper's real question is whether it will fit. Here's where fashion sites lose agents, and the eight changes that bring them back.

The shopper's question has five parts

Nobody asks an assistant for "a jacket". They ask for "a waterproof jacket in navy, size medium, under $200, that I can return for free if it doesn't fit". The agent has to find each part of that on your site, and trust it enough to repeat it:

• The style: color, material, fit and occasion, often across several stores at once.
• Their size, in stock: not the style in general, but medium in navy.
• The right size: from your size chart and fit notes, and sometimes from what the shopper wears in another brand.
• The returns: how long, who pays, and whether sale items count. In clothing, this often decides the sale.
• The right variant in the bag: then on to checkout, where the agent hands back to the shopper to pay.

• Find the style Color, material, fit, price
• Check my size In stock, in my color
• Will it fit? Size chart and fit notes
• Can I return it? Window, cost, exceptions

What gets in the way on fashion sites

Fashion sites are built to look good, and much of what makes them look good hides information from agents:

• Swatches and size boxes with no names. A row of colored circles, the chosen one marked only by a border, and a sold-out size that's merely faded.
• Stock that appears only after a click. "Only 2 left" shows up once a size is picked, so an agent reading the page sees no stock at all.
• Size guides as pictures or drawers. A chart saved as an image, or loaded by JavaScript when someone presses "Size guide", is invisible to an assistant reading the HTML.
• Marketing color names only. An agent asked for "grey" needs to be told that Fog is a light grey.
• Lookbooks with no prices or links. Editorial pages full of outfits where nothing says what each piece is or where to buy it.
• Sign-up pop-ups on arrival. In our study of 255 stores, 38% showed a pop-up or banner that covers much of the page on arrival, and on 24% of stores it had no clearly labelled button to close it.
• Express buttons only. A product page whose only button is "Buy now" or a wallet, which an agent shouldn't press without the shopper.

What a scan shows, and what it can't

AgentScore finds a product page the way an agent would: links in your home page's HTML, then your sitemap, then product cards on the home page. Four of its behaviors matter most for clothing:

Situation	What AgentScore does
Variants described as a ProductGroup with hasVariant	Reads each variant's offers when checking for price, currency and availability
The product it opened is sold out	Tries up to two other product pages. If they're all sold out, the add-to-cart check is skipped, not failed
"Add to bag" is disabled until a size is chosen	Passes the add-to-cart check if the button is real, visible and not covered, and notes that it's disabled until options are chosen
The checkout form only appears with an item in the bag	Presses add-to-cart once to see it. If the button is disabled until a size is chosen, it can't, so the checkout field and guest checkout checks are skipped with that reason

That last row is common in fashion, and nothing is marked down for it. It does mean the scan can't see your checkout form, so test that journey with an agent instead. The pop-up check runs on your home page. Reading your AgentScore report explains every check.

Eight fixes, most urgent first

These are ordered by how often they stop an agent outright, then by how much of the shopper's question they answer.

1. Make the size and color picker readable

If an agent can't choose "navy" and "medium", nothing else matters. Use named controls (native radio buttons are simplest), state the selected option in the code, and say "sold out" in words. Give each variant its own URL so an assistant can link straight to it. See size, color and variant pickers AI agents can use.

2. Put stock per size in the HTML and your product data

Describe the style as a ProductGroup whose variants each carry their own size, color, price and availability, as in Google's product variant guidance. Take availability from the same source as the button, so the data never says InStock for a size you can't sell.

{
  "@context": "https://schema.org",
  "@type": "ProductGroup",
  "name": "Harbor rain jacket",
  "productGroupID": "HRJ-01",
  "variesBy": ["https://schema.org/size", "https://schema.org/color"],
  "hasVariant": [{
    "@type": "Product",
    "sku": "HRJ-01-NVY-M",
    "color": "Navy",
    "size": "M",
    "offers": {
      "@type": "Offer",
      "price": "179.00",
      "priceCurrency": "USD",
      "availability": "https://schema.org/InStock"
    }
  }]
}

The rest of the Product fields are in product data AI shopping agents can read.

3. Publish your size guide as text

Turn size charts into HTML tables, with measurements in the units your customers use and the size system you follow (US, UK, EU). Put the chart on the product page, or link to it with an ordinary link rather than only a button that opens a drawer. Add fit notes in plain sentences: "Runs small. If you're between sizes, size up." Schema.org's size property can point to a SizeSpecification with the size system, but the visible table is what assistants quote. Honest fit notes cut returns too.

4. Make the returns policy quotable

Write the window, the cost and the exceptions in sentences that stand alone: "Return unworn items within 30 days for free. Final sale items can't be returned." Add a MerchantReturnPolicy to your Organization data, and override it on products with different terms, as Google's return policy documentation describes. See shipping, returns and FAQ pages AI assistants can quote.

5. Get the sign-up pop-up out of the way

Wait until a shopper has scrolled or opened a second page, keep it off the "Add to bag" button, and give it a real close button labelled "Close". See cookie banners and pop-ups that trap AI agents.

6. Name colors and describe photos

Pair each color name with a plain description ("Fog, a light grey"). Write alt text that says what matters for buying: color, cut, length, and the model's height and size where you have it. See alt text for AI agents.

7. Make lookbooks shoppable

Under each look, list every piece with its name, price and an ordinary link to the product. Then an assistant asked to "buy the outfit from the autumn campaign" can.

8. Keep "Add to bag" next to the express buttons

Wallet and "Buy now" buttons are great for people. Agents need an ordinary "Add to bag" too, and a checkout that doesn't force an account. See guest checkout.

Check it yourself

1. Run AgentScore and read the product data, add-to-cart and pop-up checks first.
2. Ask an assistant a real question, such as "Is the Harbor rain jacket in navy medium in stock at halyard.example, and can I return it free?", and compare the answer with the page.
3. Give a Ghost Agent the goal "Add the navy Harbor rain jacket in medium to the bag and go to checkout", and pass it when it reaches the payment step. It stops there, and the replay shows where an agent picked the wrong size or couldn't close a pop-up.

A fashion product page an agent can shop

• Picker named sizes and colors, selection and sold out stated in the code
• Stock per size, in the HTML and in ProductGroup data
• Size guide a text table with fit notes
• Returns window, cost and exceptions in plain sentences
• Pop-up covering "Add to bag" on arrival
• Buttons "Buy now" with no "Add to bag"

Repeat the test after every theme update and before every sale.