Automation bias in AI-Assisted antibiotic prescribing decisions among final-year dental students: a scenario-based experimental study

Inappropriate dental antibiotic prescribing drives antimicrobial resistance, and the integration of artificial intelligence (AI) decision support introduces critical risks of automation bias. This study evaluated how simulated AI recommendations influence prescribing decisions and confidence among final-year dental students. In a prospective, within-subjects study, 59 senior dental students completed 27 clinical vignettes across two phases: unaided (Part 1) and AI-assisted (Part 2). A simulated decision-support interface presented pre-scripted correct advice in 18 scenarios and deliberately incorrect advice in 9. Decisions and confidence (1–5 Likert) were tracked and analyzed. AI exposure reduced overall prescribing accuracy from 81.5% to 75.7% ( p < 0.001). In incorrect-AI scenarios, accuracy plummeted from 87.4% to 73.8% ( p < 0.001). Automation bias occurred in 19.4% of at-risk decisions (90/464), and significant pre- to post-AI accuracy declines were observed in six of the nine incorrect-AI scenarios in secondary scenario-level analyses ( p < 0.05). Automation bias was associated with a significant reduction in confidence (adjusted mean change = − 0.378, p < 0.001), whereas changing an initially wrong decision to a correct one following AI guidance was associated with increased confidence (adjusted mean change = + 0.328, p < 0.001). Final-year dental students were susceptible to commission-type automation bias in this simulated decision-support setting. Flawed AI guidance induces clinically meaningful commission errors accompanied by paradoxical confidence suppression, demonstrating an urgent need to integrate critical AI literacy into dental curricula.