Illustrated flow
Researcher illustration
HAI researcher
User iconUser thread
IntentConstraintsTone
Agent iconAgent thread
PlanValidateRepair
Output iconOutput thread
Next stepChecklistAction
Practitioner lane~5 min

Trust Recovery Playbook for AI Assistants

Use this when confidence has dropped after poor outputs.

5-step reset

  1. Tag high-risk intents and enforce explicit uncertainty language.
  2. Add source visibility for factual claims.
  3. Add confirmation prompts before consequential actions.
  4. Collect post-response trust feedback.
  5. Run weekly trust trend review with product + support.

Success criterion

Trust score improves for 2+ consecutive weeks without reducing task success.