Shopify metafields for AI visibility: a practical guide
Of the nine signals an AI shopping agent reads from a product, structured metafields are the least glamorous and the most valuable. A title is prose a model has to interpret. A metafield is a labeled fact — "Fit: Relaxed", "Material: 100% organic cotton", "Capacity: 1.5 L" — that needs no interpretation at all. This is the field type most Shopify catalogs leave empty, mostly because merchants don't know it's there.
What a metafield actually is
A metafield is a key-value pair attached to a product, beyond the built-in fields (title, description, price, vendor). Shopify ships with a standard product taxonomy, and every category in that taxonomy comes with a predefined set of category metafields: a "Shoes" product gets fields like size, width and sole material; a "Coffee" product gets roast level and caffeine content. You can also define your own custom metafields for anything the taxonomy doesn't cover.
The category ones matter more for AI visibility. They ride along in Shopify's product feed with a namespace AI catalog surfaces specifically read, so filling them in is the closest thing to a guaranteed signal in this list — unlike free-text fields, which a model still has to parse and trust.
Where to find them
- Set the product category first. In the Shopify admin, open a product and set "Product category" to the most specific taxonomy node that fits (not just "Apparel" — "Apparel > Clothing > Shirts > T-Shirts"). The category is what unlocks the matching metafield set — an uncategorized product has no category metafields to fill at all.
- Scroll to the "Category" section on the product page. Once a category is set, Shopify shows the relevant attribute fields right there — no admin configuration needed, no app required.
- Fill in what's true; skip what isn't. Not every attribute applies to every product. A handmade item with no official size chart doesn't need a fabricated one — an honest gap beats an invented value, and models are reasonably good at spotting invented ones.
Which ones to prioritize
If you're staring at a long attribute list and limited time, work in this order:
- Whatever the shopper would ask about first. For apparel that's size and fit; for electronics, compatibility and power; for food, ingredients and allergens. These are the attributes that show up verbatim in "does this work for me" questions — exactly what an assistant is trying to answer.
- Whatever disambiguates near-identical products. If your catalog has five variants of the same base product, the metafields (not the title) are often the only thing that tells them apart cleanly.
- Whatever a competitor is likely missing. This is the field type stores skip most often, so filling it in is disproportionately cheap relative to how rare it is.
Metafields don't replace the description
It's tempting to treat a full metafield set as "done" and leave the description thin. Don't — they do different jobs. The description is where a model finds context, use cases and the narrative it quotes in an answer. Metafields are where it finds the exact fact it checks that narrative against. A rich description with no metafields is unverifiable; a full metafield sheet with no description is unquotable. You need both.
Doing this at catalog scale
Setting metafields by hand works for a hundred products on a slow afternoon. Past that, the realistic path is: audit the catalog to see which categories and attributes are actually missing (not "add more metafields everywhere" — that's not actionable), prioritize the products with the most traffic or the thinnest data, and fill the gaps in bulk rather than product by product.