A Stay-at-Home Dad With Zero Coding Skills Built Websites With AI — After 300 Attempts, He's Now Making Over $220K/Year

A stay-at-home dad with absolutely zero coding knowledge spends his days caring for his kids and his nights building websites with AI — and he's now earning over $220,000 a year (roughly 1 million+ RMB). This isn't a marketing gimmick. It's the real story of an ordinary guy named Liang Chengwei, earned through trial and error across 300+ websites.

Blogger Guangyu visited his "studio" in person — a study room cluttered with kids' stuff, one computer, one whiteboard. That's the entire setup.

It Took 40 Websites Before He Made Any Money

Liang Chengwei's background is remarkably "ordinary": he tried learning Python in college and failed, couldn't follow his C++ classes at all, and knew nothing about code syntax. In his own words: "I don't understand code syntax, but I can recognize the letters A-B-C-D."

But he had one core conviction: From 2025 to 2035, there's likely a massive window of opportunity for AI-built websites. The code revolution brought by AI has reduced the barrier to building websites to zero, while users' willingness to pay remains strong.

What's truly impressive isn't his income figure — it's how many times he failed. His first website took over a month. Things didn't start clicking until website #40. The first 39 were all built in a state of "knowing absolutely nothing" — domains, subdomains, deployment, hosting — every single concept was learned from scratch.

His learning method is also unique: no tutorials, no courses — just chatting with AI every day. Token costs are his biggest "tuition" expense. "If you don't understand something, ask AI" — he repeated this countless times.

Background: AI Code Generation Tools Liang Chengwei primarily uses Claude, a large language model developed by Anthropic, which stands alongside OpenAI's GPT series as one of the most mainstream AI coding assistants today. These models are trained on massive code repositories and technical documentation, enabling them to understand natural language descriptions and generate corresponding HTML, CSS, JavaScript, and other frontend code. For non-technical users, the core value lies in converting "intent" directly into "executable code," bypassing the high barriers of traditional learning paths like memorizing syntax and configuring debugging environments. Since 2023, with iterations of GPT-4, Claude 2/3, and other models, code generation accuracy and contextual understanding have improved dramatically, turning "building websites with natural language" from concept into reality. "Token costs" refer to the per-text-unit charges when calling AI APIs — the more conversation and the longer the code, the more tokens consumed, which is why Liang considers this his primary learning cost.

Two Main Revenue Streams: AI Tool Sites and Game Guide Sites

Liang Chengwei's websites fall into two main categories:

The first is AI tool sites — things like AI image generation, real-person-to-animation conversion, etc. Demand for these exploded in 2024, and he precisely caught the wave.

The second is content/guide sites, especially game guides. For example, in January 2025, a game called "Xindong Town" launched, and he quickly built a guide site collecting map information, gameplay tips, bug reports, and more.

His advice for beginners: Start with guide sites. The reason is simple — guides are primarily text-based, the technical barrier is lower, and content production is faster.

Background: The SEO Window for Game Guide Sites The core competitive advantage of game guide sites lies in the "time gap." When a new game first launches, search engine competition for related keywords is extremely low. Publishing high-quality content early allows you to claim top search rankings at minimal cost. As the game gains popularity, organic traffic continues flowing in, creating a "first-mover effect." This differs from AI tool sites — AI tool sites require continuous technical iteration to stay competitive, while guide sites, once a content moat is established, have relatively low maintenance costs. For zero-experience beginners, guide sites also have a hidden advantage: content quality standards are more intuitive, user feedback is more immediate, and this helps quickly develop product intuition.

Live Demo: Building a Website From Scratch With AI in Just a Few Steps

In the video, Liang Chengwei demonstrated the entire process of building a game guide website using Claude:

1. Requirements analysis: Simply tell Claude "I want to build a guide site for this game — first help me analyze user needs," and the AI outputs complete MVP page requirements, core user profiles, and more
2. Generate code: Have AI write the frontend code directly, then download it locally
3. Upload to GitHub: Push the code to a GitHub repository
4. Deploy and go live: Deploy through a hosting platform and bind a domain name

Background: MVP Methodology and the GitHub Hosting Ecosystem When Liang has AI output "MVP page requirements," MVP (Minimum Viable Product) is a core concept in Silicon Valley startup methodology, systematically articulated by Eric Ries in The Lean Startup. The core logic: use minimal resources to build a product version that can validate core assumptions, iterate quickly through real user feedback, rather than pursuing one-shot perfection. Liang's trial-and-error journey across 300 websites is essentially executing the "Build-Measure-Learn" lean loop at high speed, with AI tools compressing the "Build" phase from days to hours.

GitHub is the world's largest code hosting platform, operated by Microsoft. For non-technical entrepreneurs, its core value isn't just storing code — it's the deep integration with modern hosting platforms like Vercel, Netlify, and Cloudflare Pages. Once code is pushed, deployment triggers automatically, making the entire workflow highly automated. This combination essentially abstracts the server infrastructure traditionally managed by DevOps engineers into a few clicks, with generous free tiers that dramatically reduce cold-start costs for indie developers.

Throughout the entire process, his core operations are just copy-paste and chatting with AI. Encounter a bug? Throw the error message to AI and let it troubleshoot. Wrong filename? AI tells you what to fix.

He used a vivid metaphor: if building a website is like constructing a house, AI is the master craftsman — you just tell it what style and features you want; GitHub is the building materials warehouse; the hosting platform is the plot of land; and the domain name is the street address.

That said, Guangyu admitted that despite how simple it looks, when he actually faced all those technical terms and English interfaces, he was "completely lost." Behind Liang's practiced fluency are 300+ websites built through countless days and nights.

Monetization: User Payments + Ad Revenue

Revenue comes from two channels:

Direct monetization — charging users. The mainstream overseas model is selling memberships or credit packages — for example, $10 for 100 credits, with different AI models consuming different amounts of credits.

Indirect monetization — ad revenue. Integrating ad platforms like Google AdSense and embedding ad code into the website. He cautions about finding the right balance: third-party ad platform pop-ups can hurt user experience and bounce rates — "don't overdo it."

Background: Google AdSense Monetization Logic Google AdSense is the world's most mainstream website ad monetization platform, using two billing models: CPC (cost per click) and CPM (cost per thousand impressions). Revenue depends on several key factors: traffic geography (a single click from a European or American user is typically worth 5-10x that of an Asian user — this is the core reason Liang targets overseas markets), content verticality (finance, legal, and software ads command the highest rates), and page quality metrics (time on page, bounce rate, etc.). Liang emphasizes "moderation" when it comes to ad placement.