← Back to scanner

How Scores Are Calculated

Every scan produces four outputs: an Analysis Result (verdict), an AI Influence Score, Detection Evidence (certainty level), and Infrastructure Signals (hosting, framework, CDN, build pipeline).

What You See on the Result Page

Block 1

Analysis Result

The verdict label, platform description, and feedback buttons.

Block 2

AI Influence Score

0–99 score estimating AI involvement, shown as a large number and a visual meter.

Block 3

Detection Evidence

Signal types, certainty label (Strong / Moderate / Limited), and collapsible technical signals.

Block 4

Infrastructure Signals

Hosting provider, framework, CDN, and build artifacts detected from HTTP headers and page content.

1

Analysis Result — Verdicts

The verdict summarises the overall AI involvement level detected on the site. It is derived from the combined AI Influence Score after all signals are evaluated.

AI-Native Build

Confirmed by strong AI builder fingerprints or identifiable bundle artifacts from platforms like Lovable, Bolt, or v0. Requires concrete evidence — not just a modern stack.

AI-Augmented Development

AI coding tool patterns or artifact signals detected. The site likely used an AI coding assistant or vibe-coding workflow, but no specific AI builder was fingerprinted.

Modern AI-Compatible Stack

A modern developer stack was detected. The tooling is compatible with AI-assisted development, but no strong artifact or code signals were found.

Template-Based Build

Built with a website builder, CMS, or e-commerce platform. Template-driven, not AI-generated.

Manual / Custom Build

Backend framework detected alongside a low AI signal profile.

Inconclusive

Very low detection confidence and no strong signals found.

Traditional Architecture

No AI fingerprints, artifact signals, or AI code patterns detected.

Traditional platforms like WordPress, Shopify, Webflow, and similar builders always resolve to Template-Based Build regardless of score — template-driven tools are not AI-generated by definition.

2

AI Influence Score (0–99)

The AI Influence Score is assembled from multiple independent layers of evidence. Artifact detection is the primary driver — bundle fingerprints and AI code patterns carry the most weight. Component libraries and stack choices are supporting context, not primary evidence on their own.

Bundle Artifact Patterns

Primary

The strongest signal. Strings embedded in compiled JavaScript bundles that uniquely identify AI builder origins — even after minification. Platform-specific CDN domains and script paths also fall here.

AI Code Signals

Primary

References to AI SDK imports, chat completion APIs, LLM environment variables, and streaming response patterns in the page source or scripts.

Detected Platform

When a builder is fingerprinted with high confidence, its AI involvement baseline anchors the score. AI-native builders (Lovable, Bolt, v0) carry a high baseline. Traditional CMS platforms carry a low baseline.

Code Architecture

Presence of component libraries and stack combinations often seen in AI-generated codebases (Shadcn UI, Radix UI, Lucide). Treated as supporting context, not primary evidence.

Build Pipeline

Build tool artifacts, CI/CD deployment signals, and hosting patterns that provide supporting context for the verdict.

Score Meter (as shown on result page)

05099

Each segment = 10 points. Score is always capped at 99 — never 100%.

3

Detection Evidence & Certainty

The Detection Evidence block shows what was found and how strongly. A Certainty Label summarises the overall confidence of the platform match.

Certainty Labels

Strong evidence
detected

Multiple independent technical signals confirmed the platform with high reliability.

Moderate evidence
detected

Some signals were detected, but the site may use custom hosting or partially obscure its stack.

Limited signals
detected

The website may hide parts of its infrastructure, use custom hosting, or block detection signals.

Signal Types

Strong

Script Pattern

JS files loaded from known builder CDNs or with unique filename patterns

Strong

CDN Domain

Asset domains that uniquely identify a specific platform

Strong

HTTP Header

Server-sent headers that identify the platform or hosting environment

Medium

HTML Pattern

CSS class names, data attributes, and markup structures in the page source

Medium

Meta Tag

Generator meta tags or platform-specific description patterns

Weak

HTML Attribute

Generic attributes that suggest a platform but may appear in other contexts

Build Pipeline & Infrastructure

Beyond HTML and JS analysis, we examine the deployment environment itself. HTTP headers and asset URL patterns reveal the hosting provider and build toolchain — strong evidence even when code is fully minified.

Hosting provider

Detected from HTTP response headers and server identifiers

Build framework

Identified from asset paths, script names, and page structure

Asset CDN patterns

Some AI builders serve assets from unique, identifiable CDN domains

Build tool artifacts

Minified bundles often retain strings that reveal the original build tool

Hosting providers like Vercel, Netlify, and Cloudflare are treated as infrastructure context — not as evidence of AI development. Many human-built sites use the same platforms.

Why We Can't Be 100% Certain

AI likelihood is a probabilistic estimate, not a lookup. Every result is an inference based on observable signals — and signals can be hidden, shared, or absent.

We analyse the rendered page, not the source code

We fetch what a browser sees — compiled, minified production output. We never access source files, build config, or git history. Developers can remove platform references during build.

Single-page apps execute code in the browser

React, Vue, and Svelte SPAs render via JavaScript running in the browser. Our scanner fetches the HTML source without executing JS — so dynamically loaded content may be invisible to us.

AI coding tools leave no traceable fingerprint

Sites built with Cursor, GitHub Copilot, or ChatGPT produce standard framework output. There is no 'Copilot tag' in the HTML — we can identify the stack pattern but not the specific tool.

Custom-domain deployments strip platform branding

When AI builder sites are deployed to custom domains, platform-specific scripts are often replaced with self-hosted assets. Minification and tree-shaking can further remove identifiable strings.

Many platforms share the same infrastructure

Cloudflare, Vercel, and AWS are used by almost every platform — AI-built or not. Shared signals increase uncertainty, and we weight them accordingly.

Detection is pattern-based, not deterministic

Every signal is a statistical indicator — not a signed certificate. We weight and cross-reference signals, but false positives and false negatives are unavoidable in any pattern-based system.

Bottom line: AI Influence Score is a best-effort estimate. Use the result as a signal, not a verdict. The reasoning panel on each scan explains what was detected and why.

Common Questions

Why can WordPress get high detection confidence but a Human-Engineered verdict?

Confidence and AI Influence Score are independent. We can be very sure a site is WordPress (high detection confidence) while still classifying it as Human-Engineered because WordPress is a traditional CMS — it doesn't meet the threshold for any AI verdict.

Can Webflow or Wix ever show AI-Assisted Development?

Not from builder detection alone. Visual builders and website builders have a low AI baseline. Only strong additional signals — like a modern vibe-coding stack or pipeline artifacts — could raise the score to that level.

Why does a modern React / Next.js stack sometimes show AI-Compatible Stack?

A modern developer stack is compatible with AI-assisted development — but using React or Tailwind alone is not treated as AI evidence. The score only rises for AI-Augmented or higher when actual AI artifact signals are detected (SDK imports, AI API patterns, or specific bundle fingerprints).

Can the score ever be 100%?

No. The AI Influence Score is always capped at 99 because fingerprint-based detection is probabilistic — some signals can appear in non-platform code.

What if nothing is detected at all?

The result defaults to a low AI score, limited certainty, and a Traditional Architecture or Inconclusive verdict. This can mean a custom stack, an unsupported platform, or stripped fingerprints.

Does a Human-Engineered verdict mean AI was never used?

No. The verdict reflects observable platform fingerprints only. Developers can use Cursor, GitHub Copilot, or ChatGPT on any stack — a custom-coded site built entirely with AI assistance will still show Human-Engineered if no platform fingerprint is detected.