---
title: "AI agents and beauty stores"
source: "https://ottom.io/verticals/beauty"
summary: "Beauty catalogues are wide, shallow and heavy on shades, which makes them easy for an agent to read and hard for it to choose from. The recurring failure is not the buy path but the evaluate stage: an agent asked for a foundation for a given skin tone needs shade and ingredient information as data, and it is usually published as pictures and prose."
---

# AI agents and beauty stores

Beauty catalogues are wide, shallow and heavy on shades, which makes them easy for an agent to read and hard for it to choose from. The recurring failure is not the buy path but the evaluate stage: an agent asked for a foundation for a given skin tone needs shade and ingredient information as data, and it is usually published as pictures and prose.

## What makes this category harder

- Shade is a variant dimension with no standard vocabulary, so it rarely appears as structured data.
- Ingredient and allergen information is a common request and is almost always unstructured prose.
- Regulatory and claims text is often an image, which makes it unreadable to a reader that needs it.

## Where agents stop, and why here

- shades are variants and are seldom published as such (https://ottom.io/learn/stalls/no-variant-data)
- product markup is often absent on heavily designed pages (https://ottom.io/learn/stalls/missing-schema-org)
- ingredient panels are photographed rather than typed (https://ottom.io/learn/stalls/policy-in-image)

## Questions

### Shades are visual. How would structured data even help?

It does not have to describe the colour. Naming the shade as a variant with its own price and stock is enough for an agent to tell you it can buy that one, which is the question it is actually asking.