Before & After
- Product descriptions written by hand
- Hours per listing
- Inconsistent copy quality
- Listings missing SEO keywords
- AI writes descriptions from photos
- Seconds per listing
- Consistent optimized output
- Built-in SEO from day one
The Challenge
Stupel sells products online. Like most e-commerce businesses, they needed good product descriptions. But writing them by hand was slow. For every new product, someone had to look at the photos, figure out what to say, and write the copy. Then someone else had to check it. Then someone had to optimize it for search.
It could take 30-60 minutes per listing. When you have hundreds of products, that adds up fast.
The quality was also inconsistent. Different writers described the same types of products in different ways. Some listings were detailed and compelling. Others were thin. Customers couldn't tell what they were buying. That hurt conversions.
They needed a way to generate high-quality, consistent product descriptions — automatically, from the product images they already had.
What We Built
We built an AI workflow that takes a product image as input and outputs a complete, SEO-optimized product description. No human writing required. The system uses Amazon Rekognition to analyze the image and extract what it sees — materials, colors, features, dimensions — then passes that data to an AI language model that writes the description.
Why use image recognition instead of a template?
Templates produce generic output. They can't tell the difference between a blue cotton shirt and a blue linen shirt. Image recognition can. It pulls the actual details from the photo — the ones that matter to buyers and to search engines.
Generic copy doesn't convert. Specific copy does. That's the core reason this approach works better than any fill-in-the-blank template.
How does the AI write good copy?
We trained the system on what good product copy looks like. Short sentences. Feature-benefit pairs. Keywords woven in naturally. The output reads like it was written by an experienced copywriter — not a robot.
How fast does it actually run?
Each listing takes seconds. Upload a product photo, get back a full description ready to publish. A catalog that used to take weeks to write can now be processed in a day. That changes the math on launching new products.
Can it handle any type of product?
Yes. Amazon Rekognition is trained on millions of images. It works across categories — apparel, home goods, electronics, food, tools. The language model adapts its output based on what it sees in the image.
The Results
Project Details
Ready to automate your product listings?
We'll show you exactly how AI can handle your catalog — at any scale.