Building a Content Machine with N8N and Claude
I automated my entire content pipeline. Here's the N8N workflow, the Claude prompts, and how it runs without me.
I check my content pipeline maybe twice a week. Everything else runs on its own.
Content ideas get generated, researched, drafted, queued for review, and scheduled without me doing anything until the review step. That step takes me 20 minutes every couple of days. Everything before it is automated.
Here’s exactly how I built it.
The Problem I Was Solving
I was the bottleneck in my own content operation. Good ideas would come up - from a customer call, a Twitter thread I read, a question someone asked in my newsletter replies - and they’d die because I didn’t have time to act on them immediately.
By the time I had time, the moment had passed or I’d forgotten why the idea was interesting. My content calendar was a graveyard of orphaned ideas.
I needed a system that could capture ideas, develop them, and hand me something that was 80% done. The last 20% - the judgment, the voice, the opinion layer - that’s still mine.
The Architecture
Before I show you the workflow, here’s the shape of the system:
- Ideas come in from multiple sources and land in one place
- A Claude prompt turns the idea into a structured brief
- Another Claude prompt drafts the post from the brief
- The draft gets routed to a review queue
- After I approve, it gets scheduled and published
Simple on paper. The N8N workflow is what makes it automatic.
Step 1: The Idea Inbox
I have a Telegram bot (built with N8N) that I text ideas to whenever they come up. Voice notes, links, one-line thoughts. Anything that might become content goes to that bot.
The N8N trigger watches for new messages to the bot. It extracts the content and writes it to a Google Sheet - my idea backlog. That’s the first node in the workflow.
I also have N8N pulling from other sources on a schedule:
- RSS feeds from industry sites I track (any article that gets high engagement goes into the backlog as a potential reaction piece or counterpoint)
- My own newsletter replies (I have a Gmail filter that tags responses with questions in them, N8N watches that label)
- A simple web form on my site where I can quickly type ideas from any device
Everything hits the same Google Sheet. One inbox, multiple sources.
Step 2: The Brief Generator
Once a day, N8N looks at the Google Sheet and picks ideas that are marked “ready to brief” - I manually do a 5-minute triage once a day to mark which ideas are worth developing.
For each ready idea, N8N sends it to the Claude API with this prompt:
“You are a content strategist. Turn this idea into a detailed blog post brief. Include: 1) The specific angle (not just the topic - the specific take), 2) The target reader and what they already know, 3) The core argument in one sentence, 4) 5-7 subheadings with 2-sentence descriptions of what each section covers, 5) 2-3 specific examples or data points I should look up, 6) Suggested internal links if relevant.
Idea: [idea text] Site context: [brief description of site voice and audience]”
The output is a structured brief that goes back into the Google Sheet next to the original idea.
Step 3: The Draft Machine
When I mark a brief “approved for draft” (another manual step, takes 2 minutes), N8N kicks off the drafting workflow.
It sends the brief to Claude with a longer prompt that includes my full voice guide. That guide is 400 words describing how the site sounds, what phrases it uses, what it never says, who the reader is. I wrote it once and it lives in an N8N variable that gets injected into every drafting prompt.
The draft prompt structure:
“You are writing for [site]. Here is the voice guide: [voice guide]. Write a blog post based on this brief. Follow the subheadings in the brief. Write in first person. Do not use em-dashes. Do not use filler phrases. Short paragraphs. Start with a hook that doesn’t begin with the subject of the article.
Brief: [brief]”
The draft goes into a shared Google Doc that follows a naming convention I can scan quickly.
Step 4: The Review Queue
N8N creates a task in my project management tool (I use Linear, but you could use Notion, Asana, anything) with a link to the Google Doc. That’s my review queue.
I open these in batches every couple of days. For each one I:
- Read it for voice issues
- Add my experience layer - specific things I’ve seen, results I’ve got, opinions the AI was too cautious to take
- Check that the intro hook actually lands
- Fix any section that’s too generic
This takes 15-20 minutes per post. Maybe 30 if it needs more work.
Step 5: Scheduling and Publishing
After I approve a post in Linear, N8N watches for that status change and moves the post into my publishing queue.
For a CMS like WordPress or a headless setup, N8N can publish directly via API. For my current site, it creates a formatted Markdown file and pushes it to a GitHub repo, which triggers a deploy.
The scheduling logic in N8N looks at my publishing calendar (another Google Sheet) and picks the next open slot. Posts go out at 9am on the site’s top traffic days based on historical data.
The Numbers
Building this took about 3 weekends spread over a month. The ongoing time cost is:
- 5 minutes/day for idea triage
- 30-60 minutes every 2 days for brief approval and draft review
- Essentially zero for everything else
Before this system, I was spending 4-6 hours per post from idea to published. That’s now less than an hour of my time, and the pipeline handles 3-4 posts per week without me scrambling.
What You Need to Build This
- N8N (self-hosted or cloud - start with cloud, $20/month)
- Claude API access (not Claude Pro - the API, so you can call it from N8N)
- Google Sheets for the idea and scheduling backlog
- A CMS with an API or a GitHub-based site
- Telegram for the idea capture bot (optional but I love it)
The Part That Surprised Me
The hardest part wasn’t the technical setup. It was writing the voice guide. When I had to put into words exactly how my content sounds, what it’s willing to say, who it’s talking to - that forced clarity I didn’t have before.
If you’re going to build something like this, start there. Write the voice guide first. Everything else is just plumbing.
Want more like this?
Get the latest AI marketing and automation insights delivered to your inbox.
Subscribe to the Newsletter →