Road traffic injuries are among the leading causes of child mortality worldwide, highlighting the importance of effective road safety education. This study aimed to examine the effectiveness of artificial intelligence (AI) compared to expert-led instruction in teaching children essential pedestrian safety behaviours. A total of 120 children aged 6–10 years (60 boys and 60 girls) from Serbia were randomly assigned to either an AI-based educational programme or a control group taught by a multidisciplinary panel of experts (pedagogues, psychologists, and traffic engineers). Both groups were trained on two critical pedestrian skills: crossing at a signalised pedestrian crossing and walking along a road without a sidewalk. Learning outcomes were evaluated through pre-test, immediate post-test, and a four-week follow-up, with statistical analyses including ANOVA and ANCOVA. Results showed significant improvements in both groups, but consistently higher performance for expert-led instruction, with correct crossing behaviour reaching 92% compared to 78% in the AI group, and correct road-walking behaviour reaching 95% compared to 80%. Effect sizes were large for both groups (Cohen’s d = 4.5 for expert-led and 3.1 for AI), while retention was superior in the expert-led group. These findings suggest that although AI-based learning offers scalability, adaptability, and interactive potential, expert-led instruction remains more effective in achieving mastery and long-term retention. The study highlights that AI may serve as a valuable supplementary tool within blended learning models, providing practical implications for policymakers and educators in designing sustainable and evidence-based approaches to children’s traffic safety education.