Slop*Detector

INCONCLUSIVE — slop score 47/100

A handful of tells surfaced (feature-tile grid, gradient backgrounds / gradient text, untouched shadcn/tailwind defaults), but not enough density to convict. The score of 37/100 says suspicious habits, not a generated site.

Findings

• Feature-tile grid — Equal-width card grids of 3–4 identical tiles (icon, subheader, blurb) are the signature layout block of generated landing pages.
• Gradient backgrounds / gradient text — Large gradient washes and gradient-clipped headline text (usually blue→purple) are the default decoration of generated UI.
• Untouched shadcn/Tailwind defaults — shadcn components, lucide icons, and Inter/Geist/Space-Grotesk font stacks left at their defaults mean the model picked the design, not a person. Instrument Serif as an accent font is a separate, independently corroborated tell.
• Vision analysis (placeholder) — No ANTHROPIC_API_KEY configured. With a key, Claude examines the screenshot for AI-generated imagery (garbled text in images, the DALL-E look, unfinished edges), hero/content mismatch, redundant sections, and fake demo personas.
• One/two-word punch sentences — Single dramatic words dropped as their own sentence ("Done." "Gone." "Period.") are an LLM emphasis habit, not a human one.
• Compulsive rule-of-three — "Fast, reliable, and secure." Triads everywhere read as machine rhythm — humans vary list length; models default to three.
• Writing analysis (placeholder) — No ANTHROPIC_API_KEY configured. With a key, Claude reads the full page copy against the inverted anti-AI style guide: rhythm, coherence between sections, specificity, and tells the regex pass can’t judge in context.