Solutions / Product descriptions

Product description generator

Descriptions generated for your whole catalogue from real attributes and supplier data, in your tone, in every language you sell in — and regenerated when the attributes change.

E-commerce Content LLM
Whole catalogue
including the long tail nobody writes
Every language
you actually sell in
Regenerated
when attributes change
Who it is for

E-commerce and marketplace teams with a catalogue large enough that most items ship with the supplier’s text or nothing at all.

Short answer

Descriptions are generated from your structured attributes and supplier documentation, not invented — the generator is constrained to the facts it is given, so it does not claim a certification or a material that is not in the data. Output follows your tone rules and your SEO structure, covers every language you trade in, and lands back in the PIM or the store as a draft your team approves.

The problem

The top hundred products have good copy and the other nine thousand have whatever the supplier sent.

01

The long tail never gets written

A copywriter covers the bestsellers, and everything else keeps a specification dump. Those pages convert badly and rank for nothing, which then justifies not writing them.

02

Every language multiplies the problem

Four markets means four times the copy, so translation gets postponed and half the catalogue sells in the wrong language.

03

Generic AI copy is worse than none

Ungrounded generation produces confident sentences about materials and certifications the product does not have. One invented claim can be a regulatory problem, not just a bad page.

How it works

From trigger to result, step by step.

01

Establish the source of truth

Attributes, supplier documents, categories, and what must never be stated without evidence. Everything the generator writes has to trace back to something in this set.

02

Capture the tone in rules and examples

Your best existing descriptions become the reference: sentence length, what you claim, what you never say, how features are ordered. Tone is defined by examples, not adjectives.

03

Generate grounded and structured

Attributes go in, and the output follows a fixed structure — short lead, benefits, specifications, use cases — with the keyword and metadata rules your SEO team set. Nothing is written from the model’s general knowledge.

04

Check before publishing

Automated checks for banned claims, missing attributes, duplicate text and length; anything flagged goes to a person. Categories with regulatory exposure are always reviewed by a human.

05

Keep it in sync

When an attribute changes, the description regenerates and flags for approval. Otherwise the catalogue drifts back into the state it was in before the project.

Before / after

What changes on the ground.

Today, by hand
×Supplier text on most of the catalogue
×Long tail with no description at all
×Other-language markets running on machine translation
×Copywriter capacity as the ceiling
With the automation running
A description on every product, in every language
Output grounded in real attributes only
Tone and SEO structure applied consistently
Automatic regeneration when attributes change
What you get

Delivered, not demoed.

An attribute and source-of-truth audit for your catalogue
A tone specification derived from your own best copy
Grounded generation across the catalogue in every language you sell in
Automated claim, duplicate and length checks before publication
Write-back into your PIM or store, with approval and regeneration
Built with

We build in your stack rather than moving you onto ours. The list below is what this solution most often connects to — other systems are a scoping question, not a blocker.

Claude OpenAI Python LangChain PostgreSQL Shopify Akeneo n8n
Time to production4–8 weeks
Build priceFixed quote
First stepFree mini-audit
Honest limits

When this is not the right solution.

·If your attribute data is thin or wrong, the descriptions inherit that. Fixing attributes is the real project and it pays for itself twice over.
·With a few dozen hand-made products, a good copywriter will write better pages than any generator and the cost difference is negligible.
·In categories where every claim is regulated, generation can draft but a qualified person must approve every line — which limits how much time it saves.

Questions we get about this one

It is constrained to the attributes and documents it is given, and claim checks run before publication. That constraint is the whole design; ungrounded generation is what makes these projects fail.

Bring us the process that hurts.

The mini-audit is free: we take your version of this process apart and tell you plainly whether automating it pays. If it does, you get a scope and a fixed price.