The Product Fields That Decide Your AI Shopping Rank

If you want the short version: AI shopping agents do not read your product page. They read a set of structured fields in Shopify's Global Catalog, and when one of those fields is blank, Shopify fills it in with a guess that you never get to see. This post is the field-by-field breakdown of which fields those are, how an assistant uses each one, and how to set it correctly in your store.

I pulled the specifics from Shopify's own Global Catalog extension (version 2026-04-08), the layer that adds Shopify fields on top of the base agent shopping protocol. It is the closest thing we have to a ranking-factor list for agentic commerce, and almost nobody has read it.

Two ways your data gets used

Every field does one of two jobs, and the distinction is the most important thing on this page.

• Hard filters decide whether you appear at all. When a shopper says "in stock, ships to Canada, under $50," the catalog drops every product that fails any of those before it ranks anything. A blank or wrong filter field is not a small penalty. It is removal.
• Ranking and inference decide which of the eligible products gets surfaced. This is where richer data, better ratings, and accurate attributes pull you ahead of the other stores that also passed the filters.

Get the filters right first. The best description in the world cannot rank a product that was filtered out three steps earlier.

The filter fields: these get you into the results

These are the fields the catalog can hard-filter on. Each one is a place you can disappear.

• Category. Shopify maps your product to its Standard Product Taxonomy through the category field, and the catalog can filter by taxonomy category. If your category is wrong or missing, Shopify infers one, and a wrong guess means you show up under someone else's shelf. Set the category explicitly in your admin for every product. This is the single highest-leverage field for matching category searches instead of only your brand name.
• Attributes: color, size, and target gender. The catalog filters variants on these three specifically, with and/or logic, so "black, size 10, women's" is a hard cut. These have to live as real variant options or mapped attributes, not as words buried in a title. If a shopper filters to a color you sell but did not model as an option, you are gone.
• Price. Filtered as an absolute range and as a relative price_tier of low, medium, or high within your category. Accurate per-variant pricing and currency are non-negotiable, and a stale price is one of the most common reasons a product is shown wrong or skipped.
• Availability. The catalog filters on available, and it also reads a running_low signal at the variant level. Real-time, accurate inventory keeps you eligible. Overselling or stale stock counts quietly remove you.
• Shipping coverage. Two filters here: ships_to (the shopper's destination country, region, or postal code) and ships_from (origin country). If your shipping profiles do not cover a market, you fail every "ships to there" search from that market. This is geography as a ranking factor.
• Condition. New or secondhand. Mostly relevant for resale and refurbished sellers, but if you sell secondhand and it is not labelled, you miss the shoppers filtering for it.
• Rating and review count. The catalog can filter to a minimum rating on a 0 to 5 scale and by review count. Reviews stopped being only social proof. They are now a literal filter you can fall below, which is why getting real reviews onto your product pages is an AI-visibility task, not just a conversion one.

The inference trap: what Shopify guesses when you leave a field blank

Here is the part that surprises people. The Global Catalog attaches a metadata object to your products that Shopify generates with machine learning. It contains inferred attributes (like material), tech_specs, top_features, and unique_selling_points. Shopify builds these from whatever you gave it: your titles, descriptions, and images.

Two things make this matter more than it looks.

First, there is no confidence score. The inferred values are presented as plain assertions. A guess that says your running shoe is made of mesh when it is knit looks identical to a correct value. The assistant has no way to know it is a guess, and neither do you.

Second, inference fills the vacuum you leave. The thinner your product data, the more the model invents, and the more of your catalog presence is written by an algorithm optimizing for plausibility rather than accuracy. The fix is not to fight the inference. It is to remove its room to operate by stating the facts yourself. A clear material, a real spec, an explicit feature in your product data gives the model less to guess and gives the assistant a fact instead of a probability.

This is also why your descriptions still matter even though no human reads them on the way to an AI recommendation. A rich, specific description is the raw material the inference runs on. Vague copy produces vague inference. For how to write copy that machines parse cleanly, see making product descriptions that AI can parse.

The identity and eligibility fields

A few fields are not about matching a search. They decide whether the sale is even yours to win.

• Seller identity. The catalog tags every result with a seller.domain, and assistants identify the seller by that domain, not by the brand in the title. If you are a brand, this is the argument for owning a clean, complete listing so a reseller's thinner one does not become the version an assistant trusts. The full mechanic is in the agentic buy box.
• Product identifiers. Shopify clusters identical products from different sellers under a Universal Product ID. A GTIN or barcode in your variant data helps Shopify match your product into the right cluster rather than stranding it alone, which matters most for commodity and resold goods.
• Checkout eligibility. Variants expose a checkout_url and a native_checkout eligibility flag. Products an agent can buy through without bouncing the shopper out are smoother to recommend, and frictionless beats friction every time an assistant is choosing what to surface.
• Compliance disclosures. Agents are required to display legal notices like Prop 65, allergens, age restrictions, and energy labels next to the product. Put those into your product data near the top rather than in a footer image, or the assistant has nothing compliant to show and may route around you.

The priority list

If you are deciding where to spend the next hour, work top down. The fields at the top remove you when wrong; the ones lower down decide close calls.

None of this is exotic. It is product-data hygiene, and it overlaps almost entirely with the work that already wins rich results in Google. The hard part has only ever been knowing which gaps actually cost you, and seeing what Shopify is inferring in the blanks.

See your gaps, including the ones Shopify is guessing

You cannot fix a wrong inference you cannot see. The fastest way to find your weak fields, and to catch where Shopify is guessing a category or attribute that does not match your admin, is to look at your store the way the catalog does.

That is what our free Shopify AI-readiness checker is for. It scores the product data and structured data the catalog reads and hands you the specific fixes, ranked by impact. If you suspect you are already missing from AI results, the companion read is why your products don't show up in AI shopping, and the big-picture context is the Shopify Global Catalog guide.

More in the AI and agentic commerce library.