The Act / Review / Ignore rule
One score, one action. An unexplained 87% is a shrug with decimals — it tells the user how the model feels and nothing about what to do. Every confidence surface I ship resolves to exactly one of three verbs: act, review, or ignore.
Confidence Score Patterns
Five ways to put a number on certainty — numeric, color, language, composite — and when each one earns or burns trust.
AI Failure States
What the screen says when the model can't deliver, and how saying it honestly keeps users from leaving.
ML Explainability
Showing non-technical people why the machine decided, at a depth they can use without a stats degree.
Human-in-Loop Patterns
Keeping the person in command when the stakes outgrow the model's confidence.
Provenance & Citations
Tracing every AI claim back to the exact source — so a person acts on what they can open, not a summary.
The Capability Contract
Saying what the system can't do, up front — the honest "no" that makes the confident answer believable.
Calibration & Track Record
Showing whether the confidence has actually been right before, so trust is earned, not assumed.
Reversibility
Making the model's suggestion cheap to walk back, so people dare to act on it.
Maheshwari, A. (2026). A Field Guide to Trust: AI Design Patterns. arpitmaheshwari.com/patterns/.
Prefer it as a book? The same patterns run as live demos in the field guide →