DEV Community

yan yan
yan yan

Posted on

How I Built an AI Content Machine That Publishes 40 Articles a Month

#ai #webdev #programming #productivity

I Built an AI Content Machine That Runs 24/7. Here Is the Blueprint.

No team. No budget. Just one laptop, three tools, and a system that never sleeps.

Three months ago, I was spending 12 hours a week writing content. Blog posts. Newsletter. Social media. The grind was real, and the results were mediocre.

Today? I spend 30 minutes a week on content. The machine does the rest.

Here is exactly how I built it — no fluff, no newsletter pitch, just the tools, the workflow, and the mistakes I made along the way.

The Problem Nobody Talks About

Most people think the hard part of content creation is writing. It is not.

The hard part is everything else:

  • Coming up with ideas worth writing about
  • Researching and organizing information
  • Editing for clarity and tone
  • Formatting for different platforms
  • Publishing consistently even when you do not feel like it

Writing is maybe 20% of the work. The other 80% is the machine around it — the system that turns raw thoughts into published content without burning you out.

Most "AI writing" advice stops at "use ChatGPT to generate outlines." That is like stopping at step one of building a car and saying "I have wheels now."

Here is the full blueprint.

The Architecture

My content machine has four layers:

Ideas --> Research --> Draft --> Polish --> Publish
  ^                                      |
  |                                      v
  +------- Feedback Loop -------+
Enter fullscreen mode Exit fullscreen mode

Each layer is automated differently. The key insight: AI is not one tool. It is a pipeline of specialized tools, each doing one thing well.

Here is the stack:

Layer Tool Why
Ideas Notion + Custom AI trigger Generates 50 ideas from 5 seeds
Research Perplexity API Gathers sources, data, quotes
Draft Claude API (long-form) Writes the first draft
Polish Grammarly + Manual review Fixes tone, flow, facts
Publish Custom script + APIs Formats and posts to all platforms

Layer 1: The Idea Engine

This was my first breakthrough. I stopped trying to "think of ideas" and built a machine to do it.

The system is dead simple:

  1. Every Monday, I write down 5 seed topics — broad areas I want to cover this week. For example: "remote work productivity," "AI ethics," "freelancing tools."

  2. I feed those seeds into an AI prompt that generates 10 article angles per seed. The prompt forces specificity:

"For each topic, generate 10 article ideas. Each must include: (1) a specific reader problem, (2) a counterintuitive claim, (3) a concrete example or data point. No generic listicles. No ultimate guides."

  1. Out of 50 generated ideas, I pick the top 5 based on a scoring system:
    • Is this actually useful? (0-3 pts)
    • Can I write it with authority? (0-3 pts)
    • Would I click this headline? (0-3 pts)
    • Is anyone else writing this exact article? (-5 pts if yes)

I spend exactly 15 minutes on this. The rest of the week is execution.

Layer 2: Research Without the Rabbit Hole

Research used to eat half my writing time. I would start Googling one thing, fall into a Wikipedia spiral, and emerge two hours later with 47 open tabs and zero paragraphs written.

Now I use a structured research prompt fed to Perplexity API:

  • Find 3-5 credible sources (academic papers, official docs, first-party data — never Medium posts)
  • Extract 2-3 key statistics or data points
  • Find one counter-argument or opposing view
  • Include source URLs

This gives me a research brief in under 60 seconds. I review it, check the sources, and move on.

The key rule: research is for facts, not for inspiration. If I do not have enough to start writing after this step, the topic is not ready. Skip it, come back later.

Layer 3: The Drafting Pipeline

This is where most people make mistakes. They either:

  • Use AI to write the whole thing and publish garbage
  • Refuse to use AI at all and stay stuck at 2 articles per month

The right approach is a drafting pipeline: AI writes the skeleton and the muscle. You add the heart.

My drafting prompt mandates:

  • Opening: Start with a personal anecdote or surprising stat. No "In today is digital age" openings. Ever.
  • Tone: Direct, opinionated, slightly informal. Write like you are explaining something to a smart friend at a bar.
  • Structure: Problem --> Why it matters --> How to fix it --> Concrete steps --> Results/Call to action.
  • Length: 1500-2000 words.
  • Rules: No fluff adjectives. Every paragraph must either teach something or advance the argument. No exceptions.

The output is about 70% there. Good structure, decent flow, but it lacks personality. That is where I come in.

Layer 4: The Human Filter

I spend exactly 20 minutes per article on the human edit. Here is my checklist:

  1. Voice injection: Rewrite the first paragraph entirely in your voice. This sets the tone for the whole piece.

  2. Example swap: Replace generic AI examples with specific personal stories. "A study found that..." becomes "Last month, a client asked me..."

  3. Contrarian check: Does this article say anything surprising? If not, add one paragraph that challenges the reader is assumptions.

  4. Fact verification: Spot-check every statistic and link. AI hallucinates. Always verify.

That is it. 20 minutes. Then it goes to Grammarly for the final polish pass, and from there to the publishing pipeline.

The Publishing Engine (The Part Nobody Talks About)

This is where 90% of people quit. Writing is done, but publishing across platforms is a nightmare of formatting, image sizing, and copy-pasting.

I wrote a lightweight Python script that:

  1. Takes the final markdown file
  2. Formats it for each platform (Medium HTML, WordPress Gutenberg, Substack, LinkedIn)
  3. Pushes it to the right API
  4. Logs the post URL

The key part — posting to Medium:

import requests

def publish_to_medium(article, token):
    headers = {"Authorization": "Bearer " + token}
    me = requests.get(
        "https://api.medium.com/v1/me",
        headers=headers
    ).json()
    payload = {
        "title": article["title"],
        "contentFormat": "markdown",
        "content": article["body"],
        "tags": article["tags"],
        "publishStatus": "draft"
    }
    resp = requests.post(
        "https://api.medium.com/v1/users/"
        + me["data"]["id"] + "/posts",
        headers=headers,
        json=payload
    )
    return resp.json()
Enter fullscreen mode Exit fullscreen mode

The whole pipeline — from idea to published draft on 3 platforms — takes me about 30 minutes per article.

The Results

After three months of running this system:

  • Output: 12 articles/month --> 40 articles/month
  • Total read time: 1,200 minutes/month --> 8,700 minutes/month
  • Subscribers: 340 --> 2,100
  • Time spent writing: 48 hours/month --> 8 hours/month

And here is the part that surprised me: quality went up, not down.

When I was doing everything manually, I was rushing. Skipping research. Publishing first drafts. The machine handles the grunt work so I can focus on what actually matters: the thinking, the voice, the human insight.

What I Would Do Differently

1. Start with a smaller scope. My first version tried to automate everything at once. It broke. Start with one layer, get it working, then add the next.

2. Log everything. I did not track which prompts worked best for the first month. That was stupid. Now every article has metadata: prompt version, research sources, edit time, performance. A/B test your prompts like you would ads.

3. The voice is the moat. Anyone can generate AI content. Nobody can replicate your specific perspective and experiences. The human edit is not where you save time — it is where you create value.

The Blueprint (Steal This) 

STEP 1: Idea generation (Monday, 15 min)
  - Feed 5 seed topics into AI prompt
  - Score and select top 5 article ideas

STEP 2: Research sprint (Monday, 30 min)
  - Run all 5 ideas through research prompt
  - Review and save research briefs

STEP 3: Draft (Tue-Wed, 10 min/article)
  - Feed research brief + style guide to AI
  - Generate first drafts for all 5 articles

STEP 4: Human edit (Wed-Thu, 20 min/article)
  - Voice injection, example swap, fact verify

STEP 5: Polish and publish (Friday, 30 min)
  - Grammarly pass
  - Run publishing script
  - Schedule social posts for the week

TOTAL: ~4 hours/week for 5-7 polished articles
Enter fullscreen mode Exit fullscreen mode

The One Rule

If you take nothing else from this article, take this:

The goal is not to replace yourself with AI. The goal is to replace the parts of your workflow that do not require you.

Your taste. Your experience. Your weird opinions. Those are irreplaceable — and they are exactly what make your content worth reading. Everything else? Automate it.

If you found this useful, follow for more on building automated systems that let you do more with less. No hustle culture. Just practical engineering applied to creative work.

Top comments (0)