Most AI design critiques fail for one simple reason: We ask AI to critique everything at once.
Prompting AI “Review this design and tell me what’s wrong” sounds reasonable, but it usually produces a generic mix of accessibility, visual, content, and interaction feedback.
I get much better results with Claude Design by splitting critique into 4 focused passes:
1️⃣ Accessibility audit: Check contrast, touch targets, focus order, semantic hierarchy, and reliance on color.
2️⃣ Visual hierarchy critique: Ignore whether the UI looks attractive. Evaluate whether the layout reflects the real importance of information and actions.
3️⃣ Content critique: Review labels and UX copy for ambiguity, unnecessary complexity, length, and unclear consequences.
4️⃣ Interaction design critique: Look for missing states: loading, empty, error, success, disabled, focus, hover/pressed, etc.
The important part: fix issues after each phase before moving to the next one.
Structured critique → better feedback → better design
📕 Complete guide to design critique with Claude Design (with prompts samples and examples): https://lnkd.in/ez9eW5Rb