Your store has 400 products. 60% have descriptions copied from the supplier, some in a foreign language, others incomplete, most with no SEO consideration at all. You know it should be better, but rewriting 240 descriptions by hand is something that never quite makes it to the top of the priority list.
This is the most common scenario in medium-sized e-commerce stores: inventory grows faster than the team can write quality content.
ChatGPT doesn't write as well as a specialized copywriter. But it writes considerably better than "Product description — imported" and it does it for a thousand products in the time it would take a human to do ten.
In this tutorial I'll explain how to build an automatic system that generates, reviews, and publishes optimized product descriptions at scale.
Why Product Descriptions Matter
Before talking about the automation, let's put in context why it's worth solving this:
SEO Impact: Product pages are the most indexed content in an e-commerce store. A well-written description with the right keywords can rank on Google for high-value purchase searches. A description copied from the supplier won't rank because hundreds of other sites have the exact same text.
Conversion Impact: E-commerce studies consistently show that product pages with detailed, benefit-oriented descriptions have 20-35% higher conversion rates than pages with minimal or generic descriptions.
Returns Impact: An accurate description that includes dimensions, materials, use cases, and limitations reduces incorrect expectations and, with that, returns.
The Prompt That Works
The quality of the descriptions ChatGPT generates depends almost entirely on the quality of the prompt. Here's the framework I use with my clients:
You are a copywriter specializing in e-commerce.
Write a product description optimized for SEO and conversion.
Product information:
- Name: {{product_name}}
- Category: {{category}}
- Technical specs: {{specs}}
- Price: {{price}}
- Brand (if applicable): {{brand}}
Description format:
1. Opening paragraph (2-3 sentences): Focused on the primary benefit for the buyer.
Include the main keyword: {{keyword}}
2. Key features list (5-7 bullet points): Convert each technical spec
into a benefit for the user. Use bullet points.
3. Closing paragraph (1-2 sentences): Call to action and who the product is ideal for.
Rules:
- Write in natural, professional American English
- Tone: informative but approachable, not corporate
- Avoid phrases like "high quality" or "excellent product" — they're generic and don't convert
- Total length: 150-250 words
- Don't repeat the product name more than 2 times
- Include the main keyword at least once naturally
Output: Only the description. No additional explanations.
Example output for a coffee machine:
Making café-quality espresso at home no longer has to be complicated. With the Express Coffee Maker Pro Model, you have full control over temperature, pressure, and extraction time to pull the perfect shot — in under 60 seconds from the moment you turn the machine on.
Key features:
- ☕ 15-bar pump pressure for professional-grade extraction at home
- 🌡️ Rapid-heat thermoblock: ready in 25 seconds
- 💧 1.5-liter water reservoir with visible level indicator
- 🫧 Integrated steam wand for frothed and steamed milk
- 🔧 51mm stainless steel portafilter (single and double baskets included)
- 🔌 1,450W with 110-220V compatibility
Ideal for anyone who drinks 1-4 coffees a day and wants a café experience without leaving home. Compatible with ground coffee and capsule adapters.
The Automation Workflow With n8n
To scale this to hundreds or thousands of products, you need the process to be completely automatic.
Option A: Process a CSV of Products
If you have your catalog in an Excel spreadsheet or Google Sheets, this is the most direct path.
- Upload your product CSV to Google Sheets with columns: name, category, specs, price, keyword
- In n8n, use the "Google Sheets Trigger" node to start when there are new rows or rows without a description
- For each row, the "OpenAI" node generates the description using the prompt
- n8n writes the generated description back into the sheet (column "generated_description")
- Review the descriptions → approve → publish
Option B: Direct WooCommerce Integration
For WooCommerce stores:
- Trigger: "WooCommerce Trigger" node activates when a new product is created
- n8n extracts the product's name, category, and specs
- OpenAI generates the description
- n8n automatically updates the product description field in WooCommerce
Option C: Direct Shopify Integration
The process is identical using the "Shopify Trigger" node and the "Shopify" node to update products.
Handling Products in Batches
If you have 500 existing products with no description, here's the process to handle them in batch without overwhelming the OpenAI API:
Workflow configuration:
- "Google Sheets" node: reads 10 products at a time (use the
limitparameter) - "SplitInBatches" node: processes them 1 at a time to control the pace
- "Wait" node: waits 1 second between each OpenAI call (avoids rate limits)
- "OpenAI" node: generates the description
- "Google Sheets" node: updates the row with the description and marks it as "processed"
At this pace, you can process approximately 200 products per hour without hitting API limits. 1,000 products in 5 hours.
Quality Control: How to Review Without Reading Everything
You shouldn't publish AI-generated descriptions without review, especially in the first batches. But reviewing 500 descriptions is unmanageable if you do it one by one.
Sampling-based review system:
- Generate the 500 descriptions and save them in Google Sheets with status "Pending review"
- Manually review a 10% sample (50 descriptions) distributed across categories
- If quality is consistent, activate automatic publishing of the rest
- Configure a problematic-word filter (the "IF" node in n8n checks whether the description contains terms you don't want to publish)
Warning signs to look for in the review:
- Descriptions with incorrect data (it invented specifications)
- Phrases in the wrong language or language mixing
- Excessive repetition of the product name
- Length outside the defined range (fewer than 100 words or more than 350)
For most standard products, GPT-4o mini has an error rate below 5% if the prompt is well-structured and the input data is correct.
Cost Optimization
GPT-4o mini is the right choice for this (not standard GPT-4o). The quality for product descriptions is practically the same and the cost is approximately 10x lower.
Estimated cost calculation:
- 1,000 products × 200 tokens per description = 200,000 output tokens
- GPT-4o mini cost: approximately $0.12 USD per 1M output tokens
- Total cost to generate 1,000 descriptions: ~$0.024 USD
In practical terms: for under $1 USD you can generate descriptions for an entire medium-sized catalog.
Variations for Different Product Types
The base prompt works for most products, but some require adjustments:
Clothing and fashion: Add instructions on how to describe the fabric (composition, texture, feel), the fit (regular, slim, oversized), and occasions for use.
Electronics: Prioritize compatibility (what devices it works with), warranty, and available technical support.
Food and supplements: Include information about ingredients, available flavors, how to use, and who it's intended for. This sector has specific marketing regulations — make sure the copy doesn't make unsupported health claims.
Industrial or technical products: The buyer is more technical. Descriptions can be more spec-dense and less emotionally benefit-oriented.
Need Help Implementing This?
If you have an e-commerce store with more than 100 products and want the system running without your team spending time on technical configuration, I can implement it for you.
The typical process is: one week to configure the workflow, one week of batch generation and review, and delivery of the complete catalog with optimized descriptions.
Free first conversation to review your catalog and give you a specific quote.