Adrian Krebs ran 500 Show HN landing pages through a headless browser and found that two-thirds carry detectable AI design fingerprints. He calls it 'slop.' Twenty-one percent of sites triggered five or more of his 15 identified AI patterns, earning the 'heavy slop' label. Another 46% hit two to four patterns. Only a third came back clean.
The telltale signs are specific. Inter font on centered hero sections. Colored borders on cards, usually top or left. Glassmorphism. Gradient everything. shadcn/ui components. And a particular shade of purple so common that designers now call it 'VibeCode Purple.' Krebs built his detector with Playwright, running deterministic CSS and DOM checks rather than feeding screenshots to an LLM. He estimates a 5-10% false positive rate based on manual review.
Show HN submissions tripled. The spike lines up with Claude Code and similar tools letting people ship projects fast. HN moderators responded by restricting Show HN posts from new accounts, a practical admission that the flood was drowning out everything else. Krebs himself acknowledges this isn't really bad, just uninspired. Before AI, everything looked like Bootstrap.
But there's a deeper issue. GitHub reported that 92% of developers now use AI coding tools, and researchers at Cambridge and Princeton have shown that models trained on their own outputs lose 10-15% diversity per generation. What Krebs spotted on Hacker News is that feedback loop playing out live, in purple gradients and identical feature card grids. He figures people will eventually craft distinct designs to stand out from the slop. That assumes humans are still the audience.