Scoring Show HN submissions for AI design patterns

Article

TL;DR

21% of Show HN submissions show 5+ AI-generated design patterns; dang wants the raw data.

Key Takeaways

• Heavy slop: 105 sites (21%), mild: 230 (46%), clean: 165 (33%) across ~500 Show HN submissions
• Each major AI model (Opus, GPT, Gemini) has distinct, detectable visual fingerprints
• Most flagged patterns — gradients, purple accents, rounded rects — predate AI; blame shadcn/TailwindUI

Discussion

Top comments:

• [simonw]: Side projects are ideal for AI tools; time-constrained work naturally uses available shortcuts
• [deaux]: Each model has its own visual archetype; mixing them into one score loses signal
• [dang]: Wants IDs for all 3 categories to build browsable HN URLs from the data
• [nomdep]: Flagged design patterns were common before AI — this conflates trends with AI output

Discuss on HN