What I let Claude write, and what I made n8n compute

Description and use

An n8n pipeline using Claude to generate SEO pages, with validation, Slack approval, and GitHub publishing, cut page cost by 50%.

Problem

Automating the creation and review of SEO pages while minimizing cost and avoiding errors.

Outcome

The pipeline processes ~50 keywords/month, reduces AI cost per page from $0.23 to $0.11, and ensures only valid pages reach review stages.

Workflow

• Read new rows from Google Sheet
• Loop over keywords
• Send keyword to Claude API
• Parse Claude response
• Validate filename, slug, placeholders, memory entry, resource card, read-next block, internal links
• Send approved pages to Slack for SEO review
• If approved, open GitHub branch, commit page, memory, sitemap, resources, token log
• Open pull request
• Web developer merges or returns
• Update status in Google Sheet
• Handle errors: log cost, delete branch, alert via Slack
• Continue loop

Evidence

A schedule trigger reads the new rows from a Google Sheet. A Loop Over Items node processes keywords one at a time.

Source

A created page goes to the SEO specialist in Slack (send-and-wait form: approve or request changes).

Source

The change that halved the cost: ... output dropped by about two thirds, input context by about 30% (29,142 to 20,660 tokens), and cost per page went from $0.23 to $0.11.

Source