Scoring Show HN submissions for AI design patterns

Article

TL;DR

21% of recent Show HN sites show 5+ AI design fingerprints; colored left-borders and Space Grotesk are the tells.

Key Takeaways

• Deterministic CSS/DOM checks flagged 21% heavy slop, 46% mild — no screenshot LLM judging needed
• Many ‘AI patterns’ (gradients, centered hero, shadcn) predate AI — attribution is noisy
• HN’s dang offered to create browsable category URLs if author open-sources the scoring project

Discussion

Top comments:

• [jerf]: 10k lines meant months of work in 2016; today it means one weekend and a credit card

  In 2016, if I saw 10,000 lines of code, that carried a certain proof-of-work with it. In 2026, 10,000 lines of code means they spent a minimum amount of money.

• [tptacek]: AI look-feel is fine; lack of commitment and thought is the real signal
• [raincole]: These patterns were common pre-AI too; no baseline comparison makes the score meaningless
• [dang]: HN already added showlim for new accounts, which bent the submission curve

Discuss on HN