Which product fields Shopify Catalog actually reads
Published
The seven product fields, verbatim from Shopify
Shopify's optimizing-products-for-Catalog doc names seven product fields that AI platforms consider: title, description, images, product organization details (Type, Vendor, Collections, Tags), barcode (ISBN, UPC, GTIN, etc.), variants (including Option name), external product URL (Shopify Agentic plan stores only). The list is structural — every field corresponds to a Shopify admin surface, not an editorial choice.
The verbatim list lives on the Catalog-optimization help page1. The same doc adds a separate guidance line: store policies should be "complete and up-to-date" so AI agents reference accurate return and other policy information. That gives the full picture as: 7 structured fields + the policies layer.
§01Fields
The structured fields
Each field is a separate admin surface. Title is the products list header field. Description is the rich-text body. Images live in the media library. Product organization (Type, Vendor, Collections, Tags) sits in the sidebar. Barcode lives under Inventory. Variants live under Variants. External product URL is a metafield/admin field on Agentic-plan stores only — Shopify online-store-plan stores don't have it.
Field
Admin surface
AI purpose
Title
Product page (header)
Discovery hook + disambiguator
Description
Product page (body)
Comprehensive spec source
Images
Media library
Visual disambiguation + alt-text input
Type
Product organization
Category classifier
Vendor
Product organization
Brand identifier
Collections
Product organization
Cluster signal
Tags
Product organization
Attribute classifier
Barcode (GTIN/ISBN/UPC)
Inventory
Cross-channel identifier
Variants (incl. Option name)
Variants
Disambiguator (size/color/etc.)
External product URL
Product page (Agentic plan only)
Deep-link target for AI checkout
§02Policies
The policies layer (the secondary AI surface)
Shopify's same doc adds a separate clause: 'These policies should be kept complete and up-to-date so that AI agents can reference accurate return and other policy information about your store.' Translation: policies feed AI answers the same way product fields do. The Knowledge Base app surfaces these answers, but the AI engines also read the raw policies directly.
Four policies feed the layer: return policy, shipping policy, refund policy, and privacy policy. Each lives under Settings > Policies in the admin. The store policies article walks the audit pass — how to rewrite each policy so AI agents quote it cleanly.
§03Checklist
Per-product checklist
Open a product. Walk these ten checks. Most PDPs pass title and description; most fail on Type, Vendor, Tags, and Barcode. The barcode field in particular is left blank on the majority of imported products, breaking the cross-channel identifier.
Title is non-empty and includes one disambiguating spec phrase.
Description body is >250 words, written by the merchant (not pasted from the manufacturer).
At least one image; every image has descriptive alt text.
Type is set (not blank).
Vendor is set (not blank or "default vendor").
At least one Collection assigned.
At least three Tags assigned (attribute, not marketing).
Barcode (GTIN/ISBN/UPC) is set on at least one variant.
Variants have an Option name (size, color, etc.) — not "Default Title".
Agentic plan only: external product URL is set per product.
