I built an AI landing page roaster in a weekend using Claude API + Netlify Functions

Source: Dev.to

Overview

Every time I shared a landing page for feedback, I got the same vague responses:

• “Your headline could be clearer”
• “Add more social proof”
• “The CTA isn’t strong enough”

Great, but which headline? What kind of social proof? How do I fix the CTA?
Vague feedback doesn’t convert visitors into customers. Specific feedback does.

PageRoast is an AI tool that scores your landing page across six conversion dimensions and gives you the exact fix for each one. You paste your URL and page copy, and you receive:

• An overall conversion score (0–100)
• A breakdown across six dimensions: Clarity, ICP Fit, Pain/Outcome Framing, CTA Strength, Social Proof, Objection Handling
• The specific problem on your page for each dimension (quoting your actual copy)
• A concrete rewrite or fix for each issue
• Your single biggest conversion killer

The Stack

• Frontend: Vanilla HTML/CSS/JS (keeping it simple)
• AI: Claude API (claude-haiku for speed)
• Backend: Netlify Functions (serverless, zero‑config)
• Email capture: Airtable via a Netlify Function
• Hosting: Netlify

The whole thing took a weekend to build and costs less than $0.01 per analysis to run.

Prompt Engineering

The hardest part wasn’t the code; it was the prompt. Generic AI feedback is as useless as generic human feedback. The key was building a structured rubric that forces the model to:

1. Quote the actual copy when identifying a problem.
2. Give a concrete rewrite, not just “improve this”.
3. Return consistent JSON so the UI can display it cleanly.

Demo

It’s completely free—no credit card, no account needed upfront.

👉

Call for Feedback

I’d love feedback from this community—especially on the prompt or the tech stack. What would you have done differently?