I Built an AI Website with an AI Assistant: 30-Day Results
A complete behind-the-scenes look at building AIGoMoon entirely with an AI assistant. 80 articles, 13 AI songs, 3 functional tools, and 468 real user interactions in 30 days. Here's everything that worked and everything that failed.
TL;DR: The Key Numbers
In 30 days, I built a complete AI website using only an AI assistant. Here are the real numbers:
- 80 articles published across 8 categories
- 13 AI-generated songs with synced LRC lyrics
- 3 functional tools: 2048 game, color blindness test, QR code generator/scanner
- 468 total user interactions (109 article views, 342 song plays, 17 favorites)
- 0 lines of code written manually — everything generated by AI
- $0 spent on development (only Cloudflare free tier)
How It Started
I had an idea for an AI content platform but zero coding experience. Instead of learning to code from scratch, I decided to try something different: describe what I wanted to an AI assistant and let it build everything.
The first conversation was simple: "I want to make an AI website called AIGoMoon with 8 sections, bilingual support, and a music player." What followed was 30 days of iterative development, debugging, and feature additions — all through natural language conversations.
The Tech Stack (That I Never Had to Learn)
The AI chose a simple but powerful stack:
- Frontend: Pure HTML, CSS, and JavaScript (no frameworks)
- Backend: Cloudflare Pages Functions (serverless)
- Database: Cloudflare KV (key-value storage for analytics)
- Hosting: Cloudflare Pages (free tier)
- Content: JSON files (data-driven, no CMS needed)
The best part? I didn't need to understand any of this. The AI explained what it was doing and handled all the implementation.
What Worked Incredibly Well
1. Iterative Development
The music player went through 15+ revisions. Each time I said "the lyrics are out of sync" or "add a floating lyrics window," the AI modified the code and redeployed. This iterative approach would have taken me months to learn and implement manually.
2. Data-Driven Architecture
All content lives in JSON files. Adding a new article means editing one JSON file — no code changes needed. This made daily automated updates possible.
3. Automated Daily Updates
I set up a scheduled task that runs every morning at 6 AM. It automatically: searches for trending AI topics, writes 2 bilingual articles, generates cover images, updates SEO metadata, and deploys to production. Zero manual intervention.
What Didn't Work (And How We Fixed It)
1. Image Path Issues
The most common bug was image paths. The AI sometimes used absolute paths when relative paths were needed, causing broken images on the homepage. Fix: standardized all image paths to relative format and added a prefix in the card rendering function.
2. Duplicate IDs
When adding new articles, the AI occasionally reused existing IDs, causing cards to link to the wrong articles. Fix: implemented an ID validation check before deployment.
3. Browser Caching
After deployments, users saw old versions because of browser caching. Fix: added version numbers to all CSS/JS file references (style.css?v=200, main.js?v=1003).
4. Lyrics Sync
The AI-generated LRC files were often out of sync with the audio. Fix: manually adjusted timestamps by listening to each song and fine-tuning the timing.
The Real Data: 30 Days of Analytics
Here's the honest data from Cloudflare KV storage:
| Metric | Number |
|---|---|
| Total article views | 109 |
| Total song plays | 342 |
| Total favorites | 17 |
| Most played song | music-02 (61 plays) |
| Most viewed article | frontier-02 (11 views) |
| Total interactions | 468 |
The music player was by far the most popular feature, accounting for 73% of all interactions. This told me that AI music is what users come for, so I doubled down on that section.
Lessons Learned
1. Be Specific with Requests
"Make it look better" gets mediocre results. "Change the card border radius to 16px, use a light gray background #f5f7fa, and make the title blue #2563eb" gets exactly what you want.
2. Test on Real Devices
The AI tested in a desktop browser, but mobile had different issues. Always test on actual phones — I found at least 5 mobile-only bugs that the AI couldn't reproduce in desktop mode.
3. Keep a Change Log
When something broke, I needed to know what changed. The AI doesn't remember previous sessions perfectly, so maintaining a simple change log saved hours of debugging.
4. Don't Trust AI-Generated Data Blindly
The AI initially put fake click numbers (hundreds) in the content JSON. I had to replace them with real KV data. Always verify analytics against the actual database.
Would I Do It Again?
Absolutely. Building this website with an AI assistant taught me more about web development than any tutorial could. The key insight: you don't need to know how to code — you need to know how to communicate what you want clearly.
The website is live at aigomoon.com and continues to update automatically every day. If you're thinking about building something but feel intimidated by coding, I highly recommend trying the AI-assisted approach.
What's Next
I'm planning to add more AI tools, expand the music library, and implement user accounts. The AI assistant handles all the technical work, so I can focus on content strategy and user experience.
Have questions about the process? Feel free to reach out — I'm happy to share more details about what worked and what didn't.