I automated a news site with n8n: from 20 RSS feeds to WordPress drafts

Description and use

An n8n workflow automates ingesting 20 RSS feeds, deduplicating, scoring relevance with a cheap LLM, rewriting selected stories with a tiered LLM, attaching an image, creating a WordPress draft, and sending a Telegram alert, leaving only editorial review.

Problem

Manually reviewing dozens of RSS feeds, scoring relevance, rewriting stories, and creating WordPress drafts was time‑consuming and repetitive.

Outcome

The pipeline reduces manual effort by automating feed ingestion, deduplication, relevance scoring, rewriting, image attachment, and draft creation, leaving only editorial review. It runs every three hours with minimal token spend by filtering early.

Workflow

• Schedule trigger every 3h
• Read list of RSS feeds
• Split feeds into individual items
• Cap to 3 items per feed (Code node)
• Hash each link (SHA256)
• Insert into Postgres with ON CONFLICT DO NOTHING to deduplicate
• If new, scrape article with Firecrawl
• Score relevance 0‑100 with cheap LLM
• If score ≥70, rewrite with tiered LLM (Claude if ≥85, else Gemini)
• Fetch cover image from Unsplash
• Create WordPress draft with cover image
• Mark as published in Postgres
• Send Telegram alert

Evidence

Deduplicate before spending: hash the link and let Postgres reject repeats with ON CONFLICT DO NOTHING.

Source

Use a cheap model to score relevance from 0 to 100 before rewriting: only what clears the threshold reaches the expensive model.

Source