Bad Prompt vs Good Prompt: How to Get Better Product Descriptions from AI
In my experience, most ecommerce store owners who try to use AI for product descriptions come away disappointed. The output is usually flat, generic, and sounds like it was written by someone who’s never seen or held the product. Then they blame the AI for giving poor descriptions but the real problem most of the time is the prompt.
Quite simply, a bad prompt gets you bad copy. This stands for both product descriptions and written copy like blog posts. That’s not a flaw in the tool, it’s just how it works. Unless you tell it, AI doesn’t know your product, your customer, your brand voice, or where the description is going to live. If you don’t specify this information, it guesses. This is where you get your bog standard boilerplate sounding descriptions such as “premium quality” and “elevate your style”, essentially, descriptions you could swap between any two products without anyone noticing.
This post is specifically about description, which is one of four principles I use when working with AI (alongside delegation, discernment and diligence). I learnt these when completing my Anthropic certification for AI Fluency. Getting your description right is where most people fall down, and it’s where the biggest gains are. I’ll show you real before-and-after examples using actual products, so you can see exactly what changes and why it matters.
What A Bad Prompt Actually Looks Like
Here are two prompts I see regularly, or versions of them:
- “Write a description for a pair of Adidas Samba trainers that are black and white.”
- “I need a product description for a new brand of coffee that I’ve made.”
Both of these will produce something. ChatGPT or Claude won’t refuse to write it for you. They’ll write you a description and hand it back as if the job’s done. Some models will ask a follow-up question or two depending on what model you’re using, but even then they’re working with barely enough to go on. You’ll get something usable, but you’ll spend more time editing it into shape than it would’ve taken to write it yourself.
The problem isn’t prompt length. It’s the information that you put into it, or lack of it to be precise. A bad prompt leaves out almost everything the AI actually needs.
The difference between a bad prompt and a good one isn’t about length — it’s about what information you give the AI to work with.
Why AI Struggles Without The Right Information
Think of it like briefing a freelance copywriter you’ve never met. If you email them “write a description for black and white Adidas Sambas,” they’d come back with a list of questions to ensure they fulfilled the brief and didn’t have to rewrite it.
- Who’s the customer?
- Where does this description go?
- How long should it be?
- What’s the tone of your store?
- Should it have bullet points?
- Should it have headers?
AI doesn’t always ask those questions, and when it does, it doesn’t always ask the right ones. You need to front-load the brief with as much of this as possible.
There are five things a good product description needs to do its job properly:
- Tell the AI exactly what the product is and where to find more about it
- Define who it’s being written for (age, mindset, what they care about)
- Specify the length and format (words, bullet points, subheadings or not)
- Set the tone and voice
- Tell it where the description will be used and how it should end
If you leave out any of those, you’re leaving it to chance. Sometimes you might get lucky but more than often you get something you’d be embarrassed to publish, doesn’t resonate with the target audience or is factually inaccurate.
This is closely related to how AI SEO mistakes happen in ecommerce too, with good tools badly instructed, that produce content which looks fine on the surface but doesn’t serve the store. The prompt is where it goes wrong.
Example One: Nike Air Max 90 (streetwear, younger audience)
Let’s take a product that sits in the streetwear and trainer culture space. The Air Max 90 has decades of heritage behind it, and if you sell it, you’re talking to a customer who knows their trainers. A generic description won’t cut through.
The bad prompt:
“Write me a product description for a Nike Air Max 90 in black and white.”
You’ll get something like:
“The Nike Air Max 90 is a classic trainer featuring a comfortable fit and iconic Air cushioning. Available in black and white, this versatile shoe is perfect for everyday wear.”
It’s not wrong. It’s just useless from a conversion and SEO point of view. You could replace “Nike Air Max 90” with almost any trainer and that description would still technically work. There’s no reason for someone to buy it over any other option on the page.
You’re leaving it up to the AI to presume:
- Gender it’s aimed at
- Age range (is it kids, or adults)
- What the materials are (is it leather with suede overlay panels or completely leather?)
The better prompt:
“I need you to write a product description for my Shopify store, which sells designer and streetwear trainers. The product is a Nike Air Max 90 in black and white. The colourway is black upper with white midsole and the iconic visible Air unit in the heel. You can find the product at [URL]. Do not copy that description; it must be unique. Write it for men aged 18 to 30 who are into trainer culture and streetwear. They know what the Air Max 90 is, so don’t over-explain it. Speak to them like they already get it. The tone should be confident, cool, and direct. No corporate language. The description should be around 150 to 200 words. After the main copy, add a bullet point list under the subheading ‘Key Details’ with specs such as colourway, sole unit, upper material and fit notes. End with the anchor text ‘Shop all Nike Air Max’ and I’ll link it to the Nike collection page. Write in British English with no em-dashes.”
That prompt takes about 90 seconds to write. The output it produces is a different category of content entirely. The AI knows the customer, the tone, the format, the length and where the description is going. There’s nothing left to guess.
Example Two: Belstaff Roadmaster Jacket (premium, mature audience)
Now a different product, different customer, different everything. The Belstaff Roadmaster is a waxed cotton jacket with serious heritage. The person buying it isn’t 21 and scrolling TikTok. They’re probably in their late 30s or older, they care about quality and provenance, and they’re spending serious money.
The bad prompt:
“Write a product description for a Belstaff Roadmaster jacket in black.”
You’ll likely get heritage mentioned. Maybe motorcycles gets a mention. Probably “iconic” (one of the most overused words in fashion copy). The description will be fine, in a completely generic, forgettable way. But it’s not necessarily targeting the right people with any kind of persuasive copy.
The better prompt:
“I need a unique product description for my Shopify designer menswear store. The product is the Belstaff Roadmaster jacket in black, with a waxed cotton outer finish. The same product can be found at [URL] – please use this to understand the product fully, but do not copy the description. It must be original. Write this for men aged 35 to 55 who buy quality outerwear and understand the value of a well-made jacket. They’re not looking for a fashion statement only. They want something that will last and that has real craft behind it. Tone should be authoritative but not stuffy. Write as though you know this product well. I need at least 250 words for the main description. Follow it with bullet points under the subheading ‘Key Details’ covering the outer material, lining, fit, and care instructions. Close with the anchor text ‘Shop the full Belstaff collection’ and I’ll link it through to the Belstaff brand page.”
It’s really the same structure as the first prompt, but completely different in tone, audience and intent. That’s the whole point. The framework stays consistent but what changes is who you’re writing for and what matters to them. This allows the AI to give you copy that will be more informative and persuasive.
You can read more about how this kind of thinking applies to your wider SEO approach in my post on ecommerce SEO in the AI age, which covers how AI search has changed what good content actually needs to do.
Frequently Asked Questions
Will AI always ask me what it needs if my prompt is too vague?
I won’t do it reliably. I find that some models will ask follow-up questions, others will make assumptions and write anyway. Either way, the output suffers. A detailed prompt from the get-go gets you better results than waiting to see what the AI asks for.
Does the same prompt structure work across different AI tools?
Yes. Whether you’re using Claude, ChatGPT, Gemini or any other tool, the same principles apply — product detail, audience, tone, format and close. The models handle prompts slightly differently but the information they need is the same.
Should I give the AI a URL to the product page?
If the model supports web browsing, yes. It gives the AI access to specs, images, and existing copy. Just always include the instruction not to copy the existing description — you want it to use the page as a reference, not a template.
How do I stop AI product descriptions sounding generic?
Two things: give it a specific audience and give it a specific tone. “Write for men aged 35 to 55 who buy quality outerwear” produces different output than “write for adults who like jackets.” The more specific your brief, the more specific the output.
Can I use this approach for bulk product descriptions?
Yes, with one adjustment. Build a master prompt template with all the consistent elements (store type, tone, format, close) and then swap out the product-specific details for each item. You can also ask the AI to write descriptions for several similar products in one session if you give it the full spec for each one.
Has AI made a mess of your store’s content?
If bad prompts have left you with thin copy, duplicate descriptions or content that’s hurting your SEO rather than helping it, I can help you fix it.