Automotive newsletter automation

Finding relevant reports from 16 consulting firms and the wider industry, removing duplicates, sorting them into chapters and writing German summaries took most of a working week every month.

An n8n workflow in three stages. Collect: scrape the insight pages of 16 named firms plus general automotive news from the last 30 days. Process: classify each article into one of eight chapters, score relevance with Gemini, drop duplicates and non-articles, then write German summaries and key insights. Generate: assemble the HTML newsletter for review and email it to about 100 consultants.

• Manual effort per issue went from about 25 hours to under 2 hours of review.
• In one recorded test run, 29 scraped items became 16 articles after removing 2 duplicates and 11 non-articles.
• Scoring rules and prompts are written down, so the client can tune relevance without editing code.

• Scraper with a rolling four-week window and a domain filter per firm
• AI classification into eight chapters, with competitor insights first
• Relevance scoring with documented definitions
• Deduplication across sources
• German summaries and key insights
• HTML newsletter output for human review before sending

Self-hosted n8n in Docker with PostgreSQL, JavaScript code nodes for deduplication and quality checks, and Gemini 2.0 Flash for classification, scoring and writing.

• Versioned code-node scripts and prompt files kept outside n8n for review
• Captured input and output of every stage for regression checks between versions
• Each rendered newsletter version compared against the last manually produced issue