Writing product descriptions with AI is a Skill job: you save one set of instructions, then feed the AI your spec sheets in batches and get back descriptions in your store's voice. Plan an afternoon to set up the instructions and work through your first few batches. The AI writes from the facts you give it, at volume. You check every number and every claim before anything goes live. For a store with dozens or hundreds of products, that split is the whole point.
- Shape
- Skill
- Sitting
- an afternoon
The shape: Skill
This job is a Skill because the work repeats many times and every instance follows the same rules. A product description for item forty should sound like the one for item four. Same voice, same length, same order of information, same words you never use.
A skill here means a saved set of instructions the AI follows every time. In Claude that is a Project with instructions; in ChatGPT it is a custom GPT or a saved project. It can also be a block of text in your notes app that you paste at the top of a fresh chat. The format matters less than the habit: you write the rules once, and every batch after that starts from them.
The AI fits this job well because the hard part of product copy is not ideas, it is stamina. Writing the first five descriptions is easy. By the fiftieth, most people start copying the manufacturer's paragraph or writing "high quality" and moving on. The AI does not get tired, and a saved set of rules pulls every batch back to the same standard.
The other half of the shape is that the AI knows nothing about your products except what you paste. It has never held the item, measured it or packed it. Every fact in a description has to come from your spec sheet, and every claim the AI adds on its own is a risk.
What this means for you: spend your effort on the instructions and the inputs. A good set of rules turns a spreadsheet of specs into usable copy. A vague set turns it into the same three sentences with different product names.
What you give the AI
Gather these before you open the chat. The first three go into the saved instructions once; the rest come with each batch.
- Three descriptions you already like. From your own store if you have them, or a paragraph from a store whose tone you admire, with a line on what you like about it. These set the voice better than any adjective you could choose.
- Who buys from you, in two or three sentences. "Mostly people buying a gift for someone who cooks" tells the AI more than "foodies".
- Your store's rules. Words you never use, claims you never make, how you write sizes and units, whether you use bullet points, and any length limit your store platform sets for titles.
- The spec sheet for each product. Materials, dimensions, weight, colors, sizes, care instructions, what is in the box, and compatibility where it applies. A row per product from a spreadsheet export works well; so does the manufacturer's spec text pasted as is.
- One line per product that only you know. Who it suits, what customers say when they pick it up, how it fits, what people pair it with, the question you get asked about it most. This line is what makes the description yours and not the manufacturer's.
- Your current description, if there is one. The AI keeps what is still true and you can see what changed.
If your store copy and your website copy should sound alike, our website copy page walks through capturing your voice from a recording first.
What comes back
For each product you get a title, a two-sentence opening, a short paragraph on who it suits and why, a bullet list of specs, and a list of claims to check. Ask for batches of five to ten products at a time. Larger batches tend to blur together toward the end.
Paste this into the saved instructions once, then paste your three examples, buyer notes and store rules under it.
You write product descriptions for a small online store.
The example descriptions, buyer notes and store rules below this prompt are the standard for every description.
For each product in the batch, write:
1. A product title, within the store's title limit (under 70 characters if none is given), with the product type and the main detail shoppers filter by.
2. Two opening sentences that say what the item is and who it suits, written to make sense on their own.
3. One short paragraph that uses the owner's note for this product.
4. A bullet list of specs, copied from the spec sheet, with units written our way.
Rules:
Use only facts from the spec sheet and the owner's note.
Do not add materials, measurements, certifications, origins, awards or health and safety claims that are not in the attachments.
Do not compare the item to other brands.
Use plain words and short sentences.
Do not use these words: premium, high quality, must-have, perfect for any occasion, game changer.
Match the voice of the example descriptions.
Write every description so it could not be mistaken for the manufacturer's text.
After each product, list:
1. Every factual claim in the description, so the owner can check it.
2. Anything a buyer would want to know that the spec sheet does not answer.Then each batch is a short message: "Batch 3, products below," with the spreadsheet rows and owner notes pasted under it.
Run the first batch on three products you know best. If those three come back wrong, fix the instructions before running the rest. A rule changed after fifty products means rechecking fifty products.
How to judge it
You can reject a bad description in under a minute once you know what bad looks like.
A bad description:
- Contains a number or material you did not give it. A weight, a dimension, "100% organic cotton" or "made in Italy" that is not on the spec sheet gets deleted on sight.
- Could sit on any store that sells the item. If you cover the store name and cannot tell whose page it is, the owner's note did not make it in.
- Opens with mood instead of the item. "Bring warmth to your everyday moments" tells a shopper on a phone nothing; "A 12-ounce stoneware mug that holds heat" tells them what they are looking at.
- Promises an outcome the item cannot guarantee, such as "keeps coffee hot all day" or "fits every body".
- Uses the banned words anyway, or their cousins: "luxurious", "stunning", "exquisite".
- Sounds identical to the description above it with the nouns swapped. Read three in a row; if the rhythm repeats, ask for variety in the openings.
- Gets the spec list wrong, which is the most expensive error, because it causes returns.
A good description says what the item is in the first sentence, answers the question you get asked most, and sounds like your three examples. The quickest test: read the specs line by line against the spec sheet, then read the opening out loud.
What stays with you
Some of this job no AI can do for you.
Every fact. The AI lists each claim so you can check it, but the checking is yours. Sizes, materials, care instructions and what is in the box are the lines that cause returns and complaints when they are wrong.
The rules for your category. Some products carry rules about what a seller may say: anything worn on skin, eaten, given to children, or sold with health or safety claims. Those rules are yours to know, and the AI will not reliably know them for you.
The owner's note. The one line per product about who it suits and what people ask is the part that makes a description sell, and only you have it. A batch without those notes comes back accurate and forgettable.
The photos. Shoppers check the photos before they read a word. Clear photos of the real item, in use and at scale, do more than any paragraph.
The pasting. You paste each description into your store, save it, and open the product page on your phone. Check that the title is not cut off, the bullets render as bullets, and the opening two sentences sit above the fold.
Once the descriptions are live, our schema markup page covers the structured data that helps search engines read each product page.
An example, made up for this page
Say you run a small online shop that sells hand-poured candles in eight scents and three sizes. You paste in three descriptions from your own site that you still like, two sentences about your buyers (mostly people sending a gift), and your rules: burn times in hours, sizes in ounces, never the word "luxury".
You export the first batch of eight products from your store as a spreadsheet: scent, wax type, wick, size, burn time. Next to each row you add one line, such as "the one people buy for a new homeowner" or "customers say it smells like a bakery, not like a candle".
The AI returns eight descriptions. Two give a burn time for the large size that does not match your spec sheet, and the claims list flags both. One opens with "Transform your space", which you send back. The other five need only small cuts.
You fix the burn times, add a line to the instructions banning openings about transforming anything, and run the next batch. By the end of the afternoon all twenty-four products have new descriptions, checked against the spreadsheet.
None of that is a real business or a real result. It shows the size and shape of the afternoon, not what it will produce.
How many at once
The right way to run this job depends on how many products you have.
| Products to write | How to run it |
|---|---|
| Under 20 | Paste spec text straight into the chat, five at a time |
| 20 to 200 | Export a spreadsheet and send it in batches of ten rows, over an afternoon or two |
| Several hundred, changing often | Treat it as a few sessions, and ask whether your store platform has its own AI description tool |
Of every job on this list, social media is the one that comes back every week, and Boomp is software that does that one for you. See ten posts made from your website, free
