Start free
Shopify · Product descriptions

It drafts.
You decide.

Most AI writers take a prompt and hand back prose. This one takes your catalog — the variants, the options, the price you actually charge — drafts from that, and then stops. Publishing to a live product is a separate, approved step.

Install on ShopifySee the whole agent ⟶
A waggle dance: one returning forager encoding what it found so the rest of the colony can act on it
The problem

A prompt-based writer does not know your store

Paste a product name into a general AI writer and it will produce confident copy about a product it has never seen. The output reads well and asserts things you cannot ship — a fabric weight, a compatibility, a delivery window. Those become returns.

Writing from a prompt

The model fills gaps with what products like yours are usually like. Every gap in your data becomes a confident sentence you now have to fact-check — and the ones you miss reach the customer.

Writing from the catalog

The draft is bounded by the record: your variants, options, price band, tags and existing copy. Where the data is thin, the description stays thin. That is the correct failure — cautious beats plausible.

The split

Draft and apply are different permissions

This is the part that matters, and it is enforced in the agent rather than promised in marketing copy.

Step
01

Draft — write scope

Ask for a description and you get one back, in the conversation, for that specific product. Nothing on your storefront changes. Read it, argue with it, ask for a shorter one.

Step
02

Apply — admin scope

Pushing a description onto a live product is a separate, higher permission and prompts for approval. An agent that can silently rewrite your catalog is a liability.

Step
03

Do the pages that matter

Work through the products that actually get traffic rather than bulk-rewriting everything. Four hundred listings rewritten at once read identically and rank for nothing.

What it uses

Your record, not a guess

Variants and options

Sizes, colours, materials and the combinations that actually exist — so the copy never describes a variant you do not stock.

Price and positioning

It knows the price band and what else you carry, so a £4 impulse item does not get copy written for a £400 considered purchase.

Your existing words

Where you have already written something, it edits rather than replaces — your terminology survives instead of being flattened into house style.

The same brain as your chat

The agent that writes the description is the agent that answers shoppers about it. A listing and a chat reply cannot contradict each other.

2
permissions — draft is not apply
Catalog
bounded — thin data, cautious copy
Approval
required before a live product changes
Related

The rest of the cluster

Shopping assistant

The same catalog knowledge, pointed at a shopper who is trying to decide.

AI discounts

The other write-capable surface — and why a discount has to be a bounded give.

Store as agent tools

The same catalog, exposed to software instead of to a person.

Questions

Product descriptions, answered

01Does it write descriptions from scratch or rewrite mine?+
Either, but it works from your actual product record — title, variants, options, price, tags and whatever description already exists. It is not generating from the product name alone, which is why the output mentions the things your listing actually has rather than inventing generic benefits.
02Will it publish changes to my store automatically?+
No, and the split is deliberate. Drafting and applying are two separate permissions: drafting sits at write scope, applying a description to a live product requires admin scope and asks for approval first. An agent that can silently rewrite 400 listings is a liability, not a feature.
03What stops it inventing product details?+
It drafts from the catalog record you already have. Where your data is thin the draft stays thin — it will describe the shape, use and materials you have recorded, not assert a fabric weight or a country of origin nobody told it. Thin input producing cautious output is the correct behaviour; a confident invention becomes a return.
04Can it do the whole catalog at once?+
It works product by product, and on purpose. Bulk-rewriting a catalog is how stores end up with 400 listings that read identically and rank for nothing. Do the products that actually get traffic, read the drafts, then apply.
05Is the output good for SEO?+
It writes for the shopper first, which is what search rewards now. Descriptions that answer real pre-purchase questions — fit, materials, what it is for — are the ones that earn the click and avoid the bounce. There is no keyword-stuffing mode, because that stopped working years ago.
06How is this different from a generic AI writer?+
A generic writer takes a prompt. This takes your catalog. It already knows the variants, the price band and what else you stock, because it is the same agent that answers shoppers about those products — so a description and a chat answer cannot contradict each other.

Draft one. Read it. Then decide.

Install the agent, point it at the product you like least, and see what it does with the data you already have.

Install on ShopifyAgent vs chatbot