Shopify metafields for AI discoverability

Metafields are Shopify's typed extension points — structured data attached to products, variants, collections, customers, orders, and more, with explicit types (single-line text, multi-line text, integer, boolean, JSON, date, measurement, rich text, reference, and so on). They are the right home for facts about a product that are specific, queryable, and reusable. That description fits most of what AI retrievers actually want: material, country of origin, fit, warranty length, compatibility, certification bodies, ingredient lists. Stuffing those facts into product description HTML is the most common alternative, and it is noticeably worse for AI retrieval because unstructured prose mixes facts and marketing and loses type information.

Product tags

Flat, untyped, limited to 250 per product. Useful for navigation and filtering. Poor for AI retrieval — no key, no type, no structured meaning.

Product description HTML

Unstructured prose. Readable by humans and models. Works for narrative content — loses badly to metafields for facts that a retriever wants to quote precisely.

Variant title / options

Structured for variant attributes but limited in field count. Great for color / size / material when those ARE the variant axis. Insufficient for cross-variant specs.

Metafields

Typed, named, scoped, and queryable. Can be exposed in theme, storefront API, schema JSON-LD, and merchant feeds. The right home for AI-relevant product facts.

Below is the metafield stack we recommend to every Shopify merchant starting on GEO. The exact twelve will not map one-to-one to every category, but the structure applies broadly — fit and care for apparel, ingredients and certifications for beauty, compatibility and specs for electronics. Use namespace 'ai' so the fields are clearly scoped and easy to filter in theme Liquid.

ai.materials

multi_line_text — materials or ingredients list. Feeds both Product schema (material / ingredient) and human-readable spec blocks.

ai.fit_notes

multi_line_text — sizing, fit, wrist-size compatibility. One of the most-quoted fields in Perplexity and ChatGPT responses.

ai.care_instructions

multi_line_text — washing, storage, or maintenance. Also the seed content for HowTo schema on products with procedures.

ai.warranty_length

single_line_text — e.g. 'Lifetime on stones, 10 years on movement'. Highly quotable.

ai.country_of_origin

single_line_text — 'Made in Italy', 'Assembled in the USA'. Feeds Product.countryOfOrigin.

ai.compatibility

multi_line_text — device, size, or accessory compatibility. Critical for electronics, gear, and accessories.

ai.certifications

list.single_line_text — 'FDA registered', 'GOTS certified', 'Fairtrade'. Feeds Product.additionalProperty or Organization.hasCredential.

ai.shipping_time

single_line_text — e.g. '2-3 business days in US, 5-10 international'. Quoted constantly on transactional prompts.

ai.return_policy_summary

single_line_text — one-sentence summary of the returns policy. Prevents retrievers from inferring stale answers from the policy page alone.

ai.faq

JSON — array of { question, answer } pairs. Renders to FAQPage schema and to an on-page accordion.

ai.howto

JSON — HowTo procedure with steps, tools, supplies. Renders to HowTo schema and to a visible step list.

ai.key_specs

JSON — array of { key, value } spec pairs. Renders to Product.additionalProperty and to a visible spec table.

A metafield sitting in Shopify admin, not rendered anywhere, is invisible to every AI retriever. Metafield value reaches retrievers through exactly three surfaces, and the right rendering strategy is to use all three because different engines read each one.

Surface 1: visible HTML

{% assign fit = product.metafields.ai.fit_notes %}
{% if fit %}
  <section class="ai-fit" aria-label="Fit and sizing">
    <h3>Fit and sizing</h3>
    <p>{{ fit | newline_to_br }}</p>
  </section>
{% endif %}

Surface 2: JSON-LD schema

{% assign specs = product.metafields.ai.key_specs.value %}
<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "Product",
  "name": {{ product.title | json }},
  "description": {{ product.description | strip_html | json }},
  "brand": { "@type": "Brand", "name": {{ product.vendor | json }} },
  "additionalProperty": [
    {% for s in specs %}
      {
        "@type": "PropertyValue",
        "name": {{ s.key | json }},
        "value": {{ s.value | json }}
      }{% unless forloop.last %},{% endunless %}
    {% endfor %}
  ]
}
</script>

Surface 3: merchant feeds

Google Merchant Center accepts additional attributes via custom feed columns; Shopify's Google & YouTube app maps metafields to Merchant Center fields through the app's attribute mapping interface. The Agentic Commerce Protocol feed similarly supports metafield-backed extended attributes. Map the ai.* metafields to the corresponding feed columns so the same data reaches ChatGPT Shopping, Google Shopping, and Bing Shopping without re-authoring.

JSON metafields are the most powerful and most misused of Shopify's types. They let you store rich structured content — FAQ pairs, HowTo steps, attribute tables — without custom app development. The catch is that unlike string or integer metafields, JSON has no built-in validation, so a typo in an admin session can break every page that renders the field. The fix is to define a strict JSON schema for each metafield, validate on save (via a small Shopify App or a theme-level check), and render defensively in Liquid.

ai.faq structure

[
  {
    "question": "Is this watch waterproof?",
    "answer": "Yes, to 5 ATM. Safe for rain, handwashing, and brief pool contact. Not rated for showering or diving."
  },
  {
    "question": "What wrist size does it fit?",
    "answer": "The bracelet fits wrists from 160mm to 210mm. Additional links can be added by a jeweller for wrists up to 240mm."
  }
]

ai.key_specs structure

[
  { "key": "Case size", "value": "42mm" },
  { "key": "Case thickness", "value": "11.5mm" },
  { "key": "Water resistance", "value": "5 ATM" },
  { "key": "Movement", "value": "Swiss quartz" }
]

Metaobjects are Shopify's reusable structured entities — think of them as custom content types you can reference from many products. When the same data appears across many products (a size guide shared by all apparel, an ingredient profile used by every serum, a certification shared by a brand line), a metaobject is the cleaner pattern. A single metaobject definition lets you edit the entity once and have the update propagate to every product that references it.

• Size guides — one size_guide metaobject per style family, referenced from 30 product SKUs. Update the sizing once; all 30 products reflect it.
• Certifications — organic, fair trade, cruelty-free, FDA-registered. Each cert is a metaobject with a name, body, and logo URL. Products reference the ones they hold.
• Sustainability claims — shared claims with sourced citations. Central update keeps the claim consistent and well-cited across the catalog.
• Designer or author profiles — one author per profile, referenced from every piece they designed. Feeds Organization and Person schema.

Most established Shopify stores we work with have years of facts embedded in unstructured product description HTML. Migrating them into metafields looks daunting but breaks down into a repeatable three-phase sequence that a small team can complete over a quarter for most catalog sizes.

1. 1Phase 1: Define the metafield stack in admin. Namespace them under 'ai' (or your chosen name). Document the type and sample value for each. 2-3 hours of work that only needs to happen once.
2. 2Phase 2: Seed values from existing content. For each product, extract the facts from the current description into the corresponding metafields. Doable manually for small catalogs; for 400+ products, use the Shopify Admin API or an app that can parse existing descriptions and write metafields.
3. 3Phase 3: Refactor theme to render from metafields. Add Liquid partials for each field type — fit, materials, specs, FAQ, HowTo. Render in visible HTML and in schema simultaneously. Eventually, prune the now-redundant content from descriptions so there is no double-source of truth.