I Built a Fully Automated AI Blog in 3 Days With Zero Coding Skills (Here’s Exactly How)

Quick Summary
Time to build	~3 days
Monthly cost	$50–60 (VPS + APIs + domain)
Articles per day	1 (fully automated)
Total articles published	21
Revenue so far	$0
Coding skills required	Zero (I had none when I started)

The Part Where I Tell You Something That Sounds Like Clickbait But Isn’t

I’m 43 years old. I live in Japan. Three weeks ago, I couldn’t write a single line of Python. I didn’t know what an API endpoint was. I thought “VPS” was some kind of shipping service. Today, I have a blog at ai-tool-hunter.com that automatically finds AI tools, writes full review articles about them, sends me a notification on Discord, waits for me to tap a button, and then publishes to WordPress, dev.to, and X — all without me writing a single word of the actual content.

Let me be clear about what this is not: this is not a story about making $10,000 in my first month. I’ve made exactly $0. This is not a story about building something perfect. Something breaks almost every day. This is the story of a non-technical person in his forties using AI to build a thing that technically shouldn’t be possible for someone like me to build. And I’m going to show you exactly how, with every cost, every error, and every ugly truth included.

Twenty-one articles are live right now. I didn’t write any of them by hand. The system produced every single one. Whether this becomes a real income source or an expensive lesson, I figured it was worth documenting either way.

Why I Did This

The honest answer is boring: I want passive income. I’ve been working in Japan for years, and I’ve watched the AI wave from the sidelines long enough. I had two major problems standing between me and a content-based business:

• I can’t write English well enough. My English is functional, but producing polished, native-level blog posts every day? Not happening. Not at the speed or quality needed to compete.
• I can’t code. Every automation tutorial I found assumed you already knew Python or JavaScript or at least understood how to set up a development environment. I didn’t know any of that.

The realization that changed everything was simple: AI can bridge both gaps simultaneously. Claude can write better English than I can, and Claude can also teach me how to write the Python scripts that make the whole system run. I’m essentially using AI to build an AI-powered system. It’s recursive in a way that still feels slightly absurd to me.

I didn’t have a grand vision. I just thought: what if I could publish one quality article per day about AI tools, and what if the whole thing ran on autopilot except for a 30-second approval step? That was the entire plan. So I started building.

The Exact Stack I Used

Here’s everything, with real costs. No affiliate links. No sponsor deals. Just what I actually pay.

• Contabo Windows VPS — ~$15/month. I chose Windows because it’s what I know. Yes, Linux would be cheaper. No, I don’t care. I needed to reduce the number of things I was learning simultaneously, and a Windows desktop environment I could remote into kept me sane.
• Python — Free. I learned the basics by asking Claude to explain things to me like I was a beginner, because I was. I still Google syntax errors constantly.
• Claude API (Anthropic) — ~$20–30/month at my current usage. This is the brain of the operation. It writes every article. The cost scales with how many articles I generate and how long the prompts are.
• WordPress — Free (self-hosted on the VPS). I’m using a standard theme. Nothing fancy.
• Domain — ~$10/year for ai-tool-hunter.com. Negligible.
• Discord — Free. This is my approval dashboard. I built a bot that sends me article previews and I tap Publish or Skip. It’s the “human-in-the-loop” part that keeps me from publishing garbage.
• Product Hunt — Free. This is where the system currently finds new AI tools to review. I’m in the process of pivoting to keyword-based sourcing, but more on that later.

Total monthly cost: roughly $50–60. That’s it. Less than most people spend on streaming subscriptions.

How It Actually Works (Step by Step)

This is the part people actually want to read, so I’ll be specific.

Step 1: Scraping AI Tools from Product Hunt

A Python script runs daily on my VPS. It checks Product Hunt for newly launched AI tools. It pulls the tool name, description, URL, and any available metadata. This gives me a list of candidates for today’s article. The script isn’t elegant — it’s functional. It breaks when Product Hunt changes something on their end, which has happened twice in three weeks.

Step 2: Claude API Generates the Full Article

The tool data gets fed into a carefully structured prompt that I send to the Claude API. The prompt instructs Claude to write a full review article, including a proprietary 4-axis scorecard I designed (covering categories like ease of use, pricing value, feature depth, and market readiness). The output is a complete blog post — title, introduction, analysis, scorecard table, verdict — usually around 1,200 to 1,500 words. This step takes about 30 to 60 seconds.

Step 3: Discord Notification

The finished article gets sent to a private Discord channel as a preview. The Discord bot formats it nicely and attaches two buttons: Publish and Skip. I get a ping on my phone.

Step 4: I Approve (30 Seconds)

I read the preview on my phone. If it looks good — and it usually does, because I spent a lot of time refining the prompt — I tap Publish. If something is off, I tap Skip. The entire decision takes about 30 seconds. This is the only manual step in the entire pipeline, and I refuse to remove it. Fully autonomous publishing with zero human oversight is how you end up with a site full of hallucinated nonsense.

Step 5: Auto-Publish Everywhere

Once I tap Publish, the system pushes the article to three places simultaneously: WordPress (the main site), dev.to (for backlinks and developer audience exposure), and X (as a tweet with the article link and a short summary). The WordPress posting uses the WordPress REST API. The dev.to posting uses their API. The X posting uses their API. All of this was built by asking Claude to write the integration code for me, then debugging it when it inevitably didn’t work on the first try.

The Honest Results

Here’s where I refuse to sugarcoat anything.

• 21 articles published as of writing this.
• $0 in revenue. Zero. Not a penny.
• Organic traffic is near zero. Google is barely aware the site exists. This is completely expected for a new domain — the “Google sandbox” is real, and most SEO experts will tell you it takes 3 to 6 months before a new site starts getting meaningful organic traffic.
• Some dev.to articles have gotten modest views, which is encouraging but not income.

I went into this knowing the first few months would produce nothing. If you’re looking for a “I made $5,000 in my first week” story, close this tab. This is a long game. The question I’m betting on is whether a site that consistently publishes quality, structured AI tool reviews every single day for 6 to 12 months will eventually rank and earn. I believe it will. But I don’t know yet. That’s the truth.

What Went Wrong (Don’t Skip This Section)

If I only told you about the pipeline working smoothly, I’d be lying to you. Here’s a partial list of things that have broken, crashed, or made me want to throw my laptop out the window:

• Discord bot crashes randomly. The bot process dies on the VPS for no clear reason. I’ve added auto-restart logic, but it still happens. I check it every morning like a farmer checking on livestock.
• API timeouts. The Claude API occasionally times out on longer prompts. I had to build retry logic, which — for someone who learned Python three weeks ago — took an embarrassing amount of time.
• Windows encoding hell. Running Python on Windows introduces character encoding issues that Linux users never deal with. Japanese system locale + English content + special characters in tool names = hours of debugging UTF-8 errors. If I could go back, I might choose Linux and just suffer through the learning curve upfront.
• Product Hunt scraping inconsistencies. Some days the scraper returns garbage. Some tools have almost no description, which means the Claude-generated article is thin. I’ve had to add validation checks, minimum content thresholds, and fallback logic.
• WordPress REST API authentication headaches. Getting the API credentials working correctly took almost an entire day of my three-day build. The documentation is fine if you already understand authentication tokens. I did not.
• Scheduling conflicts. Running multiple scripts on Windows Task Scheduler that depend on each other in sequence is more fragile than it sounds. One script takes longer than expected, the next one fires before the first is done, and everything collapses.

Something breaks almost every day. I fix it. The next day, something else breaks. This is the actual reality of running an automated system as a non-developer. If that sounds exhausting, it is. But each fix makes the system slightly more robust, and the frequency of problems is slowly decreasing.

What I’d Do Differently

If I were starting over tomorrow with what I know now, two things would change immediately:

1. Start With Keyword Research, Not Product Hunt

My current approach — scraping new tools from Product Hunt and reviewing them — produces articles about tools that often have zero search volume. Nobody is Googling a tool that launched yesterday. I’m now pivoting to a keyword-first approach: find what people are actually searching for (“best AI writing tools 2025,” “Jasper vs Copy.ai,” etc.) and then build content around that demand. The Product Hunt pipeline was a good way to build the system and get articles flowing, but it’s not an SEO strategy. I should have known this from day one.

2. Build Backlinks From Week One

I spent zero time on link building during the first three weeks. That was a mistake. Domain authority matters, and you can’t just wait for it to appear. Guest posts, HARO responses, community participation — I should have been doing all of that from the start, even while the automated system was being built.

Who Should (and Shouldn’t) Try This

This is a good project for you if:

• You have patience measured in months, not days.
• You’re comfortable with $50–60/month going to something that might not pay off for half a year.
• You enjoy problem-solving and don’t rage-quit when things break.
• You want to learn how APIs, automation, and AI content generation actually work — by building something real.
• You have a niche in mind that you genuinely find interesting. Mine is AI tools. If you pick something you don’t care about, you’ll quit in week two when the fifth error crashes your pipeline at 2 AM.

This is NOT for you if:

• You want revenue in the first month. It’s not happening.
• You think “automated” means “zero effort.” It means less effort per article, but the system needs constant babysitting — at least in the early months.
• You’re not willing to talk to an AI for hours to debug code you don’t understand. My Claude conversation history for this project is enormous. I asked dumb questions constantly. If that embarrasses you, this isn’t your project.
• You want a “set it and forget it” passive income machine. This is closer to “set it, fix it, fix it again, fix the fix, then maybe semi-forget it in six months.”

The Code Is on GitHub (Coming Soon)

I’m cleaning up the codebase right now. “Cleaning up” is generous — I’m trying to make it comprehensible to someone who isn’t me. The full pipeline — Product Hunt scraper, Claude API integration, Discord approval bot, WordPress auto-publisher, dev.to and X cross-posting — will be open-sourced on GitHub within the next few weeks. I’ll update this post with the link when it’s live.

Fair warning: the code is not pretty. It was written by a 43-year-old who learned Python from an AI assistant three weeks ago. It works. That’s the only claim I’m making about it.

The Verdict

Here’s my honest, one-sentence assessment after 21 articles and $0 earned:

Building this system taught me more about AI, automation, APIs, and content strategy in three weeks than I learned in two years of reading about them — and even if the blog never earns a dollar, that education alone was worth the $50/month.

But I think it will earn. Ask me again in six months.