Cursor, v0, and Claude generate beautiful-looking interfaces. But they silently break UX fundamentals that tank conversion rates.
I compiled 168 UX/UI principles from 2,098+ peer-reviewed academic papers. Here are the 5 most common violations I see in AI-generated UI.
1. Hick's Law — Too Many Options
AI tools love generating navigation with 12+ items, dashboards with 15 actions visible at once, and settings pages with every option exposed.
The research says: Decision time increases logarithmically with the number of choices. More options = slower decisions = higher abandonment.
The fix: Group options into 5-7 categories max. Use progressive disclosure to hide complexity until the user needs it.
2. Missing Feedback Loops
AI-generated forms rarely include inline validation, loading states, or success confirmations. The user clicks submit and... nothing visible happens for 2 seconds.
The research says: Users need system status feedback within 400ms or they assume something is broken.
The fix: Add optimistic UI updates, skeleton loaders, and explicit success/error states. Every action needs a visible reaction.
3. Cognitive Overload in Dashboards
AI tools generate dashboards that show every metric at once. No hierarchy, no progressive disclosure, no clear primary action.
The research says: Working memory holds 4±1 chunks of information. Beyond that, comprehension drops sharply.
The fix: Lead with 1-3 key metrics. Let users drill down for details. Design for scanning, not reading.
4. Fitts's Law Violations — Tiny Touch Targets
AI-generated mobile layouts frequently produce buttons and links under 44px. Looks clean in a mockup, fails in real thumbs.
The research says: The time to reach a target is a function of distance and size. Smaller targets = more errors = more frustration.
The fix: Minimum 44x44px touch targets (WCAG 2.5.5). Increase padding, not just font size.
5. No Visual Hierarchy
AI treats every element with equal visual weight. Same font sizes, same spacing, same button styles for primary and secondary actions.
The research says: Users scan in F-patterns and Z-patterns. Without clear hierarchy, they miss your primary CTA entirely.
The fix: One primary action per screen. Use size, color, and whitespace to create a clear visual path.
How I Validate AI-Generated UI
I built free tools to catch these problems automatically:
- AI Design Validator — Paste your design decisions, get matched with research-backed principles
- UX Smell Detector — 8 anti-patterns with step-by-step refactoring recipes
- AI Prompts Library — 600+ prompts to paste into Cursor, v0, or Claude for UX-aware code generation
All free, no signup required.
The full Principles Library has all 168 principles with examples, anti-patterns, and citations if you want to go deeper.
What UX violations do you see most often in AI-generated code? Curious what others are running into.
Top comments (0)