Building an attribute schema for Apparel that shoppers and AI can actually use
The apparel attributes that actually drive filters, size logic, and AI shopping answers, with a men's dress shirt before/after schema you can copy.
A men's dress shirt with "blue" in the title and nothing else is invisible to a shopper filtering by neck size, invisible to a merchandiser filtering by fit, and invisible to an AI shopping agent trying to match "slim-fit non-iron shirt, 15.5/34, for a wedding." Apparel is the category where attribute depth decides whether a product gets found at all. Here's the schema that actually matters, and what it looks like fixed.
Why apparel breaks first
Apparel has more shopper-critical variants than almost any other category: color, size, fit, fabric, and occasion all combine to decide whether an item works for a specific body and a specific event. Miss one and the product either drops out of a filtered search entirely or gets shown to the wrong shopper, which reads as a bad recommendation and often ends as a return.
The scale of that cost is well documented. Size and fit issues drive a large share of apparel returns industry-wide, with reported estimates for "fit-related" returns clustering well above other categories precisely because size, fit, and fabric behavior are attributes, not adjectives in a title. That's a data-completeness problem before it's a customer-service problem.
Google's own Merchant Center rules make the stakes concrete: for products in the Apparel & Accessories categories, color, size, gender, and age_group aren't optional metadata, they're required for a listing to be eligible for Shopping ads and free listings in the US, UK, Germany, France, Japan, and Brazil (Google Merchant Center product data specification). If those fields are blank or inconsistent, the product doesn't rank poorly, it disappears.
The attribute set that actually matters
Not every apparel field is equally load-bearing. Here's the tier that determines whether a garment surfaces in faceted search and AI answers, versus the tier that just improves the story once a shopper has already found it.
| Tier | Attribute | Why it matters |
|---|---|---|
| Facet-critical | Color (plus a normalized color family) | The single most-clicked filter in apparel navigation |
| Facet-critical | Size, size type, size system | Google requires these for Clothing; shoppers filter by them first |
| Facet-critical | Gender, age group | Required for Shopping eligibility; also the top segmentation shoppers use |
| Facet-critical | Fabric / material composition | Drives "cotton only," "no synthetics," and care-based filtering |
| Answer-critical | Fit (slim, regular, relaxed, tailored) | The attribute AI agents lean on when a shopper describes a body type |
| Answer-critical | Occasion (business, casual, wedding, athletic) | Maps shopper intent language to product inventory |
| Answer-critical | Care instructions | Answers "can I machine wash this" without a return |
| Nice-to-have | Pattern, collar type, sleeve length, closure type | Sharpens results once the shopper is already in the right bucket |
| Nice-to-have | Country of origin | Compliance and sourcing-conscious shoppers |
GS1 US frames the color and size layer specifically as a "common language" problem: standardized color and size code tables exist precisely because "Navy," "Dark Blue," and one retailer's proprietary swatch name are three incompatible values unless they're mapped to a shared vocabulary (GS1 US apparel and general merchandise standards). A facet filter can't OR across three spellings of the same color unless something upstream normalized them first.
Before and after: a men's dress shirt
Here's what a typical raw retailer feed looks like next to the same product enriched to the tier above.
| Field | Raw feed | Enriched |
|---|---|---|
| Title | Men's Shirt - Blue | Men's Slim-Fit Non-Iron Dress Shirt, Spread Collar |
| Color | Blue | Blue / color family: Blue |
| Size | M | Neck 15.5, Sleeve 34; size_type: regular |
| Gender | (blank) | Male |
| Age group | (blank) | Adult |
| Fabric | (blank) | 100% cotton, non-iron finish |
| Fit | (blank) | Slim fit |
| Collar type | (blank) | Spread collar |
| Sleeve length | (blank) | Long sleeve |
| Occasion | (blank) | Business, wedding-appropriate |
| Care | (blank) | Machine wash cold, tumble dry low |
The raw version fails the Google Merchant Center requirement outright (no gender, no age group) and can't be filtered by neck-and-sleeve size at all, because "M" isn't a real dress-shirt size, it's a guess mapped from a T-shirt scale. The enriched version can answer a specific shopper query and clears the eligibility bar for paid and free listings.
The AI answer test
Try this: ask an AI shopping assistant to recommend a slim-fit, non-iron dress shirt in a 15.5/34 for a summer wedding. A model answering that question is pattern-matching against structured fields, fit, fabric, sleeve/neck size, occasion, not against a title string. A product with "Blue" as its only attribute doesn't get excluded by the model on purpose; it simply never enters the candidate set, because there's nothing to match against. That's the same failure mode as a missing facet, just one layer downstream.
Building the schema, not just filling fields
The fix isn't a longer spec sheet. It's separating facet-critical fields (required, standardized, enforced against a controlled vocabulary) from answer-critical fields (fit, occasion, care) that AI agents and long-tail search queries depend on, and treating both as mandatory before a product goes live, not as a backlog to clean up later.
Anglera sits on top of whatever PIM or feed you already run and continuously checks apparel listings against this attribute tier, flags gaps like a missing size system or an unmapped color name, and fills them so products stay eligible for filtered search and legible to AI shopping agents. Your PIM stores the shirt. Anglera makes sure the shirt is actually findable.