Turning AI hallucinations into scaffolds for morphologic reasoning: A pilot randomized trial of prompt‐guided virtual microscopy learning during a focused histology module

Abstract

AI is a medical education tool, yet its potential to foster morphologic reasoning remains underexplored. Histology students often struggle to move beyond pattern recognition toward evidence‐based justification. This pilot trial evaluated a prompt‐guided “Prompt Loop” workflow for improving morphologic reasoning and verification‐oriented appraisal. A parallel‐design randomized controlled trial was conducted with 55 first‐year medical students. Participants were assigned to Standard Virtual Microscopy (VM) learning ( n  = 28) or VM Prompt Loop ( n  = 27). The intervention used a multimodal AI assistant (GPT‐4o) within a four‐stage cycle—Observation, Reasoning, Verification, and Consolidation—designed to treat AI inaccuracies as verification scaffolds. Outcomes included a supervised VM morphologic reasoning assessment (Identification vs. Reasoning) and a domain‐level perception questionnaire. The VM Prompt Loop group achieved higher total scores (78.0 ± 13.5 vs. 67.1 ± 12.0; p  = 0.003; Cohen's d = 0.85). This gain was mainly driven by the morphologic reasoning (38.4 ± 10.2 vs. 31.0 ± 8.2; p  = 0.005; d = 0.80), whereas identification scores showed no significant difference ( p  = 0.073). Survey data revealed higher ratings for verification and inquiry behaviors ( p  < 0.001). Process documentation showed high fidelity, with 93% of Prompt Loop students submitting required AI‐interaction screenshots and a mean of 2.1 ± 1.5 error‐correction entries. A prompt‐guided AI workflow improved morphologic reasoning in histology. When AI outputs were treated as provisional claims and routed through a structured verification process, the Prompt Loop may provide a practical model for supporting verification‐oriented morphologic reasoning in histology.