Back when I first worked with large online product catalogs, one thing always slowed us down: the content bottleneck.
Thousands of SKUs sat online with missing or vague product descriptions, inconsistent tone across categories, generic use cases like “used for baking” or “kitchen essential,” and no keyword alignment for SEO. And every time someone asked, “Why is organic traffic not growing?” — we knew the answer. But fixing it at scale was another story.
Traditional options felt slow and costly: hire multiple freelance writers, create brand guidelines no one follows, or manually copy-paste outputs into product sheets. Fast forward to now. When faced with 5,000+ SKUs across diverse categories — from baking ingredients to homeware — I didn’t start with writers. I started with a prompt. And that changed everything.
Step 1 — Build the structure first
I didn’t prompt randomly. I began by defining exactly what each product description should include:
- H1 SEO-optimized title (with brand-safe keywords)
- Meta description (under 160 characters)
- Bulleted use cases (user-friendly, conversion-focused)
- One clear paragraph describing the product in simple, benefit-first language
Then I mapped the input data — SKU title, category, subcategory, keyword tags, ingredient info, material type. This became the foundation for the prompt.
Step 2 — Engineer the prompt to think like a copywriter
Next, I wrote a structured prompt that follows a fixed format (output ready to paste into Shopify or a CMS), reflects brand tone (clear, friendly, benefit-driven), avoids restricted claims (no health promises or competitor comparisons), and includes fallback logic (e.g., if no use cases are available, suggest based on category). The goal wasn’t to “replace” a copywriter — it was to equip AI to write like one, consistently and on brand.
Step 3 — Run at scale, review at speed
Once the prompt was tuned, I ran the product sheet in batches using the OpenAI API with Excel and Python, generated clean structured output for all 5,000 SKUs, and added human review checkpoints only where needed (e.g., high-traffic items or complex SKUs).
The result
Before: copy-paste chaos, SEO blind spots, inconsistent tone across categories. Now: SEO-ready descriptions for every product, structured metadata (H1 + meta) for every SKU, bullet points that speak to real user needs — all done in days, not months.
The real learning
AI didn’t replace the strategy — it accelerated execution without compromising tone, control or quality. This wasn’t just about rewriting copy. It was about rethinking the process — using AI as a system that turns complexity into structure, speed into consistency, and SKUs into searchable, sellable assets.