Artificial intelligence-based chatbots improve the efficiency of course orientation among medical students: a cross-sectional study.

BACKGROUND

Large language models (LLMs) like ChatGPT offer new ways to improve academic and administrative workflows in medical education, particularly for students studying in a language that is not their native tongue. We set out to examine whether a custom-trained chatbot could better support course orientation compared to traditional text-based resources for an ESL student population.

METHODS

Seventy-five first-year international medical students at the University of Szeged, Hungary participated during the first session of an introductory medical informatics course to use either a custom-trained ChatGPT-based chatbot (AI group, n = 35) or the university's standard e-learning platform (Traditional group, n = 40) to locate key course information. A six-item test based on the syllabus assessed information retrieval within a 10-min window. Test scores, time per question, and question revisits were compared using independent t-tests and two-way ANOVA.

RESULTS

No differences were found in previous use of AI-based tools between the two groups (Traditional: 90.0% vs AI group: 91.4%, P = 1). The AI group scored significantly higher (5.0 [4.8 - 6.0], median[IQR]) than the Traditional group (3.5 [2.8 - 4.3], P < 0.001). The majority of AI users (24/35) achieved 5 or more correct answers, compared to just 6/40 in the Traditional group. The lowest scores (0 points) were found exclusively in the Traditional group. No significant difference was observed in total test completion time (Traditional: 380.3 ± 158.6; AI: 433.7 ± 121.4; P = 0.10), time spent per question (P = 0.90), or question revisits (P = 0.99). Notably, the AI group outperformed on specific questions related to course logistics, such as attendance (62.9% vs 17.5% correct) and grading criteria (82.9% vs 55.0% correct) (both P < 0.001).

CONCLUSIONS

Artificial intelligence-assisted information retrieval significantly improved students' ability to locate and understand key course content. Despite comparable time-on-task and navigation behaviours between groups, students using the ChatGPT-based chatbot achieved markedly higher test scores, likely due to the simplified and focused responses it provided. The absence of zero scores in the AI group and enhanced performance on logistical questions suggest that conversational interfaces may improve comprehension and retention of procedural information. These results highlight the potential of LLM-based tools to support more efficient onboarding in medical education, with a potential for broader applications across educational settings.