automation

I Built a “Blog Post Machine” in N8N (and Yeah, It Actually Works)

Marty Bostick

17 Jan 2026 • 5 min read

I’ve got a confession: I love writing… but I hate the process tax that comes with it.

You know the tax. The “open 27 tabs, skim three half-baked articles, forget what you were doing, then rewrite the intro five times” tax. It’s brutal. And if you’re trying to publish consistently without turning into a content zombie, it’s basically a recurring subscription you never asked for.

So I ran an experiment in the Amplify AI Skool community: could I build an automated blogging workflow that goes from a simple title + description to a fully drafted, on-brand post in minutes?

Turns out… yeah. It works. And it’s kind of ridiculous.

The lightbulb moment: why am I doing this manually?

This started the way most of my “projects” start: I watched a few no-code automation demos and immediately thought, “Wait… why are we still doing this the hard way?”

I’d been seeing Gumloop-style workflows floating around, where you chain research → structure → writing → output. And I realized the only reason most people aren’t doing this is because they haven’t stitched the pieces together yet. Not because it’s impossible.

So I built it in n8n (open-source automation tool, super flexible), and I used a combo of Perplexity / Google search for research and ChatGPT (plus Grok/Claude in tests) for generation. The goal was simple:

Input: blog title + short description
Output: a Google Doc draft that sounds like me, includes structure, and is SEO-aware
Time: under 2 minutes for the machine, 10–15 minutes of human polish

And no, it’s not “hit publish and pray.” It’s “let the robot do the grunt work so I can do the human work.” Big difference.

What the workflow does (in plain English)

Think of this like building a tiny content team that lives inside an automation:

Research assistant that gathers context
Editor that enforces structure and voice
Writer that drafts the actual post
Ops person that drops it into Google Docs ready to tweak

When it works, it feels like cheating. When it doesn’t, it’s usually because you didn’t give it enough context (same as a real team, honestly).

Step-by-step: how I wired this “blog post machine”

Here’s the exact flow I built and iterated on live during testing.

1) A simple form kickoff (don’t overthink it)

The workflow starts with a basic input form:

Title (example I used: “Automate Your Blog Post Workflow”)
Description (just a few lines—enough to aim the draft)
Homepage URL (so it can learn your vibe)
Optional keywords (for SEO direction)

This is important: the title alone is rarely enough. The description is the difference between “generic AI soup” and “actually useful draft.”

2) Web research: steal like a chef, not like a pirate

Next, the workflow pulls in real-world context:

It scrapes your homepage (or a doc) for tone/voice cues
It runs a Google search on the title and collects top snippets/results

Why do this? Because writing without context is how you get posts that sound confident and say nothing. Research grounds the draft in what’s trending and what competitors are saying—without you spending an hour doing it manually.

3) The “prompt sandwich”: inputs that actually matter

All of that gets fed into a custom prompt. The key ingredients:

Title + description
Top 5–10 search snippets
My tone doc (“energetic, actionable, founder-first voice”)
Keywords to rank for

Then the model generates a structured draft: intro, headings, step-by-step sections, and calls-to-action. Not a wall of text. Not a weird poem. An actual blog post skeleton with muscle on it.

I tested multiple models (ChatGPT, Grok, Claude). My take: they’re all usable, but they have different personalities. Grok gave me a punchier voice in some runs, while ChatGPT was more consistently structured. Use whatever gets you closest to your “default voice.”

After the first few tests, I added layers that made the workflow way more “ready to ship”:

Keyword extraction (so the post naturally includes terms you want to rank for)
Image generation prompts (so featured images don’t become an afterthought)
Social snippets (FB/IG/short-form versions generated from the draft)

This is where it starts feeling like a content factory—in a good way. Same core idea, multiple outputs, minimal extra effort.

5) Output to Google Docs (the “handoff” point)

The final draft lands in a Google Doc. That’s my handoff point. I read it, fix any weirdness, add personal stories, tighten the hooks, and publish.

Total human time: 10–15 minutes instead of hours.

And honestly? That’s the sweet spot. I don’t want zero involvement. I want high leverage.

The results: 80–90% drafts and 5x output

Here’s what surprised me most: the first drafts were consistently 80–90% there. Not perfect, but strong enough that polishing felt like editing—not rescuing.

I also scaled it to batch multiple posts in one run (five posts queued up like it’s nothing). That’s where you start to feel the compounding effect: consistency becomes a system, not a mood.

What broke (and what I changed)

Let me save you some pain. A few things needed tuning:

Model selection matters: some models nail tone, others nail structure. I swapped to Grok in some runs for a punchier voice.
Length control is real: “Write 1500 words” doesn’t always translate cleanly. Better prompts + section targets helped.
SEO and images need explicit steps: if you don’t add nodes for metadata and image prompts, you’ll forget them every time.

This is the same principle as building any automation: if a step is “optional,” it will eventually disappear. Automate the boring but critical stuff.

Where this goes next (and what I’d build if I had your business)

My next upgrades are pretty straightforward:

Auto-publish to WordPress (with a review gate—because I’m brave, not reckless)
Internal linking (pull from past posts and suggest links automatically)
A/B social variants (different hooks for different platforms)

If you run a business that depends on inbound traffic, this is the move. Not because AI is magical, but because systems beat motivation. Every time.

My stance: this is how creators avoid burnout

I’m going to say the quiet part out loud: most “consistent content” advice is unrealistic for founders and operators.

We’re not lazy. We’re busy. And writing is high-energy work.

This workflow doesn’t replace your brain—it protects it. It turns content from a big scary project into a repeatable pipeline. Like meal prep, but for marketing. (And yes, it’s still annoying sometimes, but at least you’re not cooking every meal from scratch.)

Actionable takeaways (do this next)

Start with a form: title + 3–5 sentence description + keywords. Don’t skip the description.
Create a “tone doc”: a one-pager that describes your voice + examples of your writing. Feed it to the model every time.
Add research inputs: pull top search snippets so the draft isn’t floating in space.
Output to a reviewable place: Google Docs is perfect for human polish.
Automate the extras: image prompt + social snippets should be part of the workflow, not an afterthought.
Keep a human gate: edit before publishing. Always.

If you want, drop a title idea and a one-paragraph description, and I’ll tell you how I’d structure the prompt (or where your workflow will likely break). That’s the fun part.

Sources

n8n (workflow automation platform)
Perplexity (AI search and research)
Amplify AI Skool post: Automated Blogging Workflow Experiment