How to Prompt AI to Write a Product Description
Generic AI prompts produce generic product descriptions. Here's how to feed the model specifics, sensory detail, and platform norms so ecommerce listings actually sell.
Generic prompts produce generic copy. If you type "write a product description for a leather wallet" into an AI model, you get sentences that could describe any leather wallet ever made: durable, stylish, perfect for everyday use. The model isn't broken. It's doing exactly what you asked, which was nothing in particular. Specificity is the whole game here. Give the model the material, the customer, a few concrete sensory details, and the platform's format, and the output stops sounding like every other listing on the internet. This post covers how to do that, plus how to iterate once you have a first draft.
One note before we go further. If you're here because you need to describe a piece of software or a feature set to an engineering team, you want how to prompt AI to write a product spec instead. That post is about internal technical documents. This one is about selling a physical or digital product to a customer on Etsy, Amazon, Shopify, or a similar marketplace. Different genre, different goals, different prompts.
Why "write a product description for X" fails
A blank prompt forces the model to invent details, and it invents the most statistically average ones it can find. That means adjectives like "premium," "high-quality," and "perfect for any occasion." None of these mean anything specific to your product. They're true of every product.
This is the same failure mode covered in our broader piece on prompt engineering: a vague instruction gets a vague answer. The fix isn't a smarter model. It's a better-loaded prompt.
The five inputs that actually move the needle
Gather these before you write a single prompt. Five minutes here saves three rounds of regeneration later.
Material or specs. Not "soft cotton" but "14oz brushed cotton fleece, garment-dyed, side-seamed." Not "durable" but "solid brass hardware, 2mm veg-tan leather." Numbers and material names read as credible because they are.
The actual customer and their problem. Not "everyone" but "a parent who needs a diaper bag that doesn't look like a diaper bag." Not "gift shoppers" but "someone buying a first anniversary gift and wants it to feel personal, not generic."
Two or three sensory or concrete details. What it smells like, sounds like, feels like in the hand, what happens the first time you use it. This is the detail a competitor selling a similar item can't copy-paste, because it's specific to how your product actually behaves.
Platform conventions and length. Etsy descriptions read like a person talking to a person, and the first 160 characters show up in search results, so lead with the strongest hook. Amazon listings are more structured, with keyword-forward bullet points doing most of the work. Shopify gives you full control, so it should match your brand's actual voice, not a marketplace's house style.
Tone and brand voice. Playful, minimal, technical, warm. Give the model three adjectives and, if you have one, a sentence from your own product page to imitate.
Bad prompt versus good prompt
Weak prompt: Write a product description for a candle.
Strong prompt: Write an Etsy listing description for a soy candle in a 9oz amber glass jar, scent is fig and cracked black pepper, hand-poured in small batches. The customer is buying this as a self-care gift for someone going through a stressful season, not for a housewarming. Mention that it burns 45 to 50 hours and the wick is cotton, no lead. Tone is warm and a little poetic but not flowery. Keep it under 500 characters for the listing body, and lead with the scent experience before the specs.
The weak version gets you three sentences about how candles make a room cozy. The strong version gets you an opening line about walking into a room that smells like a fig tree after rain, a middle section that earns the word "gift" by naming the occasion, and a close with burn time and wick material stated as fact instead of filler. The difference isn't polish. It's that the second one contains information the first one never gave the model to work with.
Second pair, this time for Amazon.
Weak prompt: Write a description for wireless earbuds.
Strong prompt: Write an Amazon bullet-point listing for wireless earbuds targeted at runners who currently deal with earbuds falling out mid-run. Specs: IPX7 waterproof, 8-hour battery per charge, 32-hour with case, ear hooks not just tips. Five bullet points, each starting with a capitalized benefit phrase, under 200 characters per bullet, and work in the keyword "wireless earbuds for running" naturally in the first and third bullets without repeating it back to back.
This produces bullets that open with the actual failure runners deal with (earbuds falling out) rather than a generic "crystal clear sound" claim every earbud listing makes. It also solves the keyword problem directly, which is the next part.
Getting a keyword in without stuffing
SEO-conscious sellers often make the mistake of asking the model to "include the keyword five times." That reads exactly as robotic as it sounds, and both Etsy's and Amazon's search algorithms weigh relevance and readability, not raw keyword density.
Instead, tell the model where the keyword should land and let it write around that constraint. For example: "Work the phrase 'organic cotton baby swaddle' into the title and once naturally in the first two sentences of the description. Do not repeat the exact phrase again after that; use variations like 'organic swaddle blanket' or 'breathable baby wrap' instead." This gets you the SEO benefit without the sentence that sounds like it was written by a keyword tool, because it was written by a keyword tool.
Iterating on the draft
Your first output is a draft, not a final. Treat it that way and you'll get better results with less frustration.
Once you have a version you mostly like, prompt for specific edits rather than regenerating from scratch:
Cut this by 30 percent and keep the sentence about the burn time, that's the strongest specific detail.
Rewrite the opening line to lead with the scent description instead of the material, that's the stronger hook for this platform.
Swap in more sensory language in the second paragraph. Right now it says "smells great," give me something more specific to fig and black pepper.
Naming often happens in the same sitting as describing, since both come up when you're launching something new; see how to prompt AI to name a product for that half of the job. Once the description is solid, you already have the raw material for an announcement post, and repurposing it is most of the work covered in how to prompt AI to write a LinkedIn post.
Each of these prompts targets one change. Asking for everything at once (shorter, punchier, more keywords, better opening) tends to produce a worse draft than three separate, specific passes.
Frequently Asked Questions
How do I get AI to write product descriptions that don't sound generic?
Feed it specifics it can't invent on its own: exact materials or specs, who the customer is and what problem they're solving, two or three sensory details, and the platform's format and length norms. A model given only a product name will fill the gaps with generic marketing adjectives, because that's all it has to work with.
What's the ideal length for an Etsy versus Amazon product description?
Etsy descriptions typically run 300 to 800 characters for the effective portion, since the first 160 characters show in search snippets. Amazon relies more on structured bullet points, usually five, each under 200 to 250 characters depending on category. Always tell the model the target length up front rather than trimming after the fact.
How do I include SEO keywords without it sounding stuffed?
Specify exactly where the keyword should appear (title, first sentence) and cap how many times the exact phrase repeats, then ask for natural variations after that. Asking a model to "include the keyword several times" produces exactly the repetitive copy that reads poorly to both shoppers and search algorithms.
Can I use the same prompt for every product in my shop?
You can reuse a template, but the specific details (material, customer, sensory language, keyword) need to change per product. A shared template with swapped-in specifics works well; a shared template with no specifics produces the same bland output for every item in your catalog.
The same specificity-over-genericity principle scales up to a full page: see our section-by-section scaffold for how to prompt AI to write a landing page, including a before/after example of generic versus specific AI copy.
How did this land?
About the author
Developer Advocate
Steve builds something with Swarmz every week and writes up what worked, what broke, and what he'd do differently. Tutorials and hands-on guides are his lane.
