Building ReviewAI: How We're Solving Amazon's Fake Review Problem with AI

The Problem: Amazon Shopping is Broken

Ever spent 30+ minutes reading Amazon reviews, only to end up more confused than when you started?

You're not alone. Here's what we discovered:

- 78% of shoppers don't trust Amazon reviews anymore
- Average research time: 30-60 minutes per product
- 25% of purchases get returned (often due to misleading reviews)
- Fakespot shut down in July 2025, leaving millions without a solution

The problem isn't just fake reviews. It's:
- Information overload (10,000+ reviews per product)
- Conflicting opinions (4.8★ but "broke in 2 weeks")
- Hidden deal-breakers buried in 3-star reviews
- No clear answer: "Should I buy this or not?"


The Solution: AI-Powered Purchase Intelligence

We built ReviewAI - an AI shopping assistant that analyzes Amazon products and gives you a decisive verdict with evidence.

### What It Does

Input: Amazon product URL
Output (in 10 seconds):
- ✅ BUY / ⚠️ CAUTION / ❌ SKIP verdict
- Trust Score (0-100) - Are reviews authentic?
- Confidence Score (0-100) - How certain is the AI?
- Deal Breakers - What could go wrong
- Perfect For - Who should buy this
- Risk Analysis - Durability, return risk, quality consistency

Why It's Different

Not just another fake review detector. We're building an AI Shopping Decision Copilot.

The Tech Stack (and Why We Chose It)

Frontend: Next.js 16 + React 19

• App Router for better performance and SEO

• Server Components for faster initial loads

• TypeScript for type safety (caught 50+ bugs before production)

AI: GPT-4.1 via Bytez SDK

• Why GPT-4? Best at nuanced analysis (not just sentiment)

• Bytez SDK for model routing and fallbacks

• Structured outputs using Zod schemas for strict JSON validation

Database: Supabase

• PostgreSQL for relational data

• Row Level Security for user data protection

• Real-time subscriptions (future: live analysis updates)

• Auth built-in (Google OAuth + Magic Links)

Analytics: PostHog + Supabase

• Dual tracking

  • PostHog → UX insights

  • Supabase → permanent audit logs

• Session replay to understand user behavior

• Event mirroring (saved to both platforms)

• Privacy-first filtering for development environment

Email: Resend

• Subdomain: notifications.reviewai.pro

• 100% deliverability (SPF, DKIM, DMARC configured)

• HTML templates (React Email had rendering issues in Next.js 16)

Hosting: Vercel

• Edge functions for geo-detection

• Image optimization for Amazon product images

• Analytics built-in

• Zero-config deployments

The Architecture

User Input
(Amazon URL or Extension)
 │
 ▼
Validation Layer
• Bot protection (botid)
• ASIN extraction
• URL validation
 │
 ▼
Data Collection
• Extension scraping (preferred)
• Server-side fallback
• Review deduplication
• Minimum 3 reviews required
 │
 ▼
AI Analysis
• Persona resolution
• GPT-4 prompt construction
• Structured JSON output
• Schema validation (Zod)
 │
 ▼
Persistence + Analytics
• Save to Supabase
• Track events (PostHog + Supabase)
• Generate public report URL
 │
 ▼
User Experience
• Report page (shareable)
• Dashboard (history + settings)
• Extension overlay (Amazon)

Key Technical Decisions

1. Why Next.js 16 App Router?

The Good

• Server Components = faster loads

• Built-in API routes

• Excellent SEO

• Seamless Vercel deployment

The Challenges

• React Email rendering issues

• Client/server component boundaries

• Caching behavior changes

2. Why Dual Analytics (PostHog + Supabase)?

PostHog

• Session replay

• Funnel analysis

• Feature flags (A/B testing)

• Heatmaps

Supabase

• Permanent audit logs

• SQL queries for custom analysis

• No data loss if PostHog quota exceeded

• Compliance (data ownership)

Implementation

// src/lib/analytics.ts
export function trackEvent(event: string, properties: object) {
 // Mirror to both platforms
 posthog.capture(event, properties);
 supabase.from('user_events').insert({ event, properties });
}

Try ReviewAI

Website:
https://reviewai.pro

Free tier includes:

• 10 analyses per month

• Chrome extension

• Full reports

• No credit card required

Questions I'm Happy to Answer

• How do you handle Amazon anti-scraping?

• How do you validate AI output?

• What's your go-to-market strategy?

• How do you compete with Fakespot?

Built with Next.js, GPT-4, Supabase, and way too much coffee.

Connect:

Website: https://reviewai.pro
Twitter/X: @aminnnn_09
Email: [email protected]