How to Write Product Descriptions With AI
Export four fields, hand Claude your brand positioning along with them, and ask for descriptions that lead with the problem the product solves. Here's the process on a real WooCommerce store, including the review step before import.
The short answer
Rewriting product descriptions with AI works when you give it your store, not the product list alone. Export four fields into a spreadsheet: product name, short description, long description and meta. Upload that alongside your brand positioning and style guide, then ask for descriptions that lead with the problem the product solves rather than the specifications. Ask for a review file before an import file, so you see every change first.
At a glance
- Time
- ≈ 30 minutes for five products
- Cost
- Free
- You need
- A store export and a positioning document
- Where
- Claude, then your store's importer
Do product descriptions matter? They do. And most stores get them wrong in one of two ways.
Some descriptions came straight from the manufacturer. That means the exact same words are sitting on every other store selling that product.
The rest you wrote yourself, for something that's yours. They still don't sell, because they read like a spec sheet. Material, size, made wherever.
Either way the page is failing at both of its jobs. It isn't selling to the person reading it, and it isn't getting found by anybody searching.
So I gave Claude every product page on a real store and asked it to rewrite them. Watch the video above and you'll see the whole run, including the import.
Key takeaways
- 01
Unclear product information caused 20% of failed purchases in Nielsen Norman Group's e-commerce studies. Not price, not shipping. People simply could not tell what the thing was or what it would do for them.
- 02
A product page has two jobs, and most pages do neither. Sell to the person reading it, and get found in search, which now means the Google kind and the AI kind.
- 03
Never hand AI a bare list of products. Upload your positioning and style guide with it, or what comes back is generic copy with your product names dropped into it.
- 04
Manufacturer descriptions are the same words your competitors are using. Rewriting them is the cheapest thing you can do to stop looking identical to every other store carrying that product.
- 05
Ask for a review file before you ask for an import file. Original beside proposed, in one document, so nothing reaches the store that you have not read.
The most useful thing it did was something nobody asked it to do.
Claude was working through five products. To write them properly it had to read the whole catalogue, because that's where the context is.
While it was in there it flagged two things nobody had asked about. Some products were filed in the wrong category. And the search titles across the store were set up in a way that was working against them.
Both went straight onto the fix list for that store.
That's the difference between a tool that rewrites text and a tool that has read your store. One returns better sentences. The other tells you the sentences were not the only problem.
It's also why the upload matters so much. The positioning document carries the customer, the benefits those products deliver, and the actual words people type when they're trying to buy this kind of thing. Without it, you get competent writing about a product Claude has never seen in context.
Prefer to watch
8:36Rewriting a catalogue without breaking it
Export the fields that matter, not the whole store
Nearly every platform has an export. WooCommerce, Shopify, Wix, Squarespace. Same idea, different button.
Don't export everything. You want four columns: product name, short description, long description, and the meta description, which is the line that shows up in search results and tells Google what the page is about.
Four fields keeps the file small enough for AI to work with properly.
Give Claude your store before you give it your products
This is the step people skip, and skipping it is why AI product descriptions usually sound like AI product descriptions.
Upload your brand positioning and your style guide first. If you don't have either, that's the job to do before this one, and Claude can build both from what you already have.
Then upload the product export. Now it knows the brand, the buyer, and the products.
Start with five products, not two hundred
Tell it explicitly to do the top five and stop.
You're checking whether the output is right before you turn it loose on a catalogue. Five is enough to see the pattern, and small enough that reading every word is realistic.
Ask for the problem, not prettier specs
The instruction that changes the output: lead with the real problem this solves for the person buying it.
Nobody buys because of the material and the dimensions. They buy because of what it does for them. Your positioning document already says what that is, which is exactly why it went in first.
Say out loud that search now means two things
Ask for language that works for Google and for AI search.
When someone asks ChatGPT or Gemini a question directly, the answer they get back is the new search result. If your page doesn't say what it is in plain language, you're not in either one.
So: speak to the human, and use the words people actually search.
Ask for a review file before an import file
Before it gives you anything importable, ask for a review document: the original description, the proposed one, and what changed.
Read it. This is your only checkpoint, and it costs five minutes.
Ask for the import file, map it, import
Once the review looks right, ask for a file formatted for your platform's importer.
In WooCommerce that's Products, then Import, choose the file, tick the option to update existing products, map the CSV columns onto the WooCommerce fields, and run it. It updates what's there rather than creating duplicates.
Check three products on the live store afterwards. Always check.
What changed on one product
What a typical page looked like first
A very short description with one generic line in it. A long description that was either a spec sheet or completely blank. A meta that was empty, or auto generated from the short description that was already thin.
That's what the buyer landed on, and it's what Google and the AI tools had to work with.
The short description
After: it leads with what the product does for you. One line, and it's about the person rather than the object.
That single line is what shows in listings, so it's doing more work than any other sentence on the page.
The long description
After: it talks to the actual buyer. It still contains the specifications, because people want them, but they sit underneath the reason to care rather than instead of it.
Compared with what was there before, it wasn't close.
The meta
After: written for what people search, not scraped from the first line of the page.
This is the one most stores never touch, and it's the one that decides whether anybody clicks.
Moving it off the browser
The browser works fine. For a whole catalogue I'd rather not re-upload the same brand files every time I sit down.
So the same job runs in Claude Cowork on the desktop, with the brand guide and the product list in one folder. It reads them straight from there and works through the catalogue page after page in the same voice.
Sources
Questions people actually ask
Can AI write product descriptions that are actually good?+
Yes, if you give it context. Handed a bare product list it writes competent, generic copy. Handed your positioning, your style guide and your product list together, it writes something that sounds like your store.
Does this work on Shopify, or only WooCommerce?+
Any platform with an export and an import, which is nearly all of them. Shopify, Wix and Squarespace all do it. The columns are named slightly differently and the process is identical.
Will Google penalise me for manufacturer descriptions?+
There's no penalty. What happens is worse in practice: Google groups pages carrying the same content and shows one of them, and it usually isn't the smallest store. Rewriting is how you stop being the one that gets folded away.
How many products should I do at once?+
Five to start, always. Once you've read those five and they're right, the batch size stops mattering much. What matters is that you've checked the output before scaling it.
What is the meta description and do I need it?+
It's the line under your page title in search results. It doesn't directly change your ranking, but it decides whether the person scrolling picks your result or the next one, and most stores leave it empty or auto generated.
Do I have to write a positioning document first?+
You need something that says who the customer is, what problem these products solve, and the words people use when they're looking for them. Claude can draft it from your existing site in an afternoon, and it makes every later step better.
What if the rewrite loses information I need?+
That's what the review file catches. You see the original and the proposed side by side before anything goes near the store, and you can tell it to put back anything it dropped.
How far should I take this?+
As far as the page is worth. In the video I put it this way: "You could do it for a couple of minutes or for hours, depending on how much this page is worth to you." A product that brings in most of your revenue deserves the hours. The rest do not.
About the author
Hey, I'm Traci
Digital Marketing Strategist
With over 25 years in this industry I help small business owners and creators turn their online presence into profit.
Keep going
The setup, the templates, and what I build next
Not more reading. The Claude setup guide you can follow tonight, and the templates I actually use. The page opens the moment you sign up.
Get the guideFree · you get it straight away