| name | ai-ui-heuristics |
| description | Nielsen's 10 usability heuristics and Don Norman's 7 principles, reinterpreted for AI-native interfaces. Use when evaluating, auditing, or reviewing the UX of any AI-powered application. Each heuristic includes the AI-specific interpretation, a common failure mode, and a concrete fix. Triggers on: "heuristic evaluation", "Nielsen heuristics", "UX audit", "review this AI UI", "is this good design", "Don Norman principles".
|
AI UI Heuristics
Nielsen's 10 heuristics and Don Norman's 7 principles, reinterpreted for AI-native interfaces.
Nielsen's 10 Heuristics for AI
Source: Jakob Nielsen, Nielsen Norman Group — https://www.nngroup.com/articles/ten-usability-heuristics/
H1: Visibility of System Status
AI-specific: Show when AI is processing, which model is used, confidence level, estimated time.
Common failure: User submits prompt, sees spinner for 30s with no idea if AI is working, stuck, or almost done.
Fix: Ambient activity indicators (subtle glow). If >5s, show updating messages ("Analyzing...", "Almost there...").
H2: Match Between System and Real World
AI-specific: Natural language prompts, not technical parameters. Name agents by role ("Copywriter" not "LLM-7B"). Avoid AI jargon ("inference", "token", "embedding").
Common failure: Interface requires users to know about "temperature", "max tokens", or "system prompts."
Fix: Hide all technical parameters behind an "Advanced" toggle. Default everything to sensible values.
H3: User Control and Freedom
AI-specific: Stop AI mid-generation. Undo every action. Edit outputs inline. Version history.
Common failure: AI generates 500 words and the user can't stop it, can't edit it, and can't go back.
Fix: Add a stop button. Make all output editable. Keep version history.
H4: Consistency and Standards
AI-specific: Same AI behavior for same inputs (or explain why different). Predictable output format every time.
Common failure: The AI gives completely different results for the exact same prompt, confusing users.
Fix: Use deterministic parameters (seed, temperature=0) for consistent outputs. If variation is expected, tell the user.
H5: Error Prevention
AI-specific: Validate inputs before sending to AI. Show cost estimates. Confirm high-risk actions.
Common failure: User accidentally pastes a 10,000-word document into a prompt and gets charged $5 for a single request.
Fix: Show input length, estimated cost, and a confirmation for expensive operations.
H6: Recognition Rather Than Recall
AI-specific: Show AI capabilities as buttons, not requiring users to remember. Show example prompts for common tasks.
Common failure: User doesn't know what the AI can do because there are no visible cues or examples.
Fix: Add a "What can I do?" button with example prompts. Include contextual hints that appear on hover.
H7: Flexibility and Efficiency of Use
AI-specific: Keyboard shortcuts for power users. Saved prompt templates. Customizable agent behaviors.
Common failure: Power users must click through the same flow 50 times a day with no shortcuts.
Fix: Add keyboard shortcuts (Cmd+K for command palette), saved templates, and batch operations.
H8: Aesthetic and Minimalist Design
AI-specific: No dashboard cards showing agent status. No permanent sidebars with AI controls. One primary action per screen. Generated content is the hero.
Common failure: Screen has 4 agent cards, 3 progress bars, 2 metric panels, and a chat sidebar — with the actual output squeezed into a corner.
Fix: The output takes 80% of the screen. Everything else is hover/gesture-revealed.
H9: Help Users Recognize, Diagnose, and Recover from Errors
AI-specific: When AI fails, explain why in plain language. Suggest what to try next. Show partial results if available.
Common failure: "An error occurred. Please try again." — with no indication of what went wrong or how to fix it.
Fix: "The AI couldn't process your image because it was too large (max 5MB). Try compressing it or using a smaller image."
H10: Help and Documentation
AI-specific: In-context help (tooltips, hints). Example prompts for every feature. "How to write good prompts" guide accessible from the UI.
Common failure: User has to leave the app and search Google to figure out how to use the AI features.
Fix: Add a "Tips" button that shows example prompts. Include contextual hints that appear on hover.
Don Norman's 7 Principles for AI
Source: Don Norman, "The Design of Everyday Things"
- Discoverability — Can users figure out what the AI can do without tutorials?
- Feedback — Is there immediate, clear feedback for every AI action?
- Conceptual Model — Do users understand how the AI works (even if simplified)?
- Affordances — Are AI capabilities visible as interactive elements (not hidden)?
- Signifiers — Do visual cues clearly indicate what is AI-generated vs human?
- Mapping — Do controls map naturally to AI actions (e.g., "regenerate this")?
- Constraints — Are dangerous AI actions prevented or gated?
Cognitive Load Check
The interface should not require users to:
- Remember information across screens (violation: recognition > recall)
- Make more than one decision per screen (violation: Hick's Law)
- Process more than 4-7 visible items at once (violation: Miller's Law)
- Learn a new interaction model just for AI features (violation: consistency)
Related Skills
- For calm technology test → use
calm-technology-ui
- For common failure catalog → use
ai-ui-anti-patterns
- For generative UI patterns → use
ai-native-ui-design
- For multi-agent UI → use
agentic-ui-patterns