Background: Cervical cancer remains a major global health issue, and colposcopy training struggles with standardization and scalability. AI has potential to improve medical education, but its effectiveness in colposcopy training remains underexplored. Objective: To evaluate the efficacy of an artificial intelligence (AI)-driven virtual patient platform for colposcopy training among resident physicians. Methods: In this randomized, assessor-blinded, wait-list-controlled trial, 60 obstetrics and gynecology residents (PGY-2/3) from two affiliated teaching hospitals were enrolled (May 2023-September 2025). Participants were randomly assigned to either the Immediate AI Intervention group (traditional teaching plus an AI platform throughout) or the Wait-List Control group (traditional teaching only during Month 1, followed by the AI platform). The primary outcome was the between-group difference in the Case Analysis score at 1 month. Secondary outcomes included theoretical knowledge, communication performance, diagnostic justification, and operative competence assessed via OSCE. Results: At 1 month, the Immediate AI group showed superior performance in Case Analysis (78.57 ± 3.72 vs 71.03 ± 5.82, mean difference = 7.54, p < 0.001, Cohen's d=1.54), Diagnostic Justification (81.63 ± 6.21 vs 74.13 ± 6.77, p < 0.001, Cohen's d=1.15), Communication Performance (80.97 ± 3.90 vs 75.77 ± 6.08, p < 0.001, Cohen's d=1.02), and Theoretical Knowledge (84.67 ± 6.27 vs 80.43 ± 5.07, p=0.006, Cohen's d=0.74). Operative competence showed no significant difference (p = 0.113). Both groups achieved high proficiency (over 85 points) within 3 months. Satisfaction was high (90.0% participants, 91.7% supervisors). Conclusion: An AI-driven virtual patient platform significantly accelerates the acquisition of diagnostic reasoning and communication skills in colposcopy training, demonstrating high educational efficacy and acceptance for scalable, standardized education.