← Guides

A product feed for AI agents: why and how

Updated 2026-08-03

What a product feed is

A product feed is a machine-readable file that holds your whole catalog: one record per product, and inside each record the price, availability, link and image. The format was invented for ad platforms — Google Shopping takes XML, local price aggregators across Moldova and Romania take YML. But the feed now has a new reader: the AI agent that picks a product and places an order on the shopper's behalf.

The key difference from a normal page is that a feed needs no parsing. The agent does not fight through layout, banners and pop-ups — it gets a clean table of products and sees price, currency and availability straight away. Where a model guesses from an HTML page, it simply reads from a feed.

Why an AI agent needs a structured catalog

AI assistants increasingly answer "find and buy me..." not with a link to search results but with a specific product and price. To land in that answer, your store has to serve data the agent can read without guessing. A feed gives three things a normal page does not:

1. Speed and accuracy. The agent pulls price and availability from the feed in one request instead of crawling dozens of pages.
2. Fresh pricing. The feed updates on a schedule, so the figure in the AI's answer matches the price in the cart. Drift is a common reason an assistant quotes an outdated price.
3. A complete catalog. The agent sees the whole range at once, not just the pages a crawler managed to reach.

The demand is already here: our AI Traffic Index counted about 1,005,146 bot hits over two weeks on a three-store panel, and the single most-read product drew 73,715 requests. Agents reach the catalog — the question is whether they can assemble an order from it.

Key fields

The minimum set, without which a record is useless:

Field	What it means	Example
id	a stable product identifier	SKU-10482
title	product name	Delonghi KG79 coffee grinder
price + currency	price and currency (ISO 4217 code)	499 MDL
availability	stock status	in_stock / out_of_stock
link	link to the product page	https://shop.md/p/10482
image	link to the image	https://shop.md/img/10482.jpg

Two fields break most often. availability has to be honest: if the item is gone, the agent must not offer it for purchase. And image is worthless when it returns 403 or 406 to bots — the number-one problem in our rating of 63 stores across Moldova and Romania, where the median score is just 38 out of 100.

Feed formats

Google Shopping (XML)

Classic XML with fields in the g: namespace. Ad systems understand it, and so do many AI crawlers:

<item>
  <g:id>SKU-10482</g:id>
  <g:title>Delonghi KG79 coffee grinder</g:title>
  <g:price>499 MDL</g:price>
  <g:availability>in_stock</g:availability>
  <g:link>https://shop.md/p/10482</g:link>
  <g:image_link>https://shop.md/img/10482.jpg</g:image_link>
</item>

YML

The format of local price aggregators. Also XML, but with its own schema (offer, price, currencyId).

agentic.json

A newer JSON format built for agentic commerce: the same catalog, but in a shape convenient for an agent to read and to check out from:

{
  "id": "SKU-10482",
  "title": "Delonghi KG79 coffee grinder",
  "price": { "amount": 499, "currency": "MDL" },
  "availability": "in_stock",
  "link": "https://shop.md/p/10482",
  "image": "https://shop.md/img/10482.jpg"
}

How to serve the feed

• Publish the feed at a stable URL and keep it open to AI bots.
• Refresh it on a schedule so price and stock never trail the site.
• Never block Googlebot or Bingbot — that kills SEO; filter AI bots deliberately instead.
• Check that the links in image open for bots rather than returning 403/406.

How Botmetria turns a feed into a UCP catalog

The WooCommerce plugin and Botmetria SDK take your existing catalog — a feed or JSON-LD Product markup — and convert it into a UCP catalog: a single structured format agents read directly and check out from. Price, currency and availability are served machine-readable instead of buried in layout. If no feed exists yet, the plugin builds one from your store data itself.

That closes two audit categories at once: Product data (30% weight) and Agentic Markdown (20% weight) — together half of the AI-readiness score.

How to check your feed

• Open the feed in a browser — every product is present and the encoding is not garbled.
• Verify the required fields: id, price, currency, availability, link, image.
• Run the free Botmetria audit: the Product data category shows whether AI can read your price and stock. Score bands: >70 green, 41–70 yellow, ≤40 red.
• Make sure the links in link and image open without 403/406.

FAQ

Related guides

• How an AI agent buys from your store
• UCP, ACP, MCP in plain words: a guide to agentic-commerce protocols