AI Tutors in Technical Training: Trust, Feedback Timing, and Skill Transfer in Maintenance Tasks

Technical maintenance training requires learners to develop procedural accuracy, diagnostic reasoning, error recovery, and confidence under realistic conditions. This study examined how AI tutor feedback timing influences trust, over-reliance, and skill transfer in maintenance tasks. Participants completed procedures under three conditions: immediate feedback, delayed feedback, and adaptive feedback triggered by hesitation, repeated errors, or safety-critical risk. Outcomes included training accuracy, recovery quality, completion time, cognitive workload, tutor trust, feedback acceptance, and unaided transfer performance. Immediate feedback improved early accuracy and reduced procedural errors but increased step-by-step dependence and weakened independent troubleshooting. Delayed feedback produced mixed effects, supporting autonomy for some learners while increasing frustration and avoidable mistakes for others. Adaptive feedback produced the strongest results by allowing independent attempts while intervening when support was needed. It improved unaided transfer, error recovery, procedural sequencing, and trust calibration. Overall, AI tutors were most effective when feedback scaffolded learning without replacing trainee judgment.