Can AI shopping agents read Shopify metafields?

Sometimes, depending on how the information is exposed and which route the agent uses. A field missing from one tool response does not establish that all agents are unable to read it. Check the actual source and returned content.

Should I move specifications into product options?

Only if they genuinely describe choices the shopper selects. First inspect the existing representation and the relevant native configuration. A missing answer alone is not a reason to reorganize a working catalog.

Which product data can AI shopping agents read? Check the route

Start with a fact, not a field type. If a shopper asks which adapter matches a particular connector, the useful question is whether the assistant received the exact connector specification for the correct product. Counting metafields will not tell you that.

Trace one deciding fact: its maintained source, the page or channel that exposes it, the content returned to the assistant, and the claim in its answer. Use the answer log for the full session record; this page focuses on the path the fact took.

Separate the routes before comparing results

Scroll sideways for columns out of view. Keyboard: focus the table and use the Left and Right arrow keys.

Route	What to inspect	What the result does not establish
An AI shopping channel using Shopify Catalog	The product’s channel representation and relevant source mapping	What a visiting browser agent receives
An assistant connected to a merchant’s server-side tools	The configured endpoint, request and returned data	What an unrelated consumer app uses
An agent visiting the storefront	The visible page and the tools available to that browser session	Catalog distribution, ranking or recommendation share

Shopify’s current server-side catalog and cart route uses UCP. Its migration guide explains that those Storefront MCP tools were replaced. WebMCP tools operate in the shopper’s browser; a server response and a browser-session response are different records.

If all you have is a screenshot from a consumer assistant, record the app, date, question and cited links. Leave the retrieval route unknown until there is evidence for it. You can still check the answer’s factual content; you cannot yet diagnose why it happened.

A custom field is not automatically hidden

Shopify supports displaying metafield content in a theme. Whether a particular assistant actually encounters that content needs its own check.

For AI-channel representation, Catalog Mapping lets merchants select sources for title, description and category, including metafields and metaobject references. Its grouping and option settings are separate controls. Mapping changes can take time to process.

Those capabilities refute the blanket claim that AI cannot read metafields. They do not promise that every custom field reaches every answer. Likewise, a tool description is a capability statement, not a capture of what happened in your session.

Keep the distinctions precise:

Documented: a platform describes an available source or operation.
Observed: a saved request returned particular content in a dated configuration.
Unknown: you do not have the response, or cannot establish which source was used.

Trace one specification all the way through

Fictional illustration, not a StoreKnows test result: the manufacturer lists a 10 mm input for adapter AX-4. The store’s maintained field also says 10 mm. Its product page displays 10 mm beside AX-4, but the assistant answers that it accepts a 12 mm input.

Scroll sideways for columns out of view. Keyboard: focus the table and use the Left and Right arrow keys.

Link in the chain	What this example lets you check
Manufacturer → store field	The maintained AX-4 value agrees with the specification.
Store field → product page	The visible page shows the right value for the right model.
Page/tool → assistant	Still unknown until you inspect what the session actually received.
Received content → answer	If the returned AX-4 value was 10 mm, the answer misread it; if it was absent, inspect retrieval and source selection.

A correct value for AX-4B is still the wrong reference for AX-4. An answer without a citation may be correct, but it does not establish that your chosen field supplied it. If the maintained specification is ambiguous, repair that source first.

Correct the broken link

For a channel using the wrong source, inspect Catalog Mapping and its preview. For page reading, put the specification beside the exact product it describes. Use options for actual choices a buyer selects, and derive additional displays from the maintained source where practical.

If the assistant received the right value but contradicted it, more copies of that value may not help. Use the diagnosis guide to decide the next intervention, then the comparison method to judge what changed.

Prepare a useful StoreKnows evaluation

Bring one important question, the exact product or variant, and the record you trust. StoreKnows focuses on simulated product-answer checks and supported proposals for compatible storefront agents. Its published information uses a saved catalog copy; it does not establish changes to consumer-assistant feeds or live inventory.

StoreKnows is available on the Shopify App Store. Install StoreKnows to run the free catalog-built check. Where a supported fix is proposed, try it with your own shopper question before publishing. You can also see how the check works first.

Updated 2 October 2026: clarified that custom questions belong to a supported proposed-fix preview, rather than the free baseline, and checked Shopify’s current server-side migration guide.

Corrected 13 September 2026: the previous version generalized public-response observations into universal metafield limitations. Those claims and the unsupported causal totals have been removed. Other platform documentation above was checked on 13 September 2026; the server-side migration guide was rechecked on 2 October 2026.

StoreKnows is independently developed and not affiliated with, endorsed by or sponsored by Shopify, OpenAI, Google or Anthropic.