How I Built an AI Blog Engine with Discord (And How You Can Too)

Let me tell you something embarrassing: I’m an AI consultant who spent most of 2024 not publishing consistently.

Not because I didn’t have things to say. I had notebooks full of half-structured thoughts. Drafts that got 80% done and then stalled. Post ideas that lived in Notion for months without anyone reading them. The problem wasn’t ideas — it was the gap between idea and published post.

Every consultant and agency owner I’ve talked to has the same problem. You know content matters. You know thought leadership drives leads. But sitting down to write a polished 1,200-word post after a full day of client work feels like being asked to run a 5K after already running a marathon. You have nothing left by the end of the day, and the blog post that was going to explain your methodology continues to not exist.

This is the problem I decided to actually solve instead of just complaining about.

Content Automation Is Not the Same as Content Outsourcing

Before I get into the architecture, I want to be clear about what this system is — and what it isn’t. It is not an AI that writes my blog for me. It’s an AI that helps me get my ideas out of my head and into publishable form faster, with a deliberate human approval gate before anything goes live. That distinction matters.

There’s a reason this series keeps coming back to human-in-the-loop systems: because fully autonomous content generation has a well-documented failure mode. It sounds like AI. The cadence is off. The opinions are generic. The examples are vague. Readers can tell. And if you’re building a reputation as a trusted advisor, AI-flavored blog posts are eroding that trust with every paragraph.

What I wanted was different: a system that captures my actual voice, my actual opinions, and my actual experience — without requiring me to cold-write 1,200 words from scratch every single week.

The Architecture: Discord as the Control Room

The system has four main components.

Discord is the command center and approval interface. I use a private Discord server with a few dedicated channels for content. When I have a thought worth turning into a post, I drop it in a #raw-ideas channel. That’s it — a voice note transcription, a bullet list, a half-formed rant, whatever I’ve got. No formatting required. The rougher the better, actually.

n8n is the workflow orchestrator. It watches the Discord channel for new messages and kicks off the content pipeline when it sees one. It handles all the plumbing: structuring the prompt, calling the LLM, extracting the draft output, and routing it to the right place.

An LLM — usually Claude — is the drafting engine. Given my rough notes plus a structured system prompt that encodes my writing style preferences, it generates a full article draft. Not my final words. A smart first draft that captures the shape of what I’m trying to say.

The approval loop is where I come back in. The draft lands in a #drafts-for-review channel, where I can read it on my phone while I’m waiting for coffee. I either react with a ✅ to queue it for publication, or I leave comments in the thread and it goes back through a revision pass. Nothing gets published without that ✅.

The whole pipeline is probably 200 lines of n8n workflow configuration and an 800-word system prompt. It’s not complicated. It took a Saturday afternoon to build the first version, and I’ve been iterating on it since.

Building the System Prompt That Doesn’t Sound Like AI

This is the part most guides skip, and it’s the most important part. The quality of your AI blog engine depends almost entirely on your system prompt. If you feed the LLM “write a blog post about AI automation,” you’ll get a generic, AI-flavored blog post. If you feed it detailed context about your voice, your audience, your opinions, and the specific thing you’re trying to say — you’ll get something that sounds like you.

My system prompt includes:

• A description of my voice: conversational, direct, uses concrete examples, has genuine opinions, occasionally references 80s and 90s pop culture
• A description of my audience: SMB owners and operations leaders who are skeptical but curious about AI
• A content template: hook → problem → solution → personal example → actionable takeaway
• A list of things I never do: generic platitudes, vague recommendations, leading with “In today’s fast-paced world”
• A sample section from a post I’m proud of, as a style reference

That last item is worth emphasizing. Showing the LLM an example of the output you want produces dramatically better results than describing it in the abstract. Think of it like showing a contractor a finished room you love instead of trying to verbally describe the aesthetic. The example does the work that a thousand adjectives can’t.

From Discord Message to Live Post in Under an Hour

Here’s what the actual workflow looks like:

1. I drop a rough idea in #raw-ideas. Something like: “Post idea: content automation ≠ outsourcing — how I use Discord + n8n + LLM approval loop. Main point is the human voice and approval gate are what make it not feel like AI slop.”

2. n8n picks that up within 30 seconds, formats it into a structured prompt combining my rough idea with the system prompt, and calls the LLM.

3. A 1,200-word draft appears in #drafts-for-review about 90 seconds later. It’s usually 70-80% of what I want. The structure is right, the voice is mostly right — but there are places that drift into generic territory, and I can spot them immediately.

4. I read through it on my phone, leave comments in the thread for the parts I want adjusted, and either send it back for a revision pass or drop the ✅.

5. A separate n8n workflow watches for that ✅ reaction, formats the post as Markdown with frontmatter, and opens a pull request in my GitHub repository. I review the PR and merge it. The post deploys automatically via Vercel.

The total time I spend per post: 15-25 minutes across two or three asynchronous check-ins. Instead of needing a two-hour uninterrupted writing block, I need a few minutes of focused attention spread across a workday.

The “Is This Cheating?” Question

I get this every time I talk about this system, and I understand the instinct. But I’d push back on the framing. Spell-check is not cheating. Dictation software is not cheating. Hiring a ghostwriter — which has been standard practice in publishing and business writing forever — is not cheating.

The ideas are mine. The opinions are mine. Every published post passes through my judgment before it goes anywhere near a reader. If I approve something that’s wrong, or generic, or not representative of how I actually think — that’s on me. The system doesn’t reduce my accountability. It reduces the friction between having something to say and actually saying it.

The alternative to this isn’t “authentic manual writing every week.” The alternative is another post that never gets written because I ran out of runway. I’d rather use a tool that lets me publish consistently and genuinely than not publish at all because the bar for “authentic” is too high to clear on a regular basis.

Lloyd Dobler held up the boombox. You just have to show up. The system is how you show up every week instead of once a quarter when the mood strikes.

The ROI

Since putting this in place, my publishing cadence went from sporadic to consistent. I’ve published something every week. My newsletter list is growing. I’ve had two inbound inquiries that cited a specific blog post as the reason someone reached out — not a landing page, not an ad, a post they found through search six weeks after I published it.

That’s the compounding nature of content: it keeps working after you’ve moved on to other things. But compounding only starts when you’re actually publishing regularly. This system is what made regular publishing something I can actually sustain.

How to Build Your Own

The minimum viable version requires:

• A private Discord server (free)
• n8n — self-hosted on a $6/month VPS, or n8n.cloud for a managed option
• An LLM API key (Claude Sonnet or GPT-4o both work well)
• 45 minutes to set up the workflow and write your system prompt

The most important investment is the system prompt — not the technical setup. Spend more time there than on the plumbing. Document your voice, your audience, your opinions. Drop in an example piece you’re genuinely proud of. The plumbing is commodity. The system prompt is the actual product.

If you want to see how an AI content engine fits into a broader automation strategy for your business — or if you’re not sure what your highest-ROI AI use case actually is — let’s talk.