AI Content Agent for Ecommerce Catalog

Challenge

An ecommerce store with several thousand SKUs had product pages that ranged from a single sentence to a copy-pasted supplier description. Search rankings were poor, conversion rates on cold traffic were low, and the marketing team had a backlog of content work they would never get through manually. Writing one good product page takes 20, 40 minutes; writing a thousand takes months.

Options Considered

1. Freelance copywriters, quality was good but cost per SKU was high and turnaround slow. Not viable for ongoing catalog growth.
2. Generic AI writing tools (Jasper, Copy.ai), fast, but disconnected from actual store data. Output required heavy editing to fix factual errors and align with brand voice.
3. Custom AI agent with direct WooCommerce API access: chosen. The agent reads product attributes, category context, and competitor positioning directly from store data, producing accurate, on-brand content with minimal review.

Decision

The agent pulls product data via WooCommerce API, title, attributes, category, price tier, existing description, and generates a full content set: long-form description, short description, SEO title, meta description, and internal link suggestions to related products. A brand voice guide is embedded in the system prompt. Output is staged for review before publishing or pushed directly for low-risk SKUs.

Implementation

A Python orchestration layer batches products by category and runs generation in parallel using GPT-4. Each batch uses a category-level prompt that adds context about the target customer and competing products. The output is structured as JSON matching the WooCommerce fields, title, description, short_description, yoast_seo, and pushed back via the REST API. A review queue lets the team approve or edit before publish.

Internal linking is generated by embedding all product titles and finding the top-k semantically related products per page, surfaced as suggested anchor text and target URL pairs for the editor to insert.

Outcome

500 product pages updated in the first run, cutting estimated manual effort from 3 months to 2 days. Organic search impressions increased 40% within 60 days. Average description length increased 4× with no increase in bounce rate.