RSS Aggregation + AI Summary Content Pipeline

The hardest part of running a content site isn't writing-it's scanning 36Kr, sspai, Jiqizhixin, Hacker News every day for topics. This n8n workflow template pulls multi-source RSS on a schedule, dedupes, runs each item through an LLM for summary + topic-value scoring, and auto-files high-score items into Feishu/Notion/Postgres. Turns "doomscrolling feeds" into "reading a ranked list".

Workflow Chain

Schedule trigger (hourly) -> parallel RSS Read nodes (36Kr / sspai / Jiqizhixin / Hacker News / TechCrunch) -> Merge -> Code node dedupes by URL stem -> Loop Over Items per-item LLM -> LLM summary + score (1-10) -> IF score >=7 -> write to Feishu Bitable / Notion / Postgres -> else drop.

Download Template

Setup Steps

1. Import: n8n -> Workflows -> Import from File, pick rss-content-pipeline.json
2. Schedule node: set hourly (topic timeliness matters-don't stretch to daily)
3. RSS Read nodes: one node per source, fill feed URL. 36Kr https://36kr.com/feed, sspai https://sspai.com/feed, Jiqizhixin https://www.jiqizhixin.com/rss, Hacker News https://news.ycombinator.com/rss
4. Merge node: combine multi-source items into one stream, mode = Append
5. Dedupe node: Code node strips URL query params then hashes; pair with Postgres/Redis of seen hashes-skip on hit
6. LLM node: fill API key (DeepSeek/Kimi/Qwen all work); 8k context is enough; send summary + score prompt per item
7. IF node: branch by returned score; >=7 goes to the write branch
8. Write node: Feishu Bitable via HTTP Request + OpenAPI; Notion via the official Notion node; Postgres via the Postgres node
9. Test run: trigger manually, check the table for high-score items

Companion Prompt (LLM Summary + Topic Scoring)

You are a topic editor for an AI content site. Read the RSS item below and output:
1. Summary: within 80 words, cover "what happened + why it's worth writing"
2. Topic-value score (1-10):
   - Timeliness (3): first published within 24h?
   - Heat (3): covered by multiple sources?
   - Writing room (2): extendable into tutorial/comparison/explainer?
   - Audience fit (2): matches AI-tools-in-practice readers?
3. Recommended angle: one sentence on the writing hook
Output JSON with fields summary / score / angle.
Don't fabricate facts not in the item; use only the item's own info for the summary.
RSS item title: {{title}}
RSS item body: {{content}}

Pitfalls

• Inconsistent RSS payload: some feeds give full content, some only a summary. In the Code node, normalize to content:encoded || content || summary, or LLM scores swing wildly
• Title-only dedupe: the same news re-titled across sources ("X launches Y" vs "Y officially debuts") slips through. Dedupe by URL stem (strip query, then hash)-far more stable than title
• LLM score drift: same item scores 8 today, 6 tomorrow. Give explicit scoring dimensions (timeliness/heat/writing room/audience) in the prompt-way more consistent than "score by feel"
• Don't forget rate-limiting on writes: hourly runs with many sources can flood hundreds of rows a day. Add a score >=7 IF filter, or your table fills up in days

References

• n8n RSS node docs: https://docs.n8n.io/integrations/builtin/core-nodes/n8n-nodes-base.rssfeedread/
• Feishu Bitable OpenAPI: https://open.feishu.cn/document/server-docs/docs/bitable-v1/bitable-overview
• Notion API create a page: https://developers.notion.com/reference/create-a-page
• RSSHub Chinese media feed aggregator: https://docs.rsshub.app/