DEV Community

Cover image for How I Automated News Discovery with n8n (439K Views, 0€ Spent)
Muhi
Muhi

Posted on

How I Automated News Discovery with n8n (439K Views, 0€ Spent)

#automation #selfhosted #programming #ai

I'm building https://AKCACHE.io, a managed database service focused on EU digital sovereignty. My marketing depends on catching relevant EU policy news early, but manually scanning news sites daily was killing my productivity.

Then I got lucky. I found a German government report about US cloud access and posted it to r/europe. The result: 439K views, 2.9K upvotes, 229 comments - all organic. Zero ad spend.

Link to the post

I've spent money on ads before. To get even 50K impressions would cost thousands of euros. This single post drove more qualified traffic than months of paid campaigns.

The problem: I can't manually hunt for articles like this every day. So I did what any developer would do: I automated it.

The Solution: n8n News Discovery Pipeline

I already host n8n on my server, so the setup cost was €0. Here's how it works:

Step 1: Aggregate News Sources

The workflow pulls from multiple RSS feeds every 6 hours:

  • heise.de (German tech news)
  • Ars Technica
  • Euractiv (EU policy)
  • EUobserver
  • TechCrunch
  • The Verge
  • Google News RSS (custom search query)

RSS-Node

Step 2: Filter by Keywords

After merging all sources, I filter for the last 6 hours (since it runs every 6 hours) and use JavaScript to check for relevant keywords:

const keywords = [
  'cloud act', 'fisa', 'digital sovereignty',
  'gdpr', 'eu cloud', 'data residency',
  'schrems', 'gaia-x', etc.
];
Enter fullscreen mode Exit fullscreen mode

This cuts down ~100 articles to maybe 10-15 relevant ones.

Step 3: AI Relevance Scoring

Each article that passes keyword filtering goes to GPT-4o-mini for analysis:

  • Score 1-10 for relevance
  • Extract key topics
  • Determine urgency (high/medium/low)
  • Suggest post angle

Cost consideration: I use GPT-4o-mini for cost efficiency, but watch the Token Per Minute limit. That's why I added a 1-minute wait between API calls.

Step 4: Store and Notify

Articles scoring 7+ get saved to PostgreSQL and sent to me via email with:

  • Title and score
  • Why it matters
  • Suggested post hook
  • One-click link to generate draft

Generated E-Mail

Step 5: Auto-Generate Reddit Posts (Human-in-Loop)

This is where it gets interesting. Each email has a clickable link that triggers a second workflow via GET request. When I click:

  1. AI fetches the full article
  2. Generates Reddit post title and body
  3. Naturally mentions my product where relevant
  4. Returns formatted post ready to copy/paste

I intentionally don't auto-post. I want automation for the tedious research part, not the human judgment part.

n8n workflow

The Results

Before automation:

  • 30+ minutes daily scanning news
  • Missed 80% of relevant articles
  • Inconsistent posting

After automation:

  • 2 minutes daily reviewing AI-filtered results
  • Catch articles within 6 hours of publication
  • Consistent content pipeline

Best part: The whole flow runs on my existing server. No SaaS subscriptions, no external dependencies.

Technical Details

Workflow triggers:

  • Schedule: Every 6 hours starting at 00:00
  • Webhook: Manual trigger via URL for generating posts

Stack:

  • n8n (self-hosted)
  • PostgreSQL (article storage)
  • OpenAI GPT-4o-mini
  • Standard email node

You can adapt this for any use case - just change the RSS sources and keywords to match what you're tracking.

Questions? Ask in the comments.

Top comments (0)