67% of recent Show HN submissions carry detectable AI design fingerprints, according to an analysis by developer Adrian Krebs. After running 500 sites through automated DOM and CSS checks, Krebs found 21% had five or more what he calls "design slop" markers - the visual tells that a site was built by asking an LLM to generate a UI rather than making deliberate design choices.
The methodology is rigorous enough to take seriously. A headless browser loaded each site, an in-page script analyzed DOM structure and computed styles, and every pattern triggered a deterministic check rather than a visual judgment call. Krebs estimates a 5-10% false positive rate after manual verification.
The Patterns That Give It Away
The patterns break into four categories. On fonts: Inter for hero headlines, serif italic accents, and specific combinations like Space Grotesk paired with Instrument Serif. On colors: purple gradients, dark mode with barely-passing contrast ratios, glowing box shadows, and colored card borders. On layout: centered hero sections with small badge labels sitting above the headline, icon-topped feature cards, numbered steps sections, and all-caps labels. On CSS: shadcn/ui component defaults and glassmorphism (the frosted-glass card effect everywhere right now).
If you've been building with Bolt.new or similar AI app builders recently, some of those probably look familiar. The full breakdown: 105 sites (21%) had five or more patterns - heavy AI fingerprinting. 230 sites (46%) had two to four patterns - mild. 165 sites (33%) had zero or one - clean.
The Interesting Part Is Not the Slop Itself
Krebs makes a point worth sitting with: Bootstrap templates existed before LLMs, and nobody panicked. Generic design has always been available to builders who don't care about visual differentiation.
The actual shift is about defaults. Bootstrap was a deliberate choice - you reached for it because you wanted a grid system or pre-built components. When an LLM generates your UI inside Cursor or any AI coding tool, the design patterns arrive whether you wanted them or not. Purple gradients and glassmorphism aren't there because someone thought they fit the product; they're there because that's what the training data leans toward.
That matters most in crowded product categories. If your SaaS homepage looks identical to a competitor's because you both scaffolded the initial UI with an AI tool, you've given up a differentiation point for free. Users can't tell from a glance what makes you different.
The practical implication: before shipping, look at your font, color, and layout choices and decide if they're right for your product. Inter, purple gradients, and glassmorphism cards aren't bad - they're just increasingly invisible because everyone has them by default.