Blog · 5 min read

AI Prompts for Handmade Sellers: The Four Jobs Nobody Writes Prompts For

Most prompt lists for makers stop at product descriptions. The jobs that actually eat a maker's week are the four below — pricing, the awkward messages, restock decisions, and photography briefs.

Search for AI prompts for handmade sellers and you get product descriptions. Twenty variations of "write a warm, story-driven description for my handmade [product]". They're fine. They're also the one part of the job most makers are already good at, because they care about the thing they made.

The parts that actually eat the week are elsewhere: working out whether a price is sane, answering the message that's been sitting unread for two days because you don't know what to say, deciding whether to make more of something, and explaining to a photographer — or to yourself at 11pm with a phone — what the picture needs to show.

Here are prompts for those four, and an honest note about which ones AI is bad at.

1. The price you're afraid to charge

Most handmade pricing is materials plus a number that doesn't feel greedy. That number is almost always too low, because it silently excludes your own hours, the listing fees, the failed batch, and the twenty minutes you spent answering questions before the sale.

The model can't know your numbers. What it's genuinely good at is naming the costs you forgot.

I make [product] by hand and sell on [platform].

What I know: materials [amount] per unit, about [number] minutes
of my time per unit, [platform] fees [%].

List every cost that belongs in the unit cost and that a solo maker
typically forgets. One line each, explaining why it belongs. Don't
calculate anything — I'll do the arithmetic myself with my own numbers.

That last sentence matters. Ask for a calculation and you'll get one, complete and confident and built on invented figures. Ask for the list and you get the thing it actually knows.

Once you have the list, the arithmetic is one rule: to get a 40% margin you divide the cost by 0.6, you don't multiply by 1.4. Multiplying gives you 29%. This is the single most common arithmetic error in handmade pricing and it costs a tenth of every sale.

2. The message you've been avoiding

Every maker has one sitting in the inbox. A customer wants a refund on something custom. Someone asks for a discount. A parcel arrived broken and it isn't your fault but it's still your problem.

These are hard not because the words are hard but because you're annoyed, and you'd rather write it tomorrow. That's what the model is for — not to be more eloquent than you, but to be less annoyed.

A customer wrote: "[paste their message]".

Context they don't know: [what actually happened].
What I can offer: [replacement / partial refund / nothing].

Write a reply in 4-5 sentences. Warm but not apologetic, no
over-explaining, no "we sincerely apologise for any inconvenience".
Say plainly what I can do and what I can't.

Two rules for using the output. Read it aloud before sending — if it doesn't sound like you, it will read as a form letter to someone who bought from you specifically because you aren't a company. And never let it invent a policy. If you don't have a returns policy, the reply shouldn't imply one.

3. Whether to make more of it

This one gets skipped because it feels like it needs a spreadsheet and a business degree. It needs neither — it needs the question asked out loud.

Here's what I sold in the last [period]:
[paste your list — item, quantity, price]

Time to make each: [rough minutes per item]

Group these by revenue per hour of my time, not by revenue. Then tell
me which items look like they're worth making more of and which are
quietly losing me time. Where the data is too thin to say, say so
instead of guessing.

The "say so instead of guessing" clause is doing real work. Without it the model will confidently rank six items off three sales each. With it, you usually get a shorter answer and a more honest one.

The result is often uncomfortable: the thing you enjoy making least earns three times what your favourite does. That's information, not an instruction — plenty of makers keep the low-earner because it's why they started. But now it's a choice.

4. The photo brief

Handmade listings live or die on the first image, and most makers shoot without a brief because they're both the client and the photographer.

I'm photographing [product] for a [platform] listing.
Buyer: [who they are, in one line].
The thing that's hard to tell from a photo: [texture / real size /
how it's used].

Give me a shot list of 5 photos, in the order they should appear in
the listing. For each: what's in frame, why that shot exists, and
what a buyer's objection it answers. Assume a phone and a window,
not a studio.

The "assume a phone and a window" line is the difference between a usable brief and a fantasy involving a lightbox you don't own.

What AI is bad at here, plainly

Your voice, unsupervised. Handmade buyers are buying a person. Text that reads like a brand will undo that, and it reads like a brand by default. Every output above needs a pass in your own words.

Anything numeric it wasn't given. Prices, market sizes, "typical" margins for your craft — all invented, all plausible-looking. If a number appears that you didn't type, delete it.

Knowing your customer. It has never met the woman who buys three of your bags every Christmas. You have. Where the model's suggestion and your gut disagree about a specific person, your gut is holding more data.

Where to start

Pick the one of the four you've been avoiding longest — that's usually the message in the inbox — and do it now, with a real message, not a made-up one. Five minutes with a real task tells you more than an hour of reading about prompts.

More on the arithmetic behind point 1: pricing and true cost. On the awkward messages specifically: replying to reviews and customer messages.

There's a free demo with 15 prompts — no signup — if you'd like to see the fill-in-the-blank format before deciding whether it's for you.