Building an attribute schema for Skincare that shoppers and AI can actually use

Skincare is the category where "just add more attributes" actually means something specific. A serum isn't just a serum: it's a skin type, a concern, a list of actives at particular concentrations, a texture, and a set of exclusions a shopper is actively filtering for. When those fields are blank, the product doesn't rank lower in search — it disappears from the facet entirely, and it never gets mentioned when someone asks an AI assistant what to buy.

The attributes that actually matter in skincare

Generic PDP templates give you brand, size, and price. Skincare shoppers filter on none of those first. They filter on:

• Skin type: dry, oily, combination, normal, sensitive
• Skin concern: acne, hyperpigmentation/dark spots, fine lines and wrinkles, redness, dullness, uneven texture, large pores, dehydration
• Key active ingredients and concentration: niacinamide (2%, 5%, 10%), hyaluronic acid, vitamin C (L-ascorbic acid vs. derivatives), retinol/retinaldehyde, bakuchiol, peptides, salicylic acid, azelaic acid, glycolic acid
• Format/texture: serum, gel-cream, lotion, oil, essence, ampoule
• Exclusion flags: fragrance-free, alcohol-free, oil-free, silicone-free, sulfate-free, paraben-free, non-comedogenic
• Claims and certifications: cruelty-free, vegan, dermatologist-tested, hypoallergenic, reef-safe (for SPF)
• Usage context: AM/PM, pregnancy-safe, suitable for sensitive/reactive skin, layering order

An academic study of ingredient-based product recommendation identified five core skin types, eleven skin concerns, and seventeen additional preference attributes (fragrance-free, cruelty-free, non-comedogenic, and similar exclusion flags) as the working vocabulary shoppers and recommendation systems actually use — see the Beauty Beyond Words paper on explainable beauty recommendations. That's the real shape of a skincare schema. Anything narrower and you're filtering on brand and price alone, which is not how anyone shops for a serum.

Why a missing attribute is worse than a missing photo

A blank skin_type field doesn't get skipped in faceted search — the product gets excluded the moment a shopper clicks "sensitive skin" in the sidebar. Same with concern, same with "fragrance-free." Faceted navigation is a series of AND filters; a product with no value in a facet field fails every query that uses it, even if the product would have been a great match.

AI shopping assistants have the same failure mode, just less visible. ChatGPT, Gemini, and Perplexity lean on structured product data — schema.org Product/Offer markup, merchant feeds — to decide what to surface and how to describe it. Reporting on ChatGPT's shopping behavior found that a majority of cited product pages carry structured data, and that schema.org markup is treated as a baseline threshold: without it, an assistant either skips the source or uses it "fragmentarily," per analysis from iPullRank on how OpenAI's product feed works. Google's own guidance for merchant listing structured data lists the fields it expects on a product page, and skin type, concern, and active ingredient aren't part of the generic required set — brands have to add them explicitly, or an AI assistant summarizing "best serum for dry skin" has nothing to point to. See Google's structured data guidance for products.

Ask an AI to recommend a niacinamide serum for oily, acne-prone skin under $30, and it will compose an answer from whichever products actually carry active_ingredient, concentration, skin_type, and concern values in machine-readable form. A serum with a beautiful product description and no structured attributes is invisible to that query, no matter how good the formula is.

Before and after: a facial serum

Here's a raw retailer feed for a niacinamide serum versus what an enriched attribute set looks like.

The raw version reads fine to a human scanning the PDP. It fails every faceted filter and gives an AI assistant almost nothing to match against a shopper's stated skin type or concern. The enriched version turns that same product into ten filterable, quotable data points — and the "cruelty-free" claim moves from unstructured body copy, which AI systems weight less, into an actual attribute field.

Structuring the schema so it holds up

A workable skincare schema separates fields into tiers instead of one flat attribute list:

1. Core identity: product name, format, size, brand
2. Shopper-facing filters: skin type, concern, active ingredient(s) with concentration
3. Exclusion/claims flags: fragrance-free, non-comedogenic, cruelty-free, vegan, pregnancy-safe
4. Regulatory/compliance: full INCI ingredient list, pH where relevant, SPF rating and broad-spectrum status for daytime products

Each tier needs a controlled vocabulary — "oily" and "oily skin" should resolve to the same facet value, not fragment into two dead-end filters. That normalization work is usually where catalogs actually fall apart: attributes exist somewhere in a spec sheet or a marketing brief, but they never make it into the structured field a facet or an AI feed reads from.

That gap between "the fact exists" and "the fact is in a queryable field" is exactly what Anglera closes. Your PIM stores the data; Anglera continuously scores product data completeness against schemas like this one, flags where skin type, concern, or active-ingredient fields are missing or inconsistent, and gap-fills them from existing content, brand specs, and ingredient lists — without requiring a rip-and-replace of whatever commerce stack or PIM you already run.