AI shopping feeds

Shoppers increasingly start in an AI assistant or in Google instead of on your site. AI Shopping Feeds publishes your live catalog in the formats those surfaces read, so your products, prices and stock are described the way your store has them. Feeds are included in the Pro and Enterprise plans.

Where it lives

Open Settings → AI Shopping Feeds in the dashboard sidebar, or go to /app/<storeId>/settings/agentic-feeds. Feeds are off until you turn them on. While they are off, or if the store’s plan no longer includes them, every feed URL returns “not found”.

The two feeds

Google Merchant Center - product data as an RSS 2.0 file with Google’s product attributes. Copy the URL from the page and add it in Merchant Center as a scheduled fetch. Merchant Center fetches on its own schedule, daily by default.
ChatGPT shopping (OpenAI product feed) - a gzip JSON Lines file in OpenAI’s product feed format. This file is the source for a separate destination submission workflow; turning feeds on does not register or approve your merchant account. Follow OpenAI’s current onboarding and upload requirements for the destination available to your store. Shoppers still buy on your own checkout.

Each feed has one item per active variant with title, description, price, sale price, availability, the product URL on your domain, image, brand, and GTIN when one is known. The page shows when each feed was last built and fetched.

What is left out

Products without an image, a product page URL, or a price above zero are skipped. Products with no brand or vendor use your store name as the brand. If your store does not track stock, every product is listed as in stock. Confirm that this fallback matches actual sellability and backorder policy. Feeds exclude product costs and numeric stock counts. Catalog sync, feed rebuild/cache refresh, and the destination’s next fetch are separate stages; a catalog change has no guaranteed end-to-end publication time.

Step by step

Make sure the catalog is synced and products have images, prices, and product URLs. See Catalog.
Click Turn feeds on. The first request to each URL builds the file.
Open each feed URL in a browser and spot-check a few products against your storefront: price, sale price, availability, and link.
Add the Google URL in Merchant Center, then check Merchant Center’s diagnostics after its first fetch.
If your store is eligible for an OpenAI destination, use the generated file in its approved submission process and inspect the destination validation result.

Validate a sample and diagnose gaps

The following is an illustrative Google item fragment, with synthetic values. Your generated feed supplies the RSS document and Google namespace. The price includes its currency; the link, image and stock state must agree with the public product page.

<item>
  <g:id>shell-blue-m</g:id>
  <g:title>Rain shell - Blue / M</g:title>
  <g:description>Lightweight rain shell with taped seams.</g:description>
  <g:link>https://store.example/products/rain-shell</g:link>
  <g:image_link>https://store.example/images/rain-shell.jpg</g:image_link>
  <g:price>79.00 USD</g:price>
  <g:availability>in_stock</g:availability>
  <g:brand>Example Outdoors</g:brand>
</item>

If an item is missing, check active product status, variant price above zero, image, and product URL before resubmitting. A “last fetched” timestamp proves that the file was requested, not that a destination approved every item. Record the catalog sync time, feed build time, fetch time, and destination validation state separately. For an incorrect price, correct the source, verify sync and the rebuilt feed, then inspect destination diagnostics after it fetches again. Do not invent a GTIN to clear an error.

Format references reviewed 21 October 2026: Google product data specification and OpenAI product upload specification. Destination requirements can change; generation is separate from account eligibility, validation, approval and actual display.

Other ways agents find your store

Store llms.txt - a plain-text index of your store, categories and products for AI assistants. The Zubby WooCommerce plugin serves it from your own domain at /llms.txt and /.well-known/agents.md. It lists the feed URLs while feeds are on.
MCP endpoint - on plans that include the public API (Pro and above), your store has a read-only Model Context Protocol endpoint that AI agents can search live for products, product details, store information and published policies. It never places orders. Details are in the API reference.

On Shopify, Shopify’s own agentic storefronts setting (Shopify admin, Sales channels) already lists eligible stores with AI assistants. The feeds here are still useful when you want a Google Merchant Center feed you control.

Feeds help supported services read your catalog. They do not guarantee listings, rankings or AI citations. Google does not require llms.txt for its AI search features; keep your public pages useful, crawlable and up to date.

Settings overview
Custom storefronts - push catalog data through the API.