I Built an AI Visibility Monitoring SaaS in 2 Days (After 10 Days of Research)
I Built an AI Visibility Monitoring SaaS in 2 Days (After 10 Days of Research)
Two days ago, I had a PRD, an architecture document, and zero lines of code.
Today, I have a fully deployed SaaS with Stripe billing, 8 AI platform integrations, 347 tests, and a staging environment that's live and accepting signups.
The Problem I'm Solving
AI search is quietly replacing Google for millions of people. When someone asks ChatGPT "What's the best coffee shop in Manchester?" or "Which SaaS handles invoice automation well?" — your business is either mentioned, or it isn't. And right now, most businesses have zero visibility into what AI is saying about them.
AImpact Monitor fixes that. It scans 8 AI platforms (ChatGPT, Perplexity, Gemini, Claude, Copilot, Meta AI, Grok, and more), runs each query 3 independent times for statistical confidence, verifies every citation, and... this is the key bit... proves which optimisations actually moved the needle.
We call that last part the Attribution Loop. It's our north star metric and the thing that makes this more than just another monitoring dashboard.
10 Days of Research (The Invisible Work)
Before I wrote a single line of code, I spent 10 days on:
- Product Requirements Document — 63 features mapped, 41 marked P0 for MVP, 23 test scenarios defined
- Ideal Customer Profile — deep dive into who this is for (solo founders, local businesses, agencies)
- Brand & Design System — Proof Green (#0F766E), consistent design language, email templates
- Architecture Design — 12-entity database schema, 12 state machines, API design, scan engine spec
- Pricing Strategy — Solo ($4.95), Growth ($14.95), Pro ($29.95) with monitoring guarantee
This research wasn't optional. It's the reason the build went fast. Every decision during implementation had already been made. I wasn't designing and coding simultaneously... I was just executing a plan.
2 Days of Building (The Visible Work)
Day 1: Foundation Through Dashboard
Phase 1 kicked off with Next.js 16, Supabase, and Railway. Within a few hours I had auth working (email + Google), a 16-table database schema deployed, pg-boss job queues configured, and a health check returning 200 from staging.
Phase 2 was the scan engine... the technical heart of the product. The N=3 execution model means every prompt runs 3 independent times against each AI platform. If 3 out of 3 mention your business, that's "High" confidence. 2 out of 3 is "Medium". This isn't a single API call and a prayer... it's validated data.
The citation verifier was satisfying to build. It does a HEAD request to check if a URL exists, then a GET with a 1MB body cap to check if your brand name actually appears on the page. Citations get classified as verified, hallucinated, broken, or partial. When AI says "according to acme.com" and that page doesn't mention your brand at all, that's a hallucination... and now you know.
Phase 3 brought the dashboard to life... score cards, trend charts, domain management, scan results with confidence badges, citation detail views. All with mock data for now, but structurally complete.
Phase 4 was the Attribution Loop, the feature that makes this product different from "here's what AI said about you" tools. Users log an optimisation (e.g., "Added FAQ schema markup"), we capture a baseline scan, then compare post-optimisation scans. If the score improved, you get an attribution receipt... visual proof that your specific change caused a specific improvement, with a confidence score.
Day 2: Billing, Launch, and Everything Else
Phase 5 wired up Stripe (3 products, 6 prices, webhook handler with idempotency), Resend for transactional emails (5 templates), the monitoring guarantee system, onboarding flow, and a free scan entry point.
Phase 6 was security review, testing, and deployment. I found and fixed 3 critical security issues (SSRF in the citation verifier, Stripe redirect URL spoofing, email template XSS), wrote 35 additional tests, built the landing page, configured security headers, and deployed to Railway with all 7 external services connected.
By end of day 2, the staging environment was live with working auth, Stripe checkout, and AI scans.
The Numbers
| Metric | Value |
|---|---|
| Total tests | 347 |
| Routes | 38 |
| Build time | 2.7 seconds |
| JS bundle | 2.1 MB |
| External services | 7 (Supabase, OpenRouter, Stripe, Resend, Upstash, PostHog, Railway) |
| Database tables | 16 |
| State machines | 12 |
| API endpoints | 13 |
| Phases completed | 6 of 6 |
What Made This Possible
1. Research-first, code-second. 10 days of research meant zero architectural debates during implementation. Every database column, every API endpoint, every state transition was pre-decided.
2. AI-assisted development. I used Claude Code as my development partner. It handled implementation while I made product decisions. The code quality is high... 347 tests, TypeScript strict mode, proper security practices... because the AI follows the architecture spec consistently.
3. The right stack. Next.js 16 + Supabase + Railway is a deployment machine. Push to git, Railway builds and deploys. Supabase handles auth and Postgres. The entire infrastructure is managed... I didn't configure a single server.
4. Lazy initialisation everywhere. Every external service (Stripe, Resend, OpenRouter) uses a lazy proxy pattern. The app builds and runs even when API keys aren't set. This meant I could deploy early and add services incrementally.
What's Next
The staging environment is live. The next steps are:
- Test the full user journey on staging with real Stripe test cards
- Wait for the custom domain SSL to provision (Railway is working on it)
- Set up production Stripe (live mode) and deploy to production
- Get the first beta customers onboarded
- Week 4: Attribution Accuracy Gate... does the attribution engine hit >70% accuracy?
- Week 6: Alert Relevance Gate... do >30% of alerts lead to user action?
The product is built. Now comes the harder part... proving it works with real users.
Try It
The staging environment is live at aimpactmonitor-staging.up.railway.app. The free scan page lets you see what AI says about your domain... no account needed.
If you're a small business wondering what ChatGPT tells people about you, this is the tool that will tell you... with proof.
Building in public. Follow the journey on X/Twitter and LinkedIn.