Use case — Descriptions

Write product descriptions
AI agents can actually parse.

A description that converts a human browser and a description that satisfies an AI agent are not the same thing. Most catalogs are written for one and invisible to the other.

Install Legible

What AI agents look for

When an AI shopping agent evaluates a product against a buyer's query, it is not impressed by marketing voice. It is extracting attributes. A shopper asking "waterproof hiking boots, wide fit, under $180, men's size 12" gets answered by an agent that scans descriptions for the specific data points that resolve the query: waterproofing standard, last width, price, sizing availability. If those facts are absent, the product does not make the shortlist.

The five attributes agents prioritize most consistently are:

SpecificityDimensions, weights, volumes, temperatures — numbers anchor the description in fact
Material compositionExact materials and percentages, not "high quality" or "premium"
Use case and contextWhen to use it, what problem it solves, who it is for
Fit and sizingHow it fits, what size system, any deviation from standard
Audience signalWho it is designed for — explicit, not implied

Red flags agents penalize

Phrases that score near zero on every rubric

"Perfect for any occasion" — no use case signal
"Luxurious feel" — no material specificity
"Great addition to any wardrobe" — no differentiator
"Made with the finest materials" — no composition data
Bullet lists of marketing claims ("stylish," "comfortable," "versatile") without factual anchors
Descriptions under 80 words with no numbers, materials, or dimensions

Keyword-stuffed descriptions — the kind written for Google Shopping in 2018 — also perform poorly. An agent scoring "100% cotton tshirt tee shirt men mens cotton casual summer gym workout athletic top" sees noise, not signal. The repetition does not improve the relevance score; it reduces it.

How Legible scores descriptions

Legible passes each product description to Claude Haiku with a structured rubric prompt. The rubric scores five dimensions on a 0–20 scale each, returning a 0–100 composite score per product. The five dimensions are specificity, material completeness, use-case clarity, audience signal, and differentiation from generic copy.

The resulting report shows your catalog's score distribution — a histogram of where your products fall — and a ranked list of the bottom quartile: the products where a description rewrite will move your overall catalog readiness score most. A Shopify apparel store with 340 SKUs typically has 60–90 products scoring under 30, and fixing those 60–90 descriptions is the single highest-leverage action available.

Before and after

Before Score: 22

A comfortable linen shirt. Perfect for summer days. Great for casual or smart-casual occasions. Available in multiple colors.

After Score: 88

100% European flax linen, garment-washed for natural drape. Relaxed fit, drops from the shoulder. Breathes in heat up to 32°C. Unstructured collar, one chest pocket. Wears untucked or half-tucked with chinos or shorts. True to size; order your usual. Ships from Paris, delivers EU in 3 days.

The rewrite is not longer for its own sake. It is longer because it contains more facts. Every sentence answers a question an agent might ask. The score improvement from 22 to 88 is not cosmetic — products scoring above 80 appear in agent shortlists for relevant queries at a measurably higher rate than products scoring below 30 (modeled estimate from apparel and home-goods catalog testing).

Score every description in your catalog.

14-day free trial. No developer required. Works with any Shopify plan.