Social Media Sentiment Monitoring Workflow Template: n8n Scheduled Scraping + LLM Sentiment Analysis + Alerts (Downloadable)

Once your brand grows, you won't know who's praising or trashing you. This n8n workflow periodically scrapes brand mentions across platforms, runs LLM sentiment analysis with a key-points summary, and pushes negative mentions to a group chat instantly-turning "passively taking hits" into "actively responding."

Workflow Chain

Scheduled trigger (hourly / daily) -> multi-platform mention scraping -> LLM sentiment analysis (positive / negative / neutral + summary) -> instant alert on negatives -> full archive + daily report.

Download Template

Usage Steps

1. Import: n8n -> Workflows -> Import from File, pick social-media-monitor.json
2. Schedule node: set scrape frequency (hourly for tight sentiment, daily otherwise)
3. Scrape node: configure per platform-Weibo / Xiaohongshu via RSS or third-party API; Zhihu / Douyin via search APIs; Twitter via API (mind domestic access)
4. LLM node: fill in API key; prompt outputs JSON (sentiment + summary + severity)
5. Alert node: severity=high pushes instantly to Feishu / DingTalk group with the original link
6. Archive node: all mentions into Feishu Bitable / database; generate a daily report
7. Test run: search one keyword and run, see whether sentiment classification is sensible

Companion Prompt (LLM Sentiment Analysis)

You are a sentiment analyst. Analyze this social mention about "{{brand}}" and output only JSON:
- sentiment: positive | negative | neutral
- severity: high (rant/complaint/PR crisis) | medium (gripe/question) | low (neutral mention/praise)
- summary: one sentence on what was said
- need_response: true (needs reply) | false
Basis: the original tone and content. If words like "scam, refund, complaint, report" appear or the tone is heated, severity must be high.
Brand: {{brand}}
Original: {{text}}

Pitfalls

• Unstable platform APIs: third-party scrape endpoints fail often. Use multi-source backups (RSS + API + search) so one source going down doesn't break the whole.
• Sentiment misjudgment: sarcasm and passive-aggression trip up the LLM (surface praise, actual mockery). Add a "watch for sarcasm" note in the prompt, and add a human second check for high-severity items.
• Alert fatigue: alert on everything and the group goes numb. Alert only on severity=high AND need_response=true; the rest go into the daily report.

References

• n8n official: n8n.io
• Feishu Bitable API: open.feishu.cn