The integration of artificial intelligence (AI) into medical education is rapidly transforming the way students acquire knowledge, develop clinical reasoning skills, and engage in lifelong learning (1). In higher education, AI-powered tools such as machine learning algorithms, large language models (LLMs), and automated tutoring programs, are increasingly being utilized to enhance traditional learning methods (2).

Recent research has shown that AI has the potential to revolutionize both medical education and clinical decision-making (5, 6). ChatGPT has been reported to perform at or above the median performance among 276,779 student test takers on the Medical College Admission Test (MCAT), and has performed at or near the passing threshold for all three steps of the United States Medical Licensing Examination (USMLE) (3, 4). In addition, AI is influencing aspects beyond education and clinical practice. AI is increasingly being used in residency applications, with 43% of applicants planning to use AI for personal statements, underscoring its growing role in professional development (7). Post-graduation, AI’s role in medical practice is expanding and currently includes dictation of notes, interpretation of laboratory results, and interpretation of medical imaging—technologies that are already improving efficiency and reducing physician workload (8). AI has also demonstrated the potential to improve healthcare quality by its ability to predict hospitalizations, emergency department visits, and mortality using electronic medical records, streamlining operational efficiency and improving risk assessment (9, 10).

The ethical discussions surrounding the rapidly expanding use of AI by students in medical school are still developing. Current discussions emphasize the importance of understanding its limitations, verifying sources, ensuring HIPAA compliance, and maintaining academic and professional integrity (11). Experts are issuing guidance on the development and use of trustworthy AI tools in healthcare, but similar guidance is not yet available for use by medical students during their training (12, 13).

Today’s medical students represent a critical group in this evolving landscape, as they will inevitably encounter AI-driven technologies in their clinical practice. Existing evidence suggests that medical students have a range of knowledge of AI tools and most agree that AI should be included in their medical training (14, 15). Inclusion of AI tools can support students in developing critical thinking, improving diagnostic accuracy, and increasing access to high-quality learning resources. A systematic review of studies on medical students’ perceptions of AI in 2023 found that medical students had a positive and optimistic attitude towards AI use in medicine; however, most students had low knowledge and limited skills in working with AI (15). By 2024, a study of 102 U.S. medical students reported high knowledge and high usage during studying (70%). In addition, they reported that those with some exposure to AI in medical school were more likely to trust AI with clinical decision-making in their future practice (16).

However, how students engage with LLMs and other AI powered tools during studying and schoolwork is limited. In the previously mentioned study of 102 U.S. medical students, 69% of respondents reported having used AI tools for medical-related purposes at least once a month (14). Another study of 415 students from 28 U.S. medical schools found that 52% reported using AI tools for medical schoolwork. The most common use was asking for explanations of medical concepts during pre-clerkship and assisting with diagnosis/treatment plans during clerkship (17). While these studies offered some information on AI use among medical students, they were limited by recall bias and did not ask detailed questions about the nature and frequency of use.

This study assessed the adoption and utilization of AI tools by medical students during studying and schoolwork through a longitudinal digital diary data collection to capture accurate accounts of the nature and frequency of AI use. By identifying the most used AI tools used during studying and their purposes, the study offers a more accurate understanding of AI use by medical students.

These findings may inform strategies on how AI can be effectively integrated into school policies, medical curricula, and ultimately into preparing future healthcare professionals to navigate AI in the healthcare environment with confidence and competence.

All currently enrolled DO students from both campuses were eligible to participate. Recruitment was conducted through email announcements distributed via institutional listservs. Interested students received an electronic informed consent form and were enrolled in the study upon providing consent. The study recruitment period spanned from April 10–28, 2025. Each participant was on their own 21-day digital diary schedule beginning upon completion of their intake survey. The final digital diary entry was collected on May 19, 2025.

Following informed consent, participants completed the baseline intake survey via Google Forms. The survey collected demographic information (e.g., age, gender, year in training, campus), prior exposure to AI tools, perceived impact of AI on learning and clinical reasoning, and self-reported frequency and use cases of AI tools in their medical education. The survey was developed for this study and is available as a supplementary file.

After completing the intake survey, participants received a brief digital survey prompt every three days over a 21-day period (seven total entries). Surveys were delivered via automated SMS notifications using a service provided by SimpleTexting.com. Each diary asked students to report the total minutes spent studying, the number of minutes using AI tools, which tools were used, and for what purposes (e.g., concept explanation, summarization, question generation). All responses were stored in a secure, password-protected cloud folder.

The study protocol was reviewed and approved by the Institutional Review Board at Touro University California. All participants provided informed consent, and all research team members completed Collaborative Institutional Training Initiative (CITI) training. No identifying information was collected, and data was stored securely in compliance with federal data protection standards. As an incentive for participation, students who complete the study were entered in a raffle to win one of twenty-eight $25 Visa gift cards and one of four $50 Target gift cards.

A total of 71 medical students completed the intake survey. The average age was 26.6 years (SD = 2.8). Among respondents, 39 (54.9%) identified as male and 32 (45.1%) identified as female. Students from all class years participated: 55 (77.4%) were in pre-clinical years MS1 or MS2 and 16 (22.6%) were in clinical years MS3, MS4 or research year. A total of 54 students (76.1%) were from Touro University California and 17 (23.9%) were from Touro University Nevada. Regarding study setting, 62 (87.3%) reported studying solo, 9 (12.6%) in a group (Table 1).

Regarding prior exposure to AI, 31 students (43.7%) reported moderate prior exposure to AI, 29 (40.8%) reported minimal exposure, 7 (9.9%) reported extensive exposure, and 4 (5.6%) reported no prior exposure. When asked whether AI tools should be formally incorporated into the curriculum, 43 students (60.5%) somewhat or strongly agreed, 28 (39.5%) somewhat or strongly disagreed.

At the time of the intake survey, 64 students (90.1%) reported having used AI tools for studying or schoolwork during their medical education. When asked if AI tools for studying and schoolwork are positively impacting their learning, 63 students (98.4%) somewhat or strongly agreed, 1 (1.6%) somewhat disagreed. When asked if AI tools for studying and schoolwork are positively impacting their clinical reasoning, 53 students (82.9%) somewhat or strongly agreed, 11 (17.2%) somewhat disagreed. (Table 1).

The average time spent studying was 185.6 mins per diary over the three-week period. Average time using AI tools was 35.8 minutes for an average of 19.4% of overall study time spent using AI.

Use of specific AI tools over the three-week study period was reported as follows: ChatGPT by 63 students (88.7%), Google Gemini by 22 (31.0%), AMBOSS AI by 11 (15.5%), Notebook LM by 6 (8.5%), Open Evidence by 5 (7.0%), UpToDate AI by 5 (7.0%), and DeepSeek by 3 (4.2%). Students reported using AI tools for a variety of purposes, including receiving simplified explanations of difficult concepts (74.6%, n = 53), learning more about complex medical concepts (71.8%, n = 51), answering practice questions (52.1%, n = 37), creating summaries of diseases and treatments (40.8%, n = 29), generating mnemonics or memory aids (Table 2).

Differences in use of tools by subgroups year in school and gender were assessed using Chi-square tests. No overall differences were observed based on gender, except that male students were more likely to use AI to create mnemonics (p = 0.034), while female students were more likely to use them to create study plans (p = 0.04). Clinical year students were more likely to use AI tools to answer practice questions (p = 0.002).

Participants’ average study time and AI use were tracked over seven diary days (Fig. 2). The average minutes spent studying ranged from 159.0 minutes on Diary 3 to 197.5 minutes on Diary 1, with a general consistency across days (Diary 1: 197.5, Diary 2: 190.5, Diary 4: 187.7, Diary 5: 192.9, Diary 6: 187.3, Diary 7: 184.1).

AI usage during study periods also varied. Average minutes spent using AI ranged from 29.1 minutes on Diary 4 to 40.8 minutes on Diary 6. The proportion of study time spent using AI ranged from 15.5% on Diary 4 to 23.6% on Diary 3. Across the seven diary entries, participants spent roughly between 15% and 24% of their study time engaging with AI.

There were no significant differences in the percentage of study time spent using AI based on gender, age, campus or attitudes towards AI use. The attitude composite was based on responses to three questions: (1 )Do you think that AI tools should be formally incorporated into curriculum by professors during medical education?, (2) Do you think your use of AI tools for studying and school work is positively impacting your learning?, (3) Do you think your use of AI tools for studying and school work is positively impacting your clinical reasoning?

However, a significant difference was observed by year in school. Third- and fourth-year students reported using AI during 30.5% of their study time, compared to the 15.8% amongst first and second year students (p = 0.003) (Table 3).

A linear regression model was used to adjust the relationship between percentage of AI use and year of training for demographic characteristics simultaneously. The relationship remained significant. Students clinical years (MS3 & MS4) reported using AI for studying approximately 19 percentage points more than those in in pre-clinical years (MS1 & MS2) (p < 0.0001) and male medical students reported using AI about 9 percentage points more than female students (p = 0.022). Age group was not strongly associated with AI use. Students at TUN used AI approximately 9 points less than those at TUC, with marginal significance (p = 0.054) (Table 4).

This study found that, on average, medical students used AI tools during 19.4% of their total study time. Students in clinical-phase peers (MS3 and MS4) were more likely to use AI for studying than their pre-clinical years (MS1 and MS2). Gender and campus also showed meaningful differences in the adjusted analysis. Students reported using primarily ChatGPT and Google Gemini to obtain simplified explanations of difficult concepts, learn more about complex medical concepts, answer practice questions, create summaries of diseases and treatments, generate mnemonics or memory aids and summarize lecture notes or textbooks.

In this study, clinical year students were more likely to use AI for answering practice questions. This subgroup is often preparing for board exams such as the Comprehensive Osteopathic Medical Achievement Test (COMAT) and United States Medical Licensing Examination Step 2 Clinical Knowledge (USMLE Step 2 CK) which coincided with the study period in April through May. These students may also be managing time constraints as they balance clinical rotations with self-directed studying. AI tools may serve as on-demand study aids by rapidly generating practice questions, explanations, and summaries to support exam review.

While AI tools have the potential to make studying more efficient and personalized, early evidence on their ability to deepen learning is mixed. Some studies suggest that AI use may hinder students’ ability to deeply understand concepts, synthesize large volumes of material, and effectively apply knowledge (18–20). There is also concern about the risk of misinterpreting AI-generated content or accepting inaccurate information, particularly when users do not critically evaluate the output (18). One study that found weaker exam performance associated with ChatGPT use advised that students should “be mindful of the potential drawbacks and consider integrating GenAI as a supplementary tool rather than a primary resource for grappling with complex topics”(21). Given that our findings show the most common uses of AI involve explaining concepts, answering practice questions, and summarizing diseases and treatments, the issue of accuracy warrants further investigation.

Taken together with emerging evidence, these findings suggest that institutions may benefit from better understanding whether and how the use of AI tools influences students’ knowledge and practice outcomes. Universities might also consider positioning AI as one part of a broader set of study resources available to students. Medical schools may find value in providing guidance on responsible AI use—emphasizing how to critically evaluate output, verify accuracy, and integrate AI effectively into evidence-based study strategies.

To date, no prior studies have reported AI adoption rates among medical students as a percentage of overall study time. A 2023 US-based cross-sectional survey involving 415 students across 28 U.S. medical schools found that 52% of participants reported having ever used ChatGPT for schoolwork, primarily for concept clarification, research assistance, and support during clinical rotations. Among students who had completed clinical rotations, 22.5% used ChatGPT during clerkships—mostly for diagnostic support and clinical queries. In terms of frequency, 39.3% of students reported using AI tools weekly, and 10.5% used them daily (17).

International studies have reported similarly wide variation in AI adoption rates among medical students: 42% in Sudan (22), 62% in China (23), and nearly 90% in Palestine (24). Usage has been found to correlate with gender (higher among males), socioeconomic status, and access to digital resources (22–24). While none of these studies employed a digital diary methodology, students consistently reported using AI to support academic tasks such as literature summarization, research workflows, clinical reasoning, explanation of medical concepts and assistance with diagnosis and treatment planning (22–24). The next generation of medical students is on track to use these tools even more. A global report from 3,839 college students found that 86% of students said they use AI for studying; 54% weekly and 24% daily (25).

In this study, over 60% of students believed that AI tools should be formally incorporated into medical school curriculum. AI-based education tools are already being tested across the spectrum, from pre-medical undergraduates to medical students and residents. For example, the extracurricular “AIM” program significantly improved AI literacy and understanding of pathology concepts among pre-med students, with large effect sizes observed in knowledge gains (26). A 14-week randomized controlled trial in China evaluated “LearnGuide,” a custom ChatGPT-based coaching tool for medical students, and found significant improvements in self-directed learning, critical thinking, and learning engagement compared to control groups (27). In another study, a tailored AI training tool improved pediatric fracture diagnostic accuracy by 6.2% among residents, though no significant improvements were observed among medical students on that task (28). Specialized tools that are carefully evaluated hold promise for enhancing medical student learning.

This study has several limitations that should be considered when interpreting the findings. First, while the digital diary method was intended to reduce recall bias and capture real-time usage patterns, self-reported data is still subject to social desirability and reporting biases. Participants may have overestimated or underestimated their AI use or selected purposes that appeared academically favorable. Second, the non-random sample was limited to students from two osteopathic medical schools, which may limit the generalizability of findings to other institutions, including allopathic programs or international medical schools. Additionally, the total sample size of 71, while adequate for exploratory analysis, may have limited statistical power to detect subtle subgroup differences or interaction effects. The wide confidence intervals observed in our regression model reflect this limitation and suggest reduced precision in our estimates. Third, although the study spanned over a three-week period, AI use may vary across different points in the academic calendar (for example, exam weeks versus routine schoolwork), and the time frame captured may not fully represent usage across a typical semester. Lastly, while the analysis included adjustments for several demographic factors, including prior AI exposure and formal instruction, there may still be unmeasured confounders, such as digital literacy, depth of AI tool proficiency, or baseline academic performance, that influenced how and how often students used AI tools.

Together with the present findings, this growing body of evidence suggests that medical schools should acknowledge the increasing role of AI in students' learning processes. We recommend further research to help education institutions better understand the use of AI for academic success which can then inform structured guidance on its safe, effective, and critical use for medical students. In addition, the novel digital diary methodology employed in this study offered real-time insights into how, and how often medical students use AI tools during their study routines. In contrast to previous research relying on retrospective surveys and recall (14, 21, 22), this approach was easy to deploy and may serve as a useful model for evaluating future AI-related educational interventions.