How to detect Cursor websites
AI-powered code editor that writes code alongside you
Cursor is an AI code editor — unlike Lovable or Bolt, it doesn't deploy your app for you. This means Cursor-built sites don't have hosting-layer fingerprints. Detection relies on code patterns and design cues that emerge from Cursor's common scaffolding suggestions, especially when users start with the 'new React app' or 'Next.js with shadcn' prompts.
Technical fingerprints
- ✓
cursor.sh or cursor.com/generated URLs in source (rare but definitive) - ✓
Vite + React + TypeScript boilerplate (most common Cursor scaffold) - ✓
shadcn/ui component patterns combined with Lucide icons - ✓
Over-commented code — Cursor often adds verbose inline comments - ✓
Highly uniform component naming: UserCard, ProductCard, DashboardCard - ✓
Deployment on Vercel or Netlify (common CI targets Cursor suggests)
Visual design patterns
- →Heavy shadcn/ui usage — Button, Dialog, Card, Badge, Input
- →Tailwind utility classes with high specificity (not @apply)
- →Dark mode by default with slate-900 background
- →Consistent 8px grid spacing (Tailwind default increments)
- →Lucide icons — especially ChevronRight, ArrowRight, Check, X
- →Toast notifications via Sonner or react-hot-toast
Typical tech stack
Frequently asked questions
Why is Cursor harder to detect than Lovable or v0?+
Because Cursor is an editor, not a deployment platform. There's no hosting signature, no generator meta tag, and no CDN fingerprint. Detection relies on the statistical patterns in the code Cursor tends to produce, which overlap heavily with hand-written code by experienced React developers.
What's the most reliable signal for a Cursor-built site?+
The combination of shadcn/ui + Lucide React + Tailwind on Vercel with a new domain is the strongest multi-signal indicator. Any single signal is weak. Our confidence score reflects this — a Cursor site typically lands in the 25–55% range unless there are many signals.
How is Cursor different from Windsurf or GitHub Copilot for detection?+
All three are code editors that assist rather than generate complete apps. Windsurf (by Codeium) has similar patterns. Copilot integrates into VS Code and leaves even fewer code-level fingerprints. The detection approach is the same for all: look for statistical patterns in the stack choices rather than definitive markers.
Detect other platforms