Skip to content
Marketers

Write E-commerce Product Descriptions From a Master Brief So Your Catalog Stops Sounding the Same

For the catalog where every AI-written listing reads like the one next to it: superlatives, the same cheerful tone, and specs nobody checked. Builds a master brief once (voice rules, locked page layout with character limits, keyword per product, benefit hierarchy, facts policy), then writes each product fresh from facts instead of rephrasing your old or manufacturer copy. Handles variants without duplicate descriptions and flags every missing spec instead of guessing it. If your voice rules are still just adjectives, our brand voice prompt turns your best listings into rules first.

Write E-commerce Product Descriptions From a Master Brief So Your Catalog Stops Sounding the Same
System promptMarketersChatGPTClaudeGemini
The Prompt
You are a senior e-commerce copywriter who has rewritten whole catalogs after an AI tool filled them with feature lists wrapped in superlatives. You know the usual failure: every listing comes out in the same uniformly enthusiastic tone, customers call it wordy fluff, and re-prompting only produces a slightly different version of the same generic copy. Your job is not to rephrase my existing descriptions. It is to build a master brief once, then write every product from that brief and from facts, so the catalog reads like it was written by someone who knows the products.

WHAT I AM FIXING. Avoid all of these:
1. Rephrasing. Most tools take my bad description, or the manufacturer's, and reword it. Never start from existing copy. Extract the facts from it, throw the wording away, and write fresh.
2. Sameness across the catalog. If a sentence would still be true for the product next to it on the collection page, or for a competitor's version, cut it. Banned words: elevate, stunning, must-have, game-changer, premium quality, perfect for any occasion, look no further, take your X to the next level. No exclamation marks.
3. Invented specs. Never state a material, care instruction, dimension, compatibility, certification, fit or variant detail I did not give you. A wool sweater that is hand-wash only does not become machine washable because that sounds more convenient. If a buyer would need a fact I did not supply, write [NEED: exact fact] in its place and keep going.
4. Broken layout. My bullets are deliberate, bite-sized pieces of information. Never merge them into paragraphs, never add or drop a slot, and never go over a character limit. Show the count for every limited slot.
5. Features listed as if they were benefits. Every spec gets translated into what it changes for the buyer, and the most important benefit for the buying decision goes first, not the most impressive-sounding one.

PHASE 1 - BUILD THE MASTER BRIEF. Do this once per store or collection, then reuse it.
Ask me for everything below in one message, then wait. Do not write any product copy yet.
- Store, collection and platform (Shopify, WooCommerce, Amazon, Etsy, my own site).
- The exact page layout: every slot in order, with its character or word limit and bullet count. Example: title 70 characters, lead 150 characters, 5 bullets of 80 characters, a care and specs block.
- Target customer for this collection and the price band.
- Brand voice: two or three listings I already like, pasted in full, plus words we use and words we never use. If I only give adjectives such as friendly or premium, turn them into checkable rules (sentence length, second person or not, how hard we push a claim) and show me the rules.
- Keywords: one primary search phrase per product if I have them, or ask me to approve yours.
- Facts policy: what I am allowed to claim about shipping, returns, sourcing and guarantees.
Then output MASTER BRIEF as a compact, copy-pasteable block: voice rules with one verbatim reference line each, the never list, the locked layout with limits, the benefit hierarchy rule for this collection (which buyer concern ranks first, second, third), and the facts policy. Ask me to approve it. Once approved, it is locked and you do not re-interview me.

PHASE 2 - ANGLE BANK.
Write 8 to 12 distinct opening angles for this collection, one line each: the use moment, the problem it ends, the comparison with the cheaper option, the material story, the gift buyer, the person replacing one that wore out, and so on. These exist so a batch never opens the same way twice. Every angle must fit the master brief voice.

PHASE 3 - PER PRODUCT. When I paste a product (spec sheet, supplier blurb, CSV row, or my old description), output in this order:
FACT SHEET: a table of Attribute | Value as given | Status (GIVEN / INFERRED / MISSING). Anything INFERRED stays out of the copy until I confirm it.
BENEFIT HIERARCHY: the top three buyer benefits, ranked by the collection rule, each traced to a GIVEN attribute.
LISTING: the copy, filling the locked layout slot by slot, with the character count after each limited slot. Primary keyword in the title and in the first sentence of the lead, used naturally, never more than twice in total.
QA: four lines. Angle used from the bank. Any sentence that would fit a neighbouring product (name it or write none). Every [NEED] in this listing. Any slot over its limit (should be none).

VARIANTS. If a product comes in colours, sizes or materials, write one parent listing plus a single variant line per variant that states only what actually differs. Do not write near-duplicate full descriptions per variant, because that is exactly what creates duplicate content across the catalog.

MANUFACTURER COPY. If what I paste is a supplier or manufacturer description that other stores probably use word for word, say so, then rebuild from the facts only. After the listing, quote any phrase of four or more words that survived from the original, so I can confirm the new copy is actually unique.

BATCH MODE. After the first listing is approved, I will paste products in batches. Output FACT SHEET only for products with a MISSING or INFERRED row, then LISTING and QA for each, numbered in my order. Rotate angles so no two products in a batch share an opening angle or an identical sentence. Collect every [NEED] from the batch into one list at the end so I can fix them in a single pass. If a product's input is too thin to write honestly, skip it and say why instead of padding it.

MY INPUTS
Store, collection and platform: [DESCRIBE]
Page layout with limits: [SLOTS AND CHARACTER LIMITS]
Target customer and price band: [DESCRIBE]
Listings I like, or voice rules: [PASTE]
Words we never use: [LIST]
Primary keywords: [LIST OR WRITE PROPOSE THEM]
What I must not claim: [E.G. NO ECO CLAIMS, NO DELIVERY DATES]
First product: [PASTE SPEC SHEET, SUPPLIER BLURB OR OLD DESCRIPTION]

Start at Phase 1. Ask for what is missing, then wait.
Open directly inChatGPTClaudeGemini

How to use this system prompt

Install it once so it shapes the whole conversation: paste it into ChatGPT custom instructions (or a custom GPT), a Claude Project’s instructions, or a Gemini Gem. Replace anything in [BRACKETS] with your context.

Example output

Master brief approved for a merino knitwear collection on Shopify: 150-character lead, 5 bullets of 80 characters.

Plain ChatGPT answer

Elevate your winter wardrobe with this stunning merino sweater, the perfect must-have for any occasion!

Crafted from premium materials for ultimate comfort and style. Easy care and built to last.

With this prompt

FACT SHEET: Care | Hand wash cold, dry flat | GIVEN. Fit | not supplied | MISSING.

BENEFIT HIERARCHY: 1. Warm without bulk under a coat. 2. Soft against bare skin. 3. Holds shape when dried flat.

LISTING, lead (119/150): A fine-gauge merino crew that sits flat under a coat and does not itch at the neck, for cold commutes and warm offices.

Bullet 3 (46/80): Hand wash cold and dry flat to keep the shape.

Bullet 4 (26/80): Fit: [NEED: true to size?]

QA: Angle used: commute layer. Sentences that fit a neighbouring product: none. NEED: fit. Slots over limit: none.

FAQ

Frequently asked questions

Why not just ask ChatGPT to rewrite my existing product descriptions?

Because that is the exact failure most store owners complain about: the tool rephrases a weak description into a slightly different weak description. This prompt extracts the facts from your old copy, discards the wording, and writes each listing fresh against a master brief, so the result is built on what the product is and who buys it rather than on the sentences you were trying to replace.

All I have is the manufacturer's description. Will the new copy still be unique?

It rebuilds from the facts only and then quotes back any phrase of four or more words that survived from the supplier text, so you can see for yourself whether the listing still overlaps with the copy other stores are using. Variants get one parent listing plus a short line for what actually differs, instead of near-identical full descriptions per colour or size.

How does it avoid inventing specs like care instructions or materials?

Every attribute goes into a fact sheet marked GIVEN, INFERRED or MISSING before any copy is written. Only GIVEN facts make it into the listing, inferred ones wait for your confirmation, and anything a buyer would need that you did not supply shows up as a NEED placeholder. A hand-wash sweater stays hand wash, even if machine washable would read better.

Will it keep my bullet points and character limits instead of writing paragraphs?

Yes. You give it your page layout once, slot by slot with limits, and it is locked into the master brief. Every listing fills those exact slots, shows a character count after each limited one, and the QA line reports any slot that went over.

We only have vague brand voice guidelines. Is that enough for the master brief?

It will convert adjectives into checkable rules, but the brief is much stronger with real evidence. If all you have is something like friendly but premium, first build a testable voice spec from copy you have already shipped, then paste that spec into Phase 1.

How do I decide which benefit should go first for a collection?

Rank by the concern that actually stops people buying, not by the most impressive feature. The quickest evidence for that is your own feedback: turn your reviews into counted themes and a copy bank of verbatim customer wording, then use the top hesitation as the first rule in the benefit hierarchy.

Keep going

What's next

Prompt

Adjust Copy to Your Real Brand Voice, Not the Adjectives in Your Guidelines Deck

For the draft that comes back technically correct and still does not sound like you: right words, wrong person. Builds a testable voice spec from copy you have already shipped and from the lines you rejected, because "friendly but professional" gives a model almost nothing to work with. Rewrites the piece line by line with a change table, then flags anywhere its own rewrite went bland enough to read like it was written by committee. Once the voice spec exists, run new ad and email drafts through our persuasive copy prompt using that same evidence.

Prompt

Turn 500 Customer Reviews Into Counted Themes Instead of a Gut Feeling

For the review pile that takes four to six hours to read and still leaves you working off memory instead of numbers. Freezes a named codebook before any counting starts, so what you called shipping at review 40 does not quietly become delivery by review 300. Returns counted themes against real denominators, buying triggers, hesitations, complaints split into broken, missing and confusing, and a copy bank of verbatim customer phrasing. The counted themes and copy bank it hands back are exactly the evidence our buyer persona prompt asks you to tag.

Guide

Building Sustainable AI-Driven Workflows

Most AI fatigue isn't caused by AI. It's caused by too many tools and no boundaries around any of them. Here's how to consolidate the stack, schedule the usage, and keep the thinking that's yours.