Trust AI UX patterns
Trust patterns help users calibrate confidence: citations, scores, human handoff, cost transparency, and auditability. Essential when mistakes are costly or regulated.
Related framework: Agentic UX framework
Essential patterns
Prove sources, show uncertainty, and disclose failure before users act.
- Trust
Citations
Attach verifiable sources to generated claims
- Trust
Confidence Score
Show how sure the model is about a claim
- Trust
Source Browser
Inspect source documents beside the answer
- Trust
Chain of Thought
Reveal step-by-step reasoning behind an answer
- Trust
Failure Disclosure
Honest signaling of AI limitations
- Trust
Data Ownership & Control
User control over AI data usage
26 patterns
Progress Steps
Collapsible thinking and tool traces
Citation Tooltips
Hover for source
Confidence Score
Show how sure the model is about a claim
Verification Next Steps
Concrete actions to validate uncertain output
Knowledge Graph
Visualizing RAG
Citations
Attach verifiable sources to generated claims
Privacy Filters
Masking PII
Source Browser
Inspect source documents beside the answer
Retrieval Context Preview
Preview retrieved context before acting on the answer
Chain of Thought
Reveal step-by-step reasoning behind an answer
Confidence Indicators
Visual confidence levels for outputs
Data Ownership & Control
User control over AI data usage
Bias Detection
Flag potentially biased outputs
Fact-Checking Indicators
Real-time fact-checking status
Source Quality Scores
Rate source reliability
Audit Trail
Complete log of AI decisions and data usage
Transparency Report
Periodic reports on AI behavior/accuracy
Scope Disclosure
Plain-language agent permissions
Permission Drift Indicator
Surface accumulated agent permissions
Failure Disclosure
Honest signaling of AI limitations
Duration-Bound Consent
Permissions that expire by default
Granular Consent
Per-capability toggles, not bundled grants
Authentication Chains
Legible identity trails across agent actions
Responsibility Attribution
Trace which agent or human caused each action
Agent Identity
Stable name, version, and capabilities per agent
Revocation Affordances
One-click revoke beside the grant
Frequently asked questions
Which trust patterns should ship first?
Lead with provenance (citations or source browser), explicit uncertainty, and a human escalation path. Add cost, autonomy, and audit signals when actions have side effects, spend, or compliance requirements.
When are trust patterns required versus nice-to-have?
Treat them as required for search, finance, health, legal, and enterprise knowledge products. Anywhere a wrong answer has external consequences. Consumer creative tools may start lighter but still benefit from confidence cues on high-stakes outputs.
How do citations differ from chain-of-thought UI?
Citations tie claims to external sources users can verify. Chain-of-thought shows reasoning steps the model took. They solve different doubts (source vs logic). Use both when answers are long or disputed.
What anti-patterns hurt trust in AI interfaces?
Fake certainty, buried sources, inconsistent confidence, and silent data use. Avoid decorative “trust” badges without actionable provenance or controls.
How do trust patterns connect to agentic products?
Agents amplify risk because they act, not only reply. Pair trust patterns with autonomy budgets, approvals, and audit trails from the Agentic UX framework so users can see and revoke what ran.
Do pattern pages include demos for trust flows?
Many trust patterns include interactive demos plus screenshots from products like Perplexity and Google AI Overviews. Open the essential patterns above for the highest-traffic conventions.