Planning and Structuring a Website with AI [Series Part 2]

Hi there. Today I want to share how I used AI to plan and structure a website from scratch.

Getting Started with AI-Powered Planning

The moment I decided to use AI to build my own site or blog, thousands of ideas flooded my head. How do I create a site that attracts lots of visitors? What kind of articles would keep people interested and make them stay longer? I agonized over these questions endlessly.

If I'd been wrestling with the same questions two or three years ago, just the planning phase would have taken two to three weeks. Brainstorming, building mind maps, filtering which ideas fit the site's overall structure, then manually typing up quality articles one by one to meet AdSense's minimum post requirements — it would have been a brutal process.

But now we have AI. Used well, it makes planning and structuring a site remarkably easy. You just talk through your thoughts — almost like venting — and AI grasps the big picture and drafts an outline for you. What I especially appreciated was that AI also flagged variables and considerations I hadn't even thought of, which made the process feel more reliable.

Let me walk you through my actual process in more detail.

1. The AI Tools I Used: Gemini + Claude Code

I used two AI tools: Gemini and Claude Code. Both on Pro plans, and both have become invaluable partners. I divided their roles based on my preferences.

Gemini handled site structure, planning, legal considerations, and other variables — essentially everything related to strategic planning and operational concerns. I think of Google Gemini as a brilliant legal advisor. I use it the way you'd consult a lawyer to get clear, precise answers.

Claude (Code) fills a different role — more like a genius friend. For writing, it produces text in the style I like best. Even though they're both AI, their output feels distinctly different in character, which is fascinating. Claude Code specifically acts like a brilliant developer friend who understands exactly what I want and builds it for me.

2. Creating the PRD (Product Requirements Document)

I didn't just chat back and forth with AI from start to finish. First, I worked through conversations to create a PRD that I could hand to Claude Code with precise requirements.

For developers, a PRD is a document detailing a product's specifications — it includes the site overview, purpose, structural components, and enough information that anyone reading it can understand what the site is, what content it'll have, and what it'll look like.

Since requesting everything at once seemed unrealistic, I split the PRD into phases — two phases for a small site, up to five for a larger project — so I could hand them to Claude in manageable chunks.

For the PRD creation itself, I asked AI to "format everything we've discussed into a PRD that I can hand to Claude Code." Ideas that were vague and floating around in my head — not fully formed, more like impressions and gut feelings — I shared honestly, and we gradually shaped them into something concrete and visual.

3. Where AI Really Shines

This is where AI's strengths become most apparent.

If I had taken the vague, unstructured thoughts in my head and dumped them on colleagues in a workplace setting — in disorganized words and messy documents — I'd probably be labeled as someone who's hard to work with, someone who doesn't even understand their own job.

In a real work environment, just preparing materials to show collaborators takes days at minimum. Client-facing materials? You might spend an entire month just on the draft.

But now, I can pour out everything in my head to AI, and it organizes, supplements, and produces a nearly perfect guide. The speed at which site development progresses is honestly astounding.

4. Curious About Something? Just Ask

One of the hardest things about building a site is taking that first step — because you don't fully know what you want, and you don't know what site structures the market responds to.

You might have ideas, but how do you implement them in a way that actually attracts visitors? Personal satisfaction isn't enough — a site that's only pleasing to you won't succeed. In the past, this meant running surveys or consulting people you know.

Now you can ask AI, and it draws on vast datasets to provide clear criteria — often based on far more data points than any survey you could run yourself.

Beyond that, any question that came up during the process — things I was curious about, concepts I'd heard of but never fully understood — I just asked as I worked. It kept the momentum going.

Summary

By planning and structuring each site this way, the process was far faster and more accurate than trying to figure it all out alone. Once the plan was ready, I just typed "Claude Code, please proceed with the build" — and AI created my own polished site.

Of course, this doesn't produce a 100% perfect site on the first pass. But the draft comes together so quickly that you can then iterate — having follow-up conversations and refining the site as you go. That's another major advantage.

Next Up

In the next post, I'll cover the process of deploying the site after building it.