Every scan returns three things — a Yes / Maybe / No verdict, an AI Score (how AI-native the platform is), and a Detection Accuracy (how confident we are). Here's exactly how each works.
The verdict is the single most important output. It answers: "Was this website built with AI?" — combining both the AI Score and Detection Accuracy into one clear answer.
AI-first builder confirmed
The site was built with an AI-native builder (Lovable, Framer, Bolt, Durable…) and our engine is sufficiently confident in the detection. AI Score ≥ 75, Accuracy ≥ 35%.
Possibly AI-assisted
The platform has AI features, uses AI coding tools (GitHub Copilot, Cursor, v0), or is an AI builder but detected with low confidence. AI Score 40–74, or AI builder with <35% confidence.
Traditional or undetected
The site uses a traditional CMS (WordPress, Drupal), an e-commerce platform (Shopify, WooCommerce), or nothing was detected at all. AI Score < 40.
The verdict is intentionally conservative. A "No" doesn't mean the site wasn't built using AI tools — developers may use AI coding assistants for any codebase. It means we couldn't confirm an AI-first builder was used.
We think about every website in one of three ways:
Tools like Lovable, Framer, Bolt, Durable, 10Web generate websites from prompts or visual input using AI at their core. These get a high AI Score (75–90) and — when detected — a high accuracy because they leave clear fingerprints (unique CDN domains, proprietary script patterns, meta tags).
Sites built on WordPress, Wix, Webflow, Shopify have rich, well-documented fingerprints — so accuracy is typically high. However, these platforms are not AI-native, so the AI Score is low (20–45) and the verdict is "No". The website may still have been built with the help of AI tools, but the hosting platform itself isn't AI-generated.
Developers using GitHub Copilot, Cursor, Codex, or similar AI coding assistants don't leave a consistent fingerprint — the output is bespoke code. If we detect signals of these tools (e.g. v0 by Vercel, Replit), the AI Score lands in the mid-range (65/100) and the verdict is "Maybe". If nothing is detected, the site gets an AI Score of 10 and verdict "No" — but that doesn't rule out human or AI coding.
The AI Score reflects how AI-native the detected platform is. It comes from the platform category — not from detection confidence. A WordPress site detected at 99% accuracy still has a low AI Score (20/100) because WordPress predates the AI era.
AI-Native (85–100): Fully AI-generated sites. Lovable, Framer, Bolt, Durable.
AI-Assisted (65–84): Platforms with AI features or AI coding tool fingerprints.
No-Code (40–64): Visual builders that aren't AI-first. Wix, Squarespace.
Traditional (20–39): CMS and e-commerce platforms. WordPress, Shopify.
Unknown (10): Custom code or undetected platform.
Detection Accuracy measures how confident our engine is in the detection. It is based on the quantity and quality of signals found in the scanned page. Accuracy is independent of AI Score — a CMS site can have 95% accuracy (we're very sure it's WordPress) while still scoring 20/100 for AI.
Script Pattern
JS files loaded from known builder CDNs or with unique filename patterns
CDN Domain
Asset domains that uniquely identify a specific platform
HTTP Header
Server-sent headers like X-Powered-By, Server, or custom platform headers
HTML Pattern
CSS class names, data attributes, and markup structures in the page source
Meta Tag
Generator meta tags, og:platform, or description patterns
HTML Attribute
Generic attributes that suggest a platform but may appear in other contexts
The raw signal score is normalized to 0–99%. We never show 100% to reflect that fingerprint-based detection is probabilistic — some platforms share technology stacks and false positives are possible.
Accuracy and AI Score are independent. A 99% accuracy means we're very sure the site uses WordPress. "No" is the correct verdict because WordPress is a traditional CMS — it wasn't built by AI, even if the developer used AI tools while building the site.
Only if we detect signals of a known AI builder. A fully bespoke React or Next.js app will get "No" or "Maybe" depending on whether any AI coding assistant fingerprints are found. Our engine can't look at raw code and determine it was written by AI.
No — we cap accuracy at 99% to avoid overconfidence. Detection is signal-based, not infallible.
The site gets AI Score 10, Accuracy 0%, and verdict "No". This typically means the site uses a custom framework, a platform not yet in our database, or has minimized its technology fingerprints.