How I Generated 10 Million Yen in Monthly Sales by Building a Landing Page Using Only Claude Code

I built a landing page (LP) for a new product I'm supporting using "only Claude Code," and in the third month after launch, monthly sales exceeded 10 million yen. No outsourcing to designers, no development costs, no requests to LP production companies. All I used was Claude Code and my "time spent writing prompts."

Here is the breakdown of the numbers:

• Product Price: 98,000 JPY (Online course, one-time purchase)
• Monthly Conversions: Approx. 105
• Monthly Sales: Approx. 10.3 million JPY
• CVR (Conversion Rate) via LP: Approx. 1.8% against ad traffic
• Actual LP Production Cost: A few thousand yen per month for Claude Code fees + domain/server

In a world where people say "it costs 300,000 to 1,000,000 JPY to hire an LP production company," the production cost was almost zero. Moreover, since I have Claude Code handle all post-launch improvements (AB testing, copy edits, section replacements), the speed of improvement is incomparable to the outsourcing era. In this article, I will reveal the entire process along with the actual prompts I used.

What You Will Learn in This Article

1. The entire flow of creating an LP with Claude Code (Requirements → Copy → Implementation → Measurement → Improvement)
2. Full text of the prompts actually used in each step
3. What I kept in mind for the LP structure to achieve a 1.8% CVR
4. How to let Claude Code handle post-launch AB testing and improvements
5. Points where I got stuck and how to avoid them

Why I Decided to Build with "Only Claude Code"

I have outsourced LP production three times in the past, costing between 350,000 and 800,000 JPY per page. While I wasn't dissatisfied with the final product, structural problems became apparent once operation began:

1. Slow and Expensive Revisions: Even small changes like fixing copy or moving a section cost time and money. Speed is impossible when every AB test requires a quote.
2. Fragmented Copy, Design, and Implementation: When the writer, designer, and coder are different people, consistency is lost in a game of telephone.
3. No One to Fix Based on Data: Production companies are pros at "making," not necessarily "selling." They don't always know how to turn GA4 data into hypotheses and fixes.

Claude Code is a terminal-based AI coding agent. It doesn't just write code; it manages file structures, previews HTML/CSS/JS, handles copywriting, and even implements improvements based on GA4 or heatmap data. This integration was the deciding factor.

Product and Prerequisites

• Product: Online course for B2B skills (98,000 JPY)
• Traffic: Meta Ads (Instagram/Facebook) + Search Listing. Monthly ad spend approx. 1.8M JPY
• Sales Flow: Ad → LP → Payment (Stripe). Simple direct sales.
• My Skill: I can read HTML/CSS but can't write it beautifully. I barely use design tools.
• Tools: Claude Code, GitHub, Cloudflare Pages, GA4, Microsoft Clarity.

Crucially, I barely write any code. Claude Code acts as the translator between the "director" and the "maker."

The 5-Step Flow

1. Requirements: Have Claude Code "interview" you about the product and customers.
2. Structure and Copy: Create the section layout and full-text copy.
3. Design and Implementation: Implement with HTML/CSS and refine via preview.
4. Measurement and Deployment: Set up GA4/Clarity and go live.
5. Improvement Loop: Feed data back to the AI for AB testing.

STEP 1: Requirements - The "Interview" Prompt

Don't just say "make an LP." AI needs specific info to avoid generic results. I used a prompt to make Claude Code interview me, asking specific questions about target audience, anxieties, and price justification. I provided raw customer survey data and Slack logs. This resulted in a structured requirements.md file.

STEP 2: Structure and Copy

I had Claude Code create a section structure based on the requirements, focusing on neutralizing anxieties rather than just following a template like PASONA. For the copy, I didn't write it all at once. I asked for 10 options per section to choose the best direction. I also strictly prohibited the AI from inventing fake testimonials or using illegal exaggerated claims.

STEP 3: Design and Implementation

I used Claude Code to implement a mobile-first, high-performance site using static HTML/CSS/JS. By having it build only the "First View" first, I could align on the design tone before expanding to the whole page. I used "verbal direction" to adjust margins, button sizes, and layouts in real-time. For design inspiration, I fed it screenshots of LPs I liked and asked it to translate their "good points" into my project.

STEP 4: Measurement and Deployment

I had Claude Code implement GA4 and Microsoft Clarity, specifically tracking "section read-through" events to see exactly where readers dropped off. Deployment was handled via Cloudflare Pages and GitHub, all guided by the AI.

STEP 5: Improvement Loop

Starting at a 0.9% CVR, I fed GA4 CSV data and Clarity screenshots back to Claude Code. It proposed three major fixes: reordering the price section, turning the FAQ into an "objection handling" section, and AB testing the hero copy. These changes pushed the CVR to 1.8%, hitting the 10M JPY sales mark. These iterations happened within hours, not weeks.

Summary: 3 Principles for Prompt Design

1. Role + Context File + Output Format: Give it a specific role and refer to files like requirements.md.
2. Don't Build All at Once: Set milestones for agreement (Interview → Structure → Copy → Implementation).
3. Prohibitions and the Right to Disagree: Tell the AI to forbid fake data and to argue if a change might lower the CVR.

Claude Code has turned LP production from "buying time with money" to "buying time with prompts."

https://動かせるサービスリンク