The receipts.
Version 1.3.1 uses 70 checks across page source and browser-generated HTML, styles and scripts. Each match includes the evidence it found.
Up to 20 page attempts, including failures, within 42 seconds. Only direct links on the same hostname and scheme are sampled; no recursive discovery. Queries, downloads, login and action links are skipped. Each sampled page has its own score.
The overall score and category bars use the strongest scored page.
Each family has a capped contribution so a cluster of related CSS effects does not get an unlimited vote. The overall score and category bars come from the strongest scored page in the sample. Simpler linked pages do not dilute a detailed homepage scan. Evidence combines matches across scored pages, and the report identifies which page supplied the score. Explicit AI-assistance disclosures are reported separately.
How points work
| Family | Maximum | Examples |
|---|---|---|
| Copy | 20 | Headline formulas, repeated phrases, paragraph uniformity |
| Layout | 25 | Hero copy + visual, CTA pairs, feature grids, section labels |
| Style | 20 | Fonts, whitespace, gradients, glass, tracking |
| Motion | 20 | Reveals, float effects, terminals, SVGs, mockups |
| Code | 15 | Class vocabulary, div composition, design tokens |
Checks have weights of 1–4. A family's points equal its maximum times the smaller of 1 and its matched weight divided by a saturation threshold: 12 for copy, style and motion; 14 for layout; 10 for code. We round each family's contribution and add them. Unavailable checks do not match and are listed in coverage; missing assets can understate resemblance. English-specific copy checks are skipped for explicitly non-English pages.
The display bands (under 40, 40–69, and 70–100) are descriptive UI labels, not calibrated authorship thresholds. We do not claim a detection accuracy, false-positive rate, or probability from these bands.
All 70 checks
Copy · 14 checks
- copy01Generic marketing vocabulary
- copy02Built for / designed for slogan
- copy03Repeated “Your…” headline
- copy04One / every headline formula
- copy05“Not just” / “more than” positioning
- copy06Repeated “No…” benefits
- copy07Mostly short slogan headings
- copy08Uniform paragraph lengths
- copy09Launch-in-minutes language
- copy10All-in-one / next-generation claims
- copy11“Say goodbye / hello” construction
- copy12“Whether you…” positioning
- copy13“Everything you need” slogan
- copy14“Ready to…” closing question
Layout · 14 checks
- layout01Explicit hero component naming
- layout02Text alongside a hero illustration
- layout03Paired hero calls to action
- layout04Eyebrow / kicker labels
- layout05Intentional headline line breaks
- layout06Accented word in main headline
- layout07Hero status / pill badge
- layout08Repeated feature-card components
- layout09Bento grid naming
- layout10Zero-padded process numbering
- layout11Alternating visual/content sections
- layout12FAQ / accordion component
- layout13Closing CTA section
- layout14Highlighted pricing plan
Style · 14 checks
- style01Dark neutral CSS palette
- style02Ambient radial gradients
- style03Decorative linear-grid background
- style04Frosted / glass panels
- style05Gradient-clipped text
- style06Large rounded cards
- style07Pill-shaped styling
- style08Inter / geometric sans font family
- style09Monospace accent font
- style10Oversized fluid headlines
- style11Large section whitespace
- style12Tight headline tracking
- style13Translucent hairline borders
- style14Glow / orb class vocabulary
Motion · 12 checks
- motion01Scroll reveal naming / attributes
- motion02IntersectionObserver entrance effects
- motion03Staggered animation delays
- motion04Floating / pulsing keyframes
- motion05Infinite ambient animation
- motion06Repeated hover-lift styling
- motion07Marquee / ticker / logo strip
- motion08Terminal mockup component
- motion09Dashboard / browser mockup
- motion10Repeated inline SVG icons
- motion11Canvas-based visual
- motion12Orbit / isometric decoration
Code · 16 checks
- code01Hero content / visual naming pair
- code02Section-header component naming
- code03Primary / secondary button naming
- code04CSS design-token vocabulary
- code05Div-heavy source composition
- code06Many styled div wrappers
- code07BEM-style component vocabulary
- code08Utility-heavy spacing / grid classes
- code09Lucide / Feather icon signature
- code10Commented section scaffolding
- code11Shared container naming
- code12Animation helper naming
- code13Card / icon / body naming family
- code14CSS breakpoint repetition
- code15Heading / subtitle class pairing
- code16Semantic accent / muted tokens
What does not increase the score
An AI product's subject matter, a domain suffix, hosting provider, framework name, HTTPS, responsive behavior, accessibility features, or valid SEO metadata are not independent authorship evidence. We keep placeholder-content observations outside the score too. A specific styling or icon signature can appear as a weak resemblance feature; its presence alone proves nothing.
Known limits
These positive examples are a starting reference, not a balanced training dataset. Some are related projects. There is no held-out human-versus-AI benchmark yet. Human SaaS pages can score high, while plain AI-built forms can yield no score. The scanner observes delivered source, not design intent, computed geometry, screenshots, or complete bundled applications.
Before publishing accuracy claims, we need documented human-made and AI-made comparison pages, a split by author and template family, and independent false-positive and false-negative evaluation. Until then, the evidence list matters more than the big number.