72 GhostScore
Product Design + Development

GhostCheck — The CRO Audit That Writes the Fix

Project ghostcheck.eu
Role Solo Founder — Design, Dev, Product
Stack React · Supabase · Claude API · Firecrawl
Year 2026
01 — Overview

What if your website audit came with the actual fixes?

GhostCheck is an AI-powered conversion audit tool that analyses any website and delivers a Premium Fix Guide — complete with rewritten copy, working code snippets, SEO schema markup, mobile UX fixes, and a prioritised implementation roadmap. Not just a score. Not a generic checklist. A consultant-grade deliverable, generated in minutes.

I designed, built, and shipped this end-to-end as a solo product — from the AI pipeline architecture to the PDF generation system, scoring methodology, and monetisation flow.

02 — Problem

Existing audit tools tell you what's wrong. None tell you how to fix it.

Website owners get a score, a list of technical issues, and a wall of jargon. Then they're left to figure out the rest. The gap between diagnosis and action is where conversion improvements die.

Generic recommendations

Tools like Lighthouse give technical scores with no context on business impact. "Eliminate render-blocking resources" means nothing to a founder.

01

No copy intelligence

Zero tools rewrite your actual hero copy. Conversion lives and dies in the first 7 seconds — yet every audit ignores messaging entirely.

02

No implementation path

Even premium audits deliver PDFs with no code, no before/after examples, no time estimates. The report sits in a folder forever.

03

Not built for DACH markets

Every CRO tool is US-centric. Swiss and DACH sites need hreflang de-CH awareness, CHF pricing context, and German keyword strategy. No competitor touches this.

04
03 — Solution

A Premium Fix Guide that does the work for you

GhostCheck crawls the site with Firecrawl, passes structured content to Claude with a carefully engineered CRO consultant system prompt, and generates a validated, structured JSON report — then renders it as a pixel-perfect PDF.

01

GhostScore™ Methodology

Scoring System
5-dimension scoring across First Impression (25pts), Copy Clarity (20pts), Trust Signals (20pts), CTA Effectiveness (20pts), and Navigation (15pts). Includes score disclaimer to contextualise results vs. brand authority.
02

Before/After Copy Rewrites

AI Copywriting
Full hero headline rewrites + Extended Copy Rewrites for Services, How It Works, and Results sections — each with psychological principle explanations. The reasoning is consultant-quality, not generic.
03

Copy-Paste Code Snippets

Developer Ready
Every fix includes working HTML/CSS/JS or JSON-LD schema markup. No placeholders, no generic examples — code is specific to the fix being implemented.
04

Annotated Hero Screenshot

Visual Evidence
Puppeteer captures the above-the-fold viewport, overlays numbered annotation circles on the hero headline, CTA, and trust bar zones, and embeds the annotated image as page 2 of the PDF.
05

Validation + Auto-Retry Pipeline

Reliability
Every generated report runs through validateFixGuideData() before PDF rendering. Empty sections, missing copy rewrites, or score math errors trigger automatic AI retry with targeted error context.
06

DACH Market Intelligence

Regional Moat
System prompt enforces hreflang de-CH/fr-CH/it-CH recommendations, CHF pricing awareness, and German keyword variants in SEO sections. No other CRO tool does this.
04 — Build Process

Shipped end-to-end in weeks, not months

1

Crawl & Extract

Firecrawl scrapes the target URL, extracts structured content including hero copy, CTAs, navigation, and meta tags for analysis.

2

AI Analysis

Claude processes the content against a 900-word CRO consultant system prompt with scoring rubrics, voice guidelines, and DACH-specific rules.

3

Validate & Retry

Output JSON is validated against the full FixGuideData schema. Missing sections or score math errors auto-retry with targeted prompts.

4

PDF Render

jsPDF renders the structured data into a branded PDF with annotated screenshot, proper page breaks, and clean typography — no truncation.

"The hardest part wasn't the AI — it was making the PDF output look like something a senior consultant would be proud to send."

— Karolina Bittmanova, Founder
05 — Results

Real reports. Real fixes.

Before — March 18
52
taoinfinity.com
Generic hero headline
Zero trust signals on page
Broken NaN carousel bug
→
After — Implementing fixes
72
potential score
Sacred geometry positioning
Trust bar + review signals
Priority nav + email capture
8fixes
Shipped in one sprint
5sites
Reports validated in testing
0blank
Sections in final reports
3tiers
Free · Premium · Concierge
06 — Tech Stack

Built lean, ships fast

Every technology choice was made to maximise output quality while minimising infrastructure cost — aligned with the solo founder, low-SaaS-cost philosophy.

React + Vite TypeScript Tailwind CSS shadcn/ui Supabase (Auth + DB + Edge Functions) Claude API (Sonnet) Firecrawl API jsPDF Puppeteer (Deno) Browserless.io Vercel Stripe
07 — Learnings

What I'd do differently

🎯

System prompt is the product

The 900-word CRO consultant system prompt — with scoring rubrics, voice rules, and DACH specifics — is what makes the output valuable. Engineering the prompt took longer than engineering the pipeline.

🔁

Validate before render, always

Blank sections in a paid report destroy trust instantly. Building the validation + auto-retry layer before launch would have saved multiple embarrassing report generations during testing.

🏢

Agencies are the real ICP

The white-label agency plan — unlimited reports for prospect pitches — is a stronger business model than selling reports to end clients. Should have designed for this from day one.