Communicate evidence and uncertainty at the decision point
AI output distinguishes source uncertainty, model uncertainty, stale data and conflict without disguising them as fluent certainty.
Guidance
Prefer provenance, observation dates and conflicting evidence over unexplained confidence scores. Calibrate language and required review to consequence.
Do
- Attach sources and freshness where decisions depend on them
- Expose conflicts and missing evidence
- Escalate review for high-stakes uncertainty
Do not
- Invent confidence percentages
- Present partial output as verified
- Use reassuring tone to mask unknowns
Verification contract
Stale, conflicting, absent-source and high-consequence scenario review
Supporting references
ref.anthropic.building-effective-agents— Building Effective Agents — Anthropic Engineeringref.apple.hig-machine-learning— Apple HIG — Machine Learningref.google.pair-guidebook— Google PAIR Guidebookref.ibm.carbon-for-ai— Carbon for AIref.intuit.content-design— Intuit Content Designref.intuit.design-system— Intuit Design Systemref.mailchimp.style-guide— Mailchimp Content Style Guideref.nist.ai-rmf— NIST AI Risk Management Frameworkref.uber.base— Uber Base design system