Study Finds No Link Between Trust and AI Verification in Chatbot Use

A recent study released on arXiv disputes the traditional view that verifying AI system outputs relies on trust. This research, based on a mixed-methods survey involving 153 regular chatbot users, discovered no significant link between trust and verification, with findings remaining consistent across various sensitivity checks. The study highlights three user practices—refinement, correction, and approval prior to automated actions—that are widely accepted and correlate positively with user satisfaction. It distinguishes between evaluative oversight, which is weakly related to satisfaction, and interventionist oversight, which shows a strong connection to satisfaction. A medium-to-large gap in satisfaction control indicates that successful task outcomes do not equate to a perceived sense of agency, suggesting a need to view verification as routine epistemic governance instead of a trust-dependent action.