Most product descriptions written by ChatGPT in 2026 look like every other product description written by ChatGPT in 2026. Generic. Bloated. Sprinkled with the same banned words - "elevate", "harness", "embark", "vivid". They all sound like an MBA student trying too hard. Buyers can smell them across the room. Search algorithms can too.

This article is about the prompt structure that produces descriptions that actually rank and convert. Same model. Same $20 a month. Completely different output, because the prompt is doing the work that most beginners outsource to wishful thinking.

Why the default prompt produces garbage

The default prompt looks like this. "Write a product description for a leather wallet." The model produces 200 words of generic copy. It hits "premium materials", "stylish design", "perfect for everyday use", "elevate your style". You paste it into Shopify and wonder why conversion is the same as the placeholder text was.

The model is not the problem. The model is following exactly what you asked for - "write a product description". A generic prompt produces a generic output. To get specific, useful copy you have to give the model the same constraints a real copywriter would work under.

The five layers of a working prompt

Every good product description prompt has five layers. Skipping any of them collapses the quality.

Layer 1: the role. Not "you are an AI". "You are a senior e-commerce copywriter who has written 500+ descriptions for a leather goods brand in the $80-$200 price range." Specific role. The model takes on the constraints of the role.

Layer 2: the brand voice. Three to five sentences describing the tone. "Direct. Confident. Slightly understated. Refuses hype words like elevate, harness, vibrant. Reads like the founder writing personally, not like a press release." The more specific, the more the output sounds like a real brand and not a Shopify template.

Layer 3: the product specifics. Everything factual about the product. Materials, dimensions, weight, country of origin, who made it, what makes it different from the 200 other similar products on Amazon. Without these the model invents details. With them, it has truth to work with.

This is the layer most people skip. They paste in the SKU name and hope. Bad copy comes from bad input.

Layer 4: the audience. Who is the customer in one specific sentence. "A 32-year-old professional in a Tier-1 US city who buys quality goods rarely but well, and would rather own one $150 wallet than three $40 ones." Now the model is writing TO someone, not FOR a search engine.

Layer 5: the constraints. The hard rules.

  • Word count - exact range, e.g. "180-240 words"
  • SEO keyword and how many times to use it (3-4 times naturally)
  • Format - paragraphs, bullets, both
  • Banned words list - paste the human-writing-rules vocabulary
  • Required elements - opening sentence that does not start with "Introducing", "Discover", "Experience"
  • What not to do - no closing wrap-up sentence, no "perfect addition to any home"

The full prompt template

Combining all five layers, here is the structure that has produced thousands of converting descriptions in my own stores and across student stores.

You are a senior e-commerce copywriter who writes for premium-but-honest brands. You refuse marketing fluff. You write like the founder is talking to a customer in person. Brand voice: [3 sentences specific to this brand]. Product: [paste all product specs - materials, dimensions, what it does, who makes it]. Customer: [one sentence about who buys this and why]. Constraints: - 200-250 words - Use the keyword "[main keyword]" 3 times naturally - No em dashes, only short hyphens - Banned words: elevate, harness, vibrant, vivid, seamless, robust, transform, embark, journey, tapestry, essence, perfect addition, must-have, statement piece, curated, crafted, captivating, breathtaking - Open with a concrete sentence about a real use case, not "Introducing X" or "Discover Y" - No closing wrap-up sentence - Mix sentence lengths - vary rhythm - Default to prose. No bullet points unless I asked for them. Write the description now.

This prompt is 6x longer than the default "write a product description for X". The output is 10x better.

The before-and-after

Default prompt: "Write a product description for a small-batch leather wallet."

Default output: "Introducing our premium leather wallet, crafted from the finest materials to elevate your everyday carry. Made with stunning attention to detail, this wallet seamlessly combines style and function. Discover a vibrant accessory that transforms your outfit. The perfect addition to any modern wardrobe..."

You can read that and feel your conversion rate dropping in real time. Banned words on every line. Generic "perfect addition". No actual information.

Full-template prompt output: "The wallet holds six cards, two folded bills and a few coins, which is roughly the contents most people in this city actually carry. We make it from full-grain leather sourced from a tannery in Tuscany that has been running since 1957. Each piece is cut and stitched by one of three leatherworkers in our small Florence workshop, hand-finished and stamped with the maker's initials inside the back fold. The leather darkens with use. Six months in, your wallet will not look like anyone else's wallet, which is the whole point. Designed to be inherited, not replaced. Available in three colours - cognac, black, and a green that is closer to dark forest than to anything we want to call a name..."

That second one converts. Not because the model is better. Because the prompt did its job.

The prompt vault concept

Smart operators do not write a new prompt every time. They build a prompt vault - one tested template per category, with placeholders for product specs and brand voice.

Then for each new product, fill in the placeholders, paste into Claude or ChatGPT, get the draft. Spend 5-10 minutes editing the draft. Total time per description: 15 minutes. Output: copy that ranks and converts.

The first template takes 2-3 hours to dial in. Run it on 10 products, see what works, iterate. By the 11th product the template runs on autopilot. By the 50th product you have a system that scales as fast as you can take photos.

For different platforms - small adjustments

The template above is for Shopify or your own store. Amazon and Etsy need small tweaks.

For Amazon: keep the description shorter (150-200 words for the description, with the heavy lifting in the bullets and A+ content). Add a constraint - "must include the keyword in the first 15 words". Amazon's algorithm rewards keyword-near-start more than Etsy's.

For Etsy: lean into the story. Etsy buyers want a sense of who is behind the product. Add a constraint - "include one sentence about how this product is made or who makes it". Reduces buyer hesitation about handmade quality.

For SEO-driven blog content (which this very article is): keep the human-edited finishing touches. Pure AI blog content gets penalized by Google in 2026. Use the prompt to draft, then edit for voice and add 1-2 personal anecdotes.

The keywords tell you what must be in there. The style rules tell you how to say it. Both hold at the same time. Do not paraphrase a keyword into something prettier if that loses the search term.

Testing whether your description is working

Two signals. Conversion rate on the listing (sales divided by visits). Click-through rate from the search results (if the algorithm shows your listing, are people clicking).

If CTR from search is low, the title and main image are the issue, not the description. Fix those first.

If CTR is fine but conversion on the listing is low, the description and photos are doing the work. Test description variations. A/B testing in Amazon is built in. On Etsy and Shopify it requires more setup.

Run two descriptions for two weeks. Pick the winner based on numbers, not preference. The descriptions you love are not always the descriptions that convert. Trust the data.

The mistakes to avoid

Using the same description across platforms verbatim. Each platform weights different things. Adapt the template per channel.

Generating 50 descriptions in one sitting without editing. The model drifts in quality over a long session. Better to do batches of 10, edit each, then start a fresh session for the next 10.

Treating the AI output as final. The last 10-15% of editing is what separates good copy from great copy. Add the personal voice the model cannot produce.

Stuffing keywords for SEO. The 2026 Amazon and Etsy algorithms penalize keyword stuffing. The natural use of the keyword 3-4 times is the right dose. More than that hurts ranking, not helps.

For the broader AI stack across categories, read the complete AI stack for e-commerce. For the Amazon-specific listing playbook, see how to write Amazon listings that convert. The full prompt vault, including templates for every category, lives in the AI module of the course. Save this prompt. Edit the role, brand, product, customer, constraints. Run it on your worst-performing listing today.