Blog›E-commerce

How Should You Structure Shopify Metaobjects So AI Shopping Assistants Can Read Them?

September 2026· 8 min read

Shopify

Metaobjects

Readable by AI

Key Takeaways

Put every answer a shopper might ask for (a measurement, a wash temperature, a spec) in its own named field on a metaobject. Don't bury it in rich text or in an image of a table.
Name the definition, its fields and its entries in the words shoppers actually use. "Washing" beats wsh_tmp. AI assistants match questions to your data partly by those names.
Keep the answer inside the entry. A reference to another metaobject is a pointer, and a reader that doesn't follow pointers sees an ID, not your size chart.
Fewer, broader entries (one size chart per product family) beat one entry per product.
Shopify now sends product data to AI channels through Shopify Catalog, and its own help centre says Catalog Mapping is "most helpful" for stores that keep custom data in metafields and metaobjects. Structure you add now is read by more than one assistant.

If you want AI shopping assistants to answer size, care and spec questions correctly, store those answers as metaobjects with one plain, well-named field per fact. Give each entry a clear display name, keep the values short, and link the entry to the products it covers. That structure is readable by your own on-site assistant, by Shopify's catalogue feeds to AI channels, and by your theme, all at once.

Why does structure matter to an AI shopping assistant?
Metafield, metaobject, page or image: which should hold a size chart?
How should you name definitions and fields?
Why should the answer live inside the entry?
What does a readable size chart look like?
How should care instructions and spec tables be modelled?
Do third-party AI agents read your metaobjects?
A checklist before you publish
Frequently Asked Questions
References

Why does structure matter to an AI shopping assistant?

An AI assistant answers "will a medium fit a 96 cm chest?" by finding a fact and quoting it. If the fact is a labelled number, the answer is exact. If it is a sentence in a product description, or pixels in a JPEG, the assistant has to guess, and a guessed size is a return waiting to happen.

The stakes are rising. Shopify reports that AI-referred orders grew nearly 13 times year over year, and that those orders convert at nearly 50% higher rates than organic search [1]. Closer to home, Australia Post's 2026 eCommerce report says six in ten Australians already use AI, and expects AI to influence up to 30% of ecommerce transactions by 2030 [2].

Shopify's own guidance on agent-ready data is blunt: agents "will often interpret literally" [1]. They read what is written, in the field it is written in. Structure is how you make sure what is written is the answer.

Metafield, metaobject, page or image: which should hold a size chart?

Shopify gives you four places to put a size chart, and they are not equally readable.

Where the chart lives	Readable by an AI assistant?	Best for
An image of a table	No. It is pixels, not text	Nothing a shopper will ask about
A page (rich text) linked from the product	Partly. The text is there but the rows and columns are lost	Long-form fit advice
A metafield on each product	Yes, one value at a time	A single fact per product, like a fabric weight
A metaobject referenced by the product	Yes, as named fields you can reuse	Charts, care guides and spec tables shared across products

Shopify's own pop-up size chart tutorial links each product to a rich-text page [3]. It looks fine to a person, but it is the least machine-readable option that still contains text. Metaobjects exist for exactly this job: Shopify describes them as a way to "create entirely new types of structured data that can be referenced and reused" [4]. Metafields attach a single value to something that already exists [4].

The practical rule: if the same table applies to many products, it is a metaobject. If it is one number about one product, it is a metafield.

How should you name definitions and fields?

Name them the way a shopper would ask the question. This matters more than it looks.

An assistant receiving "how do I wash this?" has to decide which of your tables is relevant before it reads any values. Our own assistant, QuackChat, reads each definition's type, name, description and every field's key and name, and matches the shopper's words against them. A definition called care_instructions with a field called washing is found by "how do I wash this?". The same data in a field called wsh_tmp is not.

Good naming habits:

Definition names are nouns a shopper would use: "Size guide", "Care instructions", "Tech specs".
Field names are the question's key word: "Washing", "Drying", "Chest (cm)", "Battery life (hours)".
Entry display names say what they cover: "Women's tops size guide", not "SG-03".
Use the description field. One sentence such as "Measurements for all women's tops and dresses" helps both people on your team and any reader deciding whether this table is relevant.

Field labels also become the words a shopper sees in the answer, so "Washing" reads naturally where washing_temp_c_max would not.

Why should the answer live inside the entry?

A metaobject field can hold a value, or a reference to another object. References are powerful in a theme, but a reference is only an ID until something follows it. QuackChat reads the entries of a matched definition and uses each field's value as written. It does not fetch the objects a reference points to, so a size chart whose rows are separate referenced objects would reach it as a list of IDs.

We can't tell you how every other AI agent handles references, so the safe design works for any reader:

Put the facts a shopper asks about directly in the entry's own fields.
Use references for relationships (which products use this chart), not for the answer itself.
Keep each value short. A few hundred characters per field is plenty; a whole page of text in one field is a sign it should be split.
Prefer plain text and number fields over json for anything a shopper will ask about. There is no table field type in Shopify [5], so don't try to fake one inside a JSON blob.

Scale matters too. An assistant answering a live question reads a handful of entries, not your whole catalogue of them. QuackChat reads up to ten entries of a matched type, so eight size guides, one per product family, are all seen, while four hundred per-product guides are not.

What does a readable size chart look like?

Here is a size chart that a person, a theme and an AI assistant can all read.

Definition: Size guide. Description: "Body measurements for each size, by product family."

Field name	Type	Example value
Product family	Single line text	Women's tops and dresses
Unit	Single line text	cm
Measurements	Multi-line text	XS: chest 80-84, waist 62-66, hip 86-90 (one line per size)
Fit note	Multi-line text	Relaxed fit. Between sizes, size down.
How to measure	Multi-line text	Measure around the fullest part of the chest, keeping the tape level.

Then give each product a metafield that references the right size guide entry, so the theme can show it and the product and chart stay linked.

Writing one size per line keeps the whole chart inside one entry, which is what makes it readable by a reader that doesn't follow references. If your theme needs a fully structured grid, keep that as well, and keep the plain-text version alongside it.

How should care instructions and spec tables be modelled?

Care instructions repeat across many products, so they are a natural metaobject. Use one field per question a shopper asks:

Washing (single line): "Cold machine wash, 30°C max"
Drying (single line): "Line dry in shade. Do not tumble dry"
Ironing (single line): "Low heat, inside out"
Notes (multi-line): anything else, such as "Wash dark colours separately for the first three washes"

Spec tables need a decision first. Shopify's category metafields (the product taxonomy attributes, such as neckline or sleeve length for shirts) are designed to help discovery "on your site, on marketplaces, and on search engines" [6]. Use those first for anything the taxonomy already covers. Add custom metafields for single facts it doesn't, and reach for a metaobject only when a block of specs repeats across products, such as a battery model shared by five devices.

Do third-party AI agents read your metaobjects?

Partly, and the honest answer is that it depends on the channel.

Shopify's help centre says eligible products become discoverable to AI channels through Shopify Catalog, carrying title, description, options, images, price, availability "and other key attributes". It adds that Catalog Mapping "is most helpful for stores with custom data and grouping logic", naming metafields and metaobjects [7]. Shopify's Spring '26 Edition also expanded the Catalog API with richer size, colour and delivery data for AI agents [8].

What Shopify does not say is that an unmapped metaobject is ignored, or that it is read. So:

Structure your data as above, which helps every reader.
If you sell through AI channels, look at Catalog Mapping for the fields that matter most.
Check storefront access is on for definitions your theme or public channels need. It's on by default when you create a definition in the Shopify admin, but definitions created by apps default to off [4][9]. (QuackChat reads through the store connection you approve, so it doesn't depend on this setting.)

A checklist before you publish

Every fact a shopper asks about is its own field, not a sentence in a paragraph.
Definitions, fields and entries are named in shopper language.
Each entry has a descriptive display name.
The answer lives inside the entry, and references only link things together.
Values are short, plain text or numbers, not images or JSON.
There is one entry per product family, not one per product.
Products reference the right entry through a metafield.
Storefront access is on where your theme or channels need it.

If you run QuackChat on Shopify, it reads your metaobjects automatically once your store is connected. There is no separate setup for them, though a store connected before mid-August 2026 may be asked to approve the app's access once more. Store connections are on the Pro plan and above, and every paid plan starts with a 30-day free trial. See pricing, the Shopify setup guide, or how QuackChat compares with Shopify Inbox. For the launch details, read QuackChat now connects to Shopify.

Frequently Asked Questions

What is the difference between a metafield and a metaobject?

A metafield adds one value to something that already exists, such as a product. A metaobject is a new type of structured record with several fields that you can reuse across many products. A size chart shared by twenty products is a metaobject; a fabric weight for one product is a metafield.

Should a size chart be a metaobject, a page or an image?

A metaobject. An image can't be read as text at all, and a rich-text page loses its rows and columns. A metaobject keeps each measurement as a named value that a theme, a catalogue feed and an AI assistant can all read.

Can ChatGPT and other AI agents read my Shopify metaobjects?

Shopify sends product data to AI channels through Shopify Catalog and says Catalog Mapping is most helpful for stores that keep custom data in metafields and metaobjects. It doesn't publish exactly which metaobject fields each agent receives, so structure the data well and use Catalog Mapping for the fields that matter.

Do metaobjects need storefront access turned on?

For your theme and public storefront channels, yes. It's on by default for definitions you create in the Shopify admin, but check it for any created by an app. An assistant that reads through an approved store connection, such as QuackChat, doesn't depend on it.

Which field type should I use for measurements?

For a size chart that any reader can use, a multi-line text field with one size per line works well, alongside a single "Unit" field. For single measurements on one product, Shopify's number and dimension types are fine.

Should I use category metafields or custom metafields for specs?

Category metafields first, because they follow Shopify's product taxonomy and feed discovery on your site, marketplaces and search engines. Add custom metafields for facts the taxonomy doesn't cover, and use a metaobject when a block of specs repeats across products.

Does structured size data reduce returns?

It removes one cause of them: a shopper choosing the wrong size because the chart was hard to read or the answer was guessed. Whether your return rate drops depends on your products and how often fit is the reason, so measure it before and after.

References

Shopify Enterprise blog, agentic-ready product data (5 June 2026). https://www.shopify.com/enterprise/blog/agentic-ready-product-data (retrieved 30 September 2026)
Australia Post, 2026 eCommerce Report summary. https://auspost.com.au/business/business-ideas/selling-online/ecommerce-trends (retrieved 30 September 2026)
Shopify Help Center, pop-up size chart tutorial. https://help.shopify.com/en/manual/custom-data/metafields/pop-up-tutorial (retrieved 30 September 2026)
Shopify.dev, metaobjects documentation. https://shopify.dev/docs/apps/build/custom-data/metaobjects (retrieved 30 September 2026)
Shopify.dev, list of metafield data types. https://shopify.dev/docs/apps/build/custom-data/metafields/list-of-data-types (retrieved 30 September 2026)
Shopify Help Center, category metafields. https://help.shopify.com/en/manual/custom-data/metafields/category-metafields (retrieved 30 September 2026)
Shopify Help Center, agentic storefronts: products. https://help.shopify.com/en/manual/online-sales-channels/agentic-storefronts/products (retrieved 30 September 2026)
Shopify, Spring '26 Edition for developers (17 June 2026). https://www.shopify.com/news/spring-26-edition-dev (retrieved 30 September 2026)
Shopify Help Center, building a metaobject. https://help.shopify.com/en/manual/custom-data/metaobjects/building-a-metaobject (retrieved 30 September 2026)