Content Generation · Read + Write

Product description from spec

Augment Rep · High Reason · Medium

The problem

E-commerce catalogs need 3,000 product descriptions every month. Copywriters can't keep up; suppliers send terse spec sheets; SEO suffers.

The AI approach

Vision-LLM reads the spec sheet (table of attributes), brand voice prompt enforces tone, output is a 100-word description plus 3 bullet highlights and a meta tag.

The outcome

~60s per SKU

Catalog throughput from ~30/day per writer → ~300/day with 1 writer reviewing. Brand voice stays consistent.

Try itInput → Process → Output

Input — supplier spec sheet
Bamboo Cutting Board — 18"SKU BB-018 · supplier: Tipa V Co.
materialMoso bamboo, sustainable
dimensions45 × 30 × 1.6 cm
weight1.1 kg
finishfood-safe mineral oil
carehand wash; oil monthly
originThailand · FSC certified
price (whsl)฿420
categoryKitchen / Cookware
Process — AI pipeline
1Read spec sheet (vision + table)ReadGenerative
2Apply brand voice + claim rulesRulesSymbolic
3Generate description + bulletsWriteGenerative
4Run brand-compliance checkRulesSymbolic
Output — description + bullets
Click Run demo to generate description, bullets, and SEO meta.
Catalog draft v1 · brand-checked
Title

Bamboo Cutting Board · 18″ · Sustainable Moso

Description

A solid 18-inch board cut from sustainable Moso bamboo, finished with food-safe mineral oil. Knife-friendly, lighter than hardwood, FSC-certified at the source. The full 45 × 30 cm work surface gives you room for a whole pumpkin — or a careful brunch line-up.

Highlights
  • FSC-certified bamboo · grown in 5 years, not 50
  • Food-safe mineral oil finish — no petroleum coatings
  • Hand-wash and re-oil monthly to keep the surface tight
SEO meta (160 chars)
Sustainable 18″ bamboo cutting board. FSC-certified Moso, food-safe oil finish, large 45×30 cm surface. Light, knife-friendly, hand-wash.
Compliance check: ✓ no banned claims · ✓ length under cap · ✓ origin verified.

Three AI types in this use case

SymbolicBrand voice rules (banned words, tone, length); regulatory claim list (no "medical-grade" without certification); SEO meta length cap.
PredictiveSpec-sheet table extractor (predicting cells); image classifier confirms photo matches description.
GenerativeLLM writes the description, 3 bullets, and SEO meta from the spec + brand voice prompt.

The stack

  • Vision-LLM · Claude Sonnet
  • Brand voice · prompt library
  • Compliance · banned-claim checker
  • Catalog · Shopify / Magento API

When this works

  • Spec sheets are mostly tabular and consistent
  • Brand voice is documented, not vibes
  • Compliance review catches rare claim violations

When it fails

  • Hallucinated specs (wattage, dimensions) — catalogue accuracy drops
  • Voice drift across categories — needs per-category prompts
  • Compliance violations slip if rules not codified
  • Auto-publish without human → catalogue trust dies