ai newsSa., 17. Okt.
EilmeldungDeepseek V4.1 Flash 1017 beats larger rivals in agentic tasks while once again massively cutting per-task costs

Turn Your Blog RSS into a Week of Social Posts with n8n and Any LLM API

Posting consistently is a distribution problem, not a writing problem. You already wrote the thinking (the blog post). This workflow turns each new post into platform-native drafts automatically, and a human reviews before anything goes out. The workflow Schedule (daily) -> Read Blog RSS -> Take ne

dev.toSamstag 10:45 · 17.10.2 Min. Lesezeit
Bild: dev.to

Posting consistently is a distribution problem, not a writing problem. You already wrote the thinking (the blog post). This workflow turns each new post into platform-native drafts automatically, and a human reviews before anything goes out.

Schedule (daily) -> Read Blog RSS -> Take newest post -> LLM draft -> Parse -> Save to Google Sheets.

1. RSS Read node with your feed URL. n8n's RSS node outputs title, link, contentSnippet per item.

2. Take the newest. A Code node: return items.slice(0, 1);

3. The LLM call. HTTP Request node, POST to your provider's chat completions endpoint. I use an OpenAI-compatible shape so you can swap providers (OpenAI, DeepSeek, Groq all accept the same body):

{ "model": "gpt-4o-mini", "response_format": { "type": "json_object" }, "messages": [ { "role": "system", "content": "You repurpose blog posts into social drafts. Return ONLY valid JSON with keys: x_post, linkedin_post, facebook_post. x_post: max 240 chars, hook first. linkedin_post: 80-150 words, hook on first line. facebook_post: 40-80 words, conversational." }, { "role": "user", "content": "Title:... Link:... <first 2000 chars of post>" } ] } Enter fullscreen mode Exit fullscreen mode Send the API key as an Authorization: Bearer header. Keep the key in your CONFIG node, not hardcoded.

4. Parse. A Code node: JSON.parse(items.json.choices.message.content), then shape one clean item with date, source_title, source_url, the three drafts, and status: 'draft'.

5. Save. Append to a Google Sheets tab with matching columns. Review the drafts over coffee, copy, post.

AI drafts are 80% there and 20% off-brand. The sheet is the review queue. This keeps your voice intact and avoids the classic automation embarrassment of a bot posting nonsense at 3am. Total human time: 10 minutes a day.

One post a day through a mini model costs cents. If you are watching every dollar, point the HTTP node at DeepSeek; same request shape, roughly a tenth of the price.

This is one of five workflows in my automation template pack ( lead capture, Stripe dunning, this repurposer, review drips, lead reactivation ), all with setup checklists and docs.

Quelle: dev.to (dev.to). Original öffnen