Uptake and day-level patterns of instructor-provided and other generative AI chatbot use in an undergraduate medical research methodology course: a cross-sectional survey with retrospective day-level assessment

Generative artificial intelligence chatbots are increasingly used by medical students. Educational programs are also beginning to provide course-specific chatbots, but little is known about how students incorporate these tools into their existing AI practices. This study quantified uptake of an instructor-provided chatbot and characterized retrospective day-level patterns of chatbot engagement during an undergraduate course. We conducted a survey among medical students enrolled in a 6-day research-methodology course. Students retrospectively reported their daily use and duration of use of the instructor-provided chatbot and other self-selected chatbots. Uptake of the instructor-provided chatbot was described using alternative frequency thresholds ranging from use on at least one to at least four course days. Daily chatbot-use states, intensity, and overlapping use were described. Satisfaction and open-ended feedback were assessed among students who had used the instructor-provided chatbot. We further explored factors associated with daily use of any AI chatbot using repeated measures analysis. Eighty-nine students were included. Thirty-one students (34.83%) used the instructor-provided chatbot on at least one course day, while 17 (19.10%) used it on at least two days, 11 (12.36%) on at least three days, and eight (8.99%) on at least four days. Daily use of any AI chatbot ranged from 61.80% to 87.64% across the course period and was driven primarily by self-selected chatbots. Student-level patterns showed that engagement with the instructor-provided chatbot was concentrated among a relatively small subgroup, whereas self-selected chatbot use was more widespread and sustained. In exploration analysis, daily AI chatbot use was lower during the weekend and was positively associated with more frequent prior AI use for non-medical purposes. Higher mean overall satisfaction was observed among repeated users than among one-time users, although the trend in overall satisfaction did not reach statistical significance ( p = 0.064). Open-ended feedback indicated a need for clearer guidance regarding the chatbot’s scope, information sources, prompting, and limitations. Uptake of the instructor-provided chatbot was selective and usually occurred alongside other chatbots rather than in isolation. Course-specific AI tools may therefore be more readily incorporated when their distinct role is clearly communicated and aligned with students’ existing AI practices. Assessing frequency, intensity, timing, and overlapping use provides a more informative account of self-reported AI engagement than ever-use measures alone.