← The journalField note·September 15, 2026·9 min read

How shopping agents actually get your Shopify product data

Agents read Shopify Catalog, Merchant Center attributes and vendor feeds, not your theme HTML. Your structured fields decide what they can say about you.

By Rank Sniper

Agents do not read your theme

The most expensive misconception in Shopify SEO right now is that a shopping agent visits your product page, renders your theme, and reads the copy.

It does not work that way. Shopify's Help Center states the mechanism directly: "Shopify Catalog is the primary method for agentic storefronts to receive your product data."

That single sentence relocates the entire optimisation problem. Your hero image, your section layout, your carefully sequenced product story, your review widget: none of that is the interface an agent consumes. The interface is a structured record.

Shopify describes what that record contains. Products are listed with "their title, description, options, images, price, availability, and other key attributes, all structured in a way that AI agents can parse and understand."

Title. Description. Options. Images. Price. Availability. That list is the surface area. If a fact about your product is not in one of those fields, an agent working from the catalogue does not have it.

What is already switched on

Two facts from the same Help Center documentation that merchants are frequently unaware of.

Agentic storefronts reach "AI channels, such as ChatGPT, Google AI Mode and Gemini, Microsoft Copilot, and Meta."

And they are active by default for eligible stores.

So the question for most merchants is not whether to participate. It is whether the data already being distributed represents the products accurately.

Controls exist. Shopify documents a per-channel opt-out under Sales channels, then Agentic. That lets you remove a specific AI channel.

Full suppression is a heavier instrument. Shopify states that removing products entirely requires marking them Unlisted, which also removes them from search engines. Read that consequence carefully before using it. You cannot exit the agent surface for a product without also exiting organic search for the same product.

Two catalogue endpoints, two jobs

Shopify's agent documentation frames the platform side as "Build unified agentic experiences that securely act on behalf of buyers with the Universal Commerce Protocol (UCP) and Shopify's UCP-compliant MCP servers."

There are two catalogue MCP servers, and the difference matters for understanding when your store is visible.

Storefront Catalog MCP

Scoped to one merchant. The documentation says to "Use Storefront Catalog MCP whenever buyer intent and query are scoped to a single merchant."

It returns "products with title, description, price range (minor units), media, and variants."

Note what is in that return shape. Price is expressed in minor units, so currency precision is handled numerically rather than as display text. Variants are returned, which means variant-level data is part of what the agent sees. Media is returned as media, not as rendered page imagery.

The tools exposed are search_catalog, lookup_catalog and get_product. Those are the three shapes of agent behaviour: find by query, resolve a known reference, fetch full detail.

Access has a gate. "The /api/ucp/mcp endpoint requires an agent profile — every request must include a meta.ucp-agent.profile URL."

Global Catalog MCP

Scoped to the whole ecosystem. The documentation describes it as a way to "Search and discover products across the entire Shopify ecosystem."

Its search is broader in input type as well as scope: "search_catalog supports a text query, an image, or a set of product IDs to find similar items."

Image-based and similarity-based retrieval are worth pausing on. If an agent can search by image and by similar-product ID, then your product imagery and your variant structure participate in discovery, not just your text.

Shopify has described Shopify Catalog as comprising "billions of products". Discovery at that scale is a retrieval problem, and retrieval works on fields.

The approval requirement is gone

In its Spring '26 developer edition, Shopify wrote: "Building on Shopify's agentic commerce layer used to require approval. That requirement is gone." And: "Developers don't apply for UCP access, they register their agent profile in the Developer Dashboard and call the public MCP endpoint."

For a merchant, the implication is about the population of agents, not about anything you configure. The set of parties that can query catalogue data is now defined by registration rather than by approval. Your product data is addressable by any registered agent profile.

That is an argument for treating catalogue fields as public-facing copy with the same care you give a product page, because that is now what they are.

Metafields: Catalog Mapping

The default field set is the default field set. Shopify documents Catalog Mapping as the mechanism that exposes metafields and metaobjects to the catalogue.

This is the lever for merchants whose differentiating data lives outside the standard fields. Material composition, compatibility lists, certifications, dimensions, ingredient lists, care instructions: on most Shopify stores these live in metafields or in theme sections, and only the metafield route is documented as reachable by the catalogue.

If a specification is currently rendered by a theme block and stored nowhere structured, an agent working from the catalogue cannot cite it. Moving it into a metafield and mapping it is the documented path.

Google's lane: UCP, Universal Cart, and one attribute

Google's agentic commerce announcement in January 2026 described UCP as "a new open standard for agentic commerce that works across the entire shopping journey", and said it was "co-developed with industry leaders including Shopify, Etsy, Wayfair, Target and Walmart."

The UCP specification itself is published at ucp.dev under Apache 2.0, described as "The common language for platforms, agents, and businesses."

In May 2026 Google introduced Universal Cart as "an intelligent shopping cart and your new hub for shopping on Google."

The merchant-relevant mechanics are in Merchant Center documentation, and they are specific. "You remain the Merchant of Record." And: "Only product listings using the native_commerce(checkout_eligibility) product attribute will display the 'Buy' button."

Two things follow.

First, merchant of record status stays with you. This is not a marketplace arrangement where the platform sells your goods.

Second, transactability on Google's agentic surface is controlled by a named attribute on a listing. Not by page quality, not by schema on your product template, not by content. A field.

OpenAI's lane: nine required fields

OpenAI's commerce documentation takes the feed route. Merchants "provide a secure, regularly refreshed feed (CSV or JSON) containing key details".

The required discovery fields are nine:

  • item_id
  • title
  • description
  • url
  • brand
  • seller_name
  • image_url
  • availability
  • price

That is the complete discovery contract. Nine fields decide whether a product is discoverable in that channel.

Two defaults are documented and they differ: "Search defaults to enabled and checkout to disabled." Discovery is opt-out. Transaction is opt-in.

On responsibility: "OpenAI is not the merchant of record in the Agentic Commerce Protocol. Merchants are expected to bring their own PSP."

The Agentic Commerce Protocol is published at agenticcommerce.dev under Apache 2.0, with a spec directory dated 17 April 2026. That site credits Stripe and OpenAI; Stripe's own ACP documentation, which also records Meta's involvement, is the source for the wider attribution.

So there are two open protocols in the market with overlapping goals and different authorship. UCP, co-developed with Shopify and Google among others. ACP, created by Stripe, OpenAI and Meta. Both Apache 2.0. Both structured around feeds and field contracts rather than page crawling.

The comparison that makes the lesson obvious

Line up what the three documented interfaces consume.

  • Shopify Catalog: title, description, options, images, price, availability, plus mapped metafields.
  • Storefront Catalog MCP return: title, description, price range in minor units, media, variants.
  • OpenAI discovery feed: item_id, title, description, url, brand, seller_name, image_url, availability, price.
  • Google Merchant Center agentic checkout: the native_commerce(checkout_eligibility) attribute.

Every one of these is a field list. Not one of them is your theme.

Title and description appear in all three catalogue interfaces. Availability and price appear in all three. Images or media appear in all three. Brand and seller name are required by OpenAI and are frequently blank or auto-filled on Shopify stores that never set a vendor value.

What to check in your own admin

Concrete, in the order that usually pays.

Descriptions written as product data, not as page copy. An agent receives the description field as text. If your description assumes surrounding page context, or consists mainly of marketing framing, the agent has less to work with than a competitor whose description states materials, dimensions, compatibility and use case in plain sentences.

Vendor and brand values. OpenAI's feed requires brand and seller_name. Check whether your Shopify vendor field is set deliberately across the catalogue or left at an import default.

Availability accuracy. Availability is in every field list above. Stale availability is not a cosmetic error in an agent context. It is the difference between an agent recommending a product and an agent recommending a competitor's.

Options and variants. Storefront Catalog MCP returns variants. If your options are named in ways that only make sense next to a swatch image, the structured record is weaker than the page.

Images as data. Global Catalog MCP supports image-based search. Product photography is retrieval input, not only merchandising.

Metafields and Catalog Mapping. Inventory your specification data. Anything meaningful that lives only in a theme section is invisible to the documented catalogue path.

Your channel settings. Open Sales channels, then Agentic, and look at which channels are on. Agentic storefronts are active by default for eligible stores, so this is a review rather than a setup task.

The honest limit

None of the documentation in this article promises that complete fields produce citations, recommendations or sales. Shopify documents what the catalogue contains and how agents query it. Google documents which attribute produces a Buy button. OpenAI documents which nine fields are required for discovery.

What all of it establishes is a floor. Incomplete or inaccurate structured fields constrain what an agent can say about your products, because those fields are the only thing it receives.

That is a narrower claim than the market is making, and it is the one the documentation supports.

Related reading

Sources

Run a free scan →Get the next dispatch →
Keep reading
Field note·September 15, 2026

What Google actually says about being surfaced in AI Overviews and AI Mode

Google documents one entry condition for generative AI features, no special markup, and three separate control planes. Here is what that means for a Shopify store.

Read →Field note·September 15, 2026

What can and cannot be measured about AI citations in 2026

Google reports impressions, Bing reports citations, Cloudflare reports crawler hits. Position within an answer and agent-completed purchases are not reported.

Read →
How shopping agents actually get your Shopify product data — Rank Sniper