Human-Centered AI
People + AI Guidebook
Maintained by Google PAIR
Google’s research-backed patterns for designing AI-powered products, covering mental models, explainability, feedback, and graceful failure.
Why it matters
AI features fail on trust, not accuracy. PAIR is the most complete public answer to "how do we set expectations, show confidence, and recover when the model is wrong".
Key principles
- User Needs + Defining Success - confirm AI is genuinely the right solution before committing to it.
- Data Collection + Evaluation - your training data is a design material; its gaps become user-facing failures.
- Mental Models - set accurate expectations early so users know what the system can and cannot do.
- Explainability + Trust - show the why behind output, and calibrate confidence honestly rather than optimistically.
- Feedback + Control - always give users a way to correct, override, or opt out of automated decisions.
- Errors + Graceful Failure - design the wrong-answer path with the same care as the happy path.
Resources
Foundations, principles, tutorials, community Figma files, icon sets, and templates.