Product descriptions are one of the most underrated SEO and conversion levers. Unique, well-written descriptions improve your Google rankings, lift conversion, and help customers actually understand what they’re buying. The problem? Writing 800 of them by hand is brutal.

A wholesale client came to us with exactly this problem. Their Shopify store had 800 SKUs of industrial supplies imported from their supplier feed. Every product description was the manufacturer’s default text — dry, jargon-heavy, duplicated across thousands of other stores selling the same products. Google saw this as duplicate content and ranked them nowhere.

The traditional approach

A decent ecommerce copywriter charges R80-R150 per product description in South Africa. For 800 products at R100 each, that’s R80,000. Timeline: 3 months minimum, because no one writes 800 descriptions quickly.

The client didn’t have R80k or 3 months. They had a slow-bleeding SEO problem and limited budget. So we proposed an AI-led approach.

What we actually did

We built a small AI workflow using Claude (Anthropic’s API) and the Shopify API. Here’s the process at a high level:

1. Exported all 800 products via Shopify CSV — took 5 minutes
2. Built a prompt template that gave Claude the product title, key specs, target audience (wholesale buyers), brand voice notes, and instructions to write a unique 80-120 word description focused on benefits and use cases
3. Wrote a script that fed each product into the prompt one at a time, captured the response, and saved it back to a new CSV
4. Spot-reviewed the first 50 outputs — about 6 needed tweaks. We refined the prompt to fix those patterns
5. Ran the full batch overnight — 800 descriptions generated in about 4 hours, total API cost around R2,800
6. Quick human review of every description by the client’s team — around 30 needed manual edits. The rest went live as-is
7. Bulk re-imported the new descriptions via Shopify CSV

Why this worked (and when it doesn’t)

AI-generated descriptions work brilliantly when:

• You have a large catalogue of similar products
• Your products have clear, structured specs (size, material, weight, use case)
• You provide good brand voice guidance in the prompt
• You have a human review step at the end
• The competitive landscape uses generic manufacturer descriptions

AI is not the right answer when:

• Your brand voice is highly distinctive and idiosyncratic
• Your products require specialised technical knowledge to describe accurately
• You’re a luxury or storytelling brand where every word should be hand-crafted
• You can’t or won’t review the output before publishing

What we actually wrote in the prompt

The prompt structure mattered more than the AI model. We told Claude: who the customer is (wholesale buyers, not retail shoppers), what they care about (durability, bulk pricing, lead times, technical specs), the tone (professional, direct, no marketing fluff), the length (80-120 words), the structure (one benefit-focused opening sentence, then product details, then a closing line about ordering).

We also included three example descriptions in the prompt to anchor the style. This is called few-shot prompting and it dramatically improves output quality.

What this means for your store

If you have a large catalogue with weak descriptions, you’re sitting on serious SEO and conversion upside. The cost of fixing it with AI is now a fraction of what it was 18 months ago, and the quality is comparable to a mid-level copywriter.

It’s not magic. You still need someone to design the prompt, review outputs, and integrate it back into your store. But the economics are completely different from 2023.