This study is a cross-sectional, descriptive survey conducted in April and May 2025 among medical students and recent graduates in Syria. The primary aim was to explore their experiences with AI tools and perceptions of AI’s role in medical education and clinical practice.

A total of 400 participants were enrolled, including clinical-year medical students (fourth and fifth years) and preresidency students (graduates and near-graduates). Participants were recruited through convenience sampling using academic student networks and widely used social media platforms, such as closed university-affiliated groups.

Exclusion criteria were preclinical students (first to third years) and postgraduate specialty trainees (residents). Inclusion criteria included current enrollment in clinical medical training or graduation within the past 2 years, and those who voluntarily agreed to complete the questionnaire. Participation was anonymous and uncompensated.

Participants were drawn from multiple public medical faculties across various regions of Syria. The eligible population was estimated to include several thousand clinical-year students and recent graduates nationwide. Due to security and privacy considerations, specific university names are withheld.

Data were collected using a structured, self-administered online questionnaire delivered via a shared Google Forms (Google LLC) link, in accordance with CHERRIES (Checklist for Reporting Results of Internet E-Surveys) guidelines. The questionnaire was originally developed in English based on published literature addressing medical students’ experiences and perceptions of AI in medical education and clinical practice. It was professionally translated into Arabic by experts in medical education and linguistics. The Arabic version was reviewed and refined by bilingual subject-matter experts to ensure linguistic accuracy, cultural appropriateness, and conceptual equivalence.

The perception items were intended to capture attitudinal views rather than to constitute a formal psychometric scale. Accordingly, internal consistency of the multi-item perception items was assessed using Cronbach α as an indicator of reliability. The final questionnaire was pilot-tested with a small group (≈30 students) to assess clarity and contextual relevance. Internal consistency of the multi-item perception scales was evaluated using Cronbach α, yielding acceptable reliability (α=.77 in pilot, α=.75 in full sample). As no modifications were made to the questionnaire following this assessment, these responses were retained and included in the final analytic dataset.

The survey consisted of 52 questions over 4 pages, organized into the following domains aligned with the Results section:

Access to the survey was restricted to researchers with password-protected accounts. The open survey link was disseminated via online academic student groups. Participation was voluntary, anonymous, and uncompensated, with no incentives offered. All questions were mandatory except for 2 optional items. A back button was available, but no summary review or adaptive questioning was used.

To protect participant confidentiality and minimize reidentification risk, no personally identifiable information (names, contact details, or institutional identifiers) or IP addresses were collected. Each Google Forms submission was treated as a unique response; duplicate or partial submissions were excluded after data cleaning.

Quantitative data were analyzed using IBM SPSS Statistics (version 27; IBM Corp). Descriptive statistics (frequencies, percentages, means, or SDs) summarized participant characteristics and AI use. Inferential analyses included:

Given the exploratory nature of this study and the large number of comparisons performed, no formal adjustment for multiple comparisons was applied. Instead, exact P values were reported throughout to allow transparent interpretation of statistical significance.

Missing data were negligible, and all analyses were conducted using complete-case analysis without imputation.

Ethical approval was obtained from the Ethical Approval for Biomedical Researchers committee (1612). All participants provided informed digital consent before survey commencement, after reading an introductory consent statement outlining this study’s purpose and duration. Responses were collected anonymously and securely stored to protect privacy. Participants received no compensation for participation.

Data collection took place between April and May 2025. By the end of the recruitment period in May 2025, a total of 400 clinical-year students and preresidency students had participated in this study. Data cleaning and verification were conducted, followed by statistical analyses completed in June 2025. The analysis focused on patterns of AI use, perceptions toward AI in medical education and patient care, and their associations with demographic characteristics. In the following analyses, the composite prior AI-use variable represents the number of different educational and research tasks for which participants reported using AI; higher scores indicate AI use across a greater number of distinct tasks.

The sample was nearly gender-balanced, with 51.5% (206/400) of participants identifying as male and 48.5% (194/400) of participants as female. Most participants (336/400, 84%) were aged between 20 and 24 years. Half of the respondents were clinical-year students (200/400, 50%), and the other half were preresidency students. Notably, a large proportion (290/400, 72.5% of participants) expressed a desire to pursue postgraduate specialization abroad. In terms of technology and language exposure, 68.8% (275/400) of participants reported having access to a PC. German language study was reported by 39.8% (159/400) of participants, of whom 50.9% (81/159) of participants used AI tools to support their learning. Research experience was reported by 28.2% (113/400) of participants. Knowledge about AI was most commonly obtained through social media (294/400, 73.5% of participants), followed by reading research papers (60/400, 15% of participants) and discussions with IT-oriented peers (39/400, 9.8% participants; Table 1).

Participants reported using AI for a variety of academic and clinical tasks. Use for studying or examination preparation was reported by 57% (228/400) of participants, while 40% (160/400) of participants used it to complete written assignments. AI was used to suggest research topics or questions by 35.5% (142/400) of participants and to assist in writing research papers by 33.8% (135/400) of participants. Additional uses included writing case reports (93/400, 23.3% participants), generating self-assessment questions (87/400, 21.8% participants), and drafting patient notes (47/400, 11.8% participants).

Use of AI-supported educational technologies during medical school was also common. Digital anatomy tools were used by 72.3% (289/400) of participants, computational pathology tools by 42.8% (171/400) of participants, and AI-generated clinical simulation cases by 22.5% (90/400) of participants (Table 2).

When asked about future use of AI during residency training, participants expressed high levels of interest. AI was considered useful for answering medical questions by 85.5% (342/400) of participants, exploring new medical topics or research by 82.8% (331/400) of participants, and supporting studying or examination preparation by 80.8% (323/400) of participants.

Further intended uses included writing research papers (300/400, 75% participants), writing case reports (261/400, 65.3% participants), writing patient notes (211/400, 52.8% participants), and assisting in clinical decision-making (208/400, 52% participants; Table 3).

A series of chi-square tests examined associations between demographic characteristics and patterns of AI use (academic level, gender, research experience, computer availability, and German language exposure). Significant associations were identified between subgroups and multiple AI use cases (Table 4).

Clinical-year students reported higher use of AI for studying or examination preparation compared with preresidency students (66% vs 48%; χ²₁=13.21; P<.001). Similar differences were observed for generating self-assessment questions (26.5% vs 17%, χ²₁=5.30; P=.02), suggesting research topics (45% vs 26%, χ²₁=15.76; P<.001), writing research papers (39.5% vs 28%, χ²₁=5.91; P=.02), completing written assignments (48.5% vs 31.5%, χ²₁=12.04; P<.001), creating simulated clinical cases (27% vs 18%, χ²₁=4.64; P=.03), and writing patient notes (15.5% vs 8%, χ²₁=5.42; P=.02).

Gender differences were observed across several AI-use domains. Male participants reported greater use of AI for writing research papers (40.3% vs 26.8%; χ²₁=8.12; P=.004), suggesting research topics (40.3% vs 30.4%; χ²₁=4.25; P=.04), and using digital anatomy tools (77.2% vs 67%; χ²₁=5.15; P=.02).

Participants with research experience were significantly more likely to use AI for suggesting research topics (43.4% vs 32.4%; χ²₁=4.25; P=.04), writing research papers (56.6% vs 24.7%; χ²₁=36.89; P<.001), writing case reports (33.6% vs 19.2%; χ²₁=9.50; P=.002), and using digital anatomy platforms (84.1% vs 67.6%; χ²₁=10.97; P<.001).

Access to a PC was associated with greater use of AI for writing research papers (38.2% vs 24%; χ²₁=7.73; P=.005) and for learning via digital anatomy platforms (77.1% vs 61.6%; χ²₁=10.28; P<.001).

Students studying German reported slightly higher use of AI for language learning and for completing written assignments (47.2% vs 35.3%; χ²₁=5.65; P=.02) and to help write research papers (42.1% vs 28.2%; χ²₁=8.30; P=.004). This trend reflects a culturally motivated behavior, particularly among Syrian students, who often aspire to pursue postgraduate training in Germany due to the ongoing crisis. These students appeared to use AI to simulate German language acquisition, highlighting AI’s potential role in educational mobility and international preparation.

Perceptions of AI for career and educational domains were compared across student subgroups (academic level, gender, research experience, access to a PC, and German language exposure) using Mann-Whitney U tests.

Preresidency students reported higher agreement that ChatGPT improves learning during residency compared with clinical-year students (median 4, IQR 4‐4, vs 4, IQR 3‐4; U=17,736; P=.03). They also showed stronger agreement that medical schools and residency programs should establish formal policies regulating AI use by trainees (median 4, IQR 4‐4, vs 4, IQR 3‐4; U=17,497; P=.02). In contrast, clinical-year students rated AI as more effective in meeting their needs during medical school than preresidency students (median 4, IQR 3‐4, vs 3, IQR 3‐4; U=16,501; P<.001).

Gender-based differences were observed across several items. Male students reported higher ratings for AI supporting learning during residency (median 4, IQR 4‐4, vs 4, IQR 3‐4; U=17,254.5; P=.007), preference for ChatGPT over other search engines (median 4, IQR 3‐4, vs 3, IQR 2‐4; U=16,667; P=.003), and enthusiasm toward future versions of AI (median 4, IQR 4‐4, vs 4, IQR 3‐4; U=17,264; P=.009).

Students without PCs reported greater influence of AI on residency specialty choice compared with those with computers (median 3, IQR 2‐3, vs 3, IQR 2‐3; U=15,059; P=.04). Conversely, students with PCs were less likely to agree that their peers consistently used AI ethically (median 3, IQR 3‐3, vs 3, IQR 3‐4; U=14,595; P=.007).

Students who had learned German rated AI as more helpful for learning during residency than those who had not (median 4, IQR 4‐4, vs 4, IQR 3‐4; U=16,350; P=.005), which may reflect AI’s perceived value in language acquisition for migration-preparing students. Additionally, they showed enthusiasm toward future versions of AI (median 4, IQR 4‐5, vs 4, IQR 3‐4; U=16,136; P=.003).

Research experience was also associated with perception differences. Students with research experience were less confident in the accuracy of AI’s outputs (median 4, IQR 4‐5, vs 4, IQR 3‐4; U=13,042; P<.001), more supportive of institutional AI policies (median 4, IQR 4‐4.5, vs 4, IQR 3‐4; U=13,535; P=.005), and more likely to agree that AI improves learning during residency (median 4, IQR 4‐4, vs 4, IQR 3‐4; U=14,272; P=.03). In contrast, students without research experience were more likely to perceive their peers as using AI ethically (median 3, IQR 3‐4, vs 3, IQR 2‐4; U=13,902; P=.01) and to report greater influence of AI on residency specialty choice (median 2, IQR 2‐3, vs 2, IQR 1‐3; U=14,175; P=.04). Detailed Mann-Whitney U test results for perceptions of AI in education and career are provided in Multimedia Appendix 1.

A series of Mann-Whitney U tests were conducted to compare perceptions of AI for patient care and professional practice among medical students’ subgroups (academic level, gender, research experience, access to a PC, and German language exposure).

Gender again played an important role where male students reported significantly higher agreement that AI will improve diagnostic accuracy compared with female students (median 4, IQR 3‐4, vs 3, IQR 3‐4; U=15,489; P<.001), improve patient care (median 4, IQR 3‐4, vs 4, IQR 3‐4; U=15,281; P<.001), AI will have a major impact on health care (median 4, IQR 3‐4, vs 4, IQR 3‐4; U=16,171; P<.001), reduce medical errors (median 3, IQR 3‐4, vs 3, IQR 3‐4; U=16,931; P=.005), whereas female students worry more about the ethical impact of AI on health care (median 4, IQR 3‐4, vs 3, IQR 3‐4; U=17,410; P=.02).

Academic level was also associated with differences in perception. Preresidency students believed more that AI will have a major impact on health care during residency (median 4, IQR 3‐4, vs 4, IQR 3‐4; U=17,760; P=.03).

German language learners expressed higher agreement with the fact that AI will improve patient care (median 4, IQR 3‐4, vs 4, IQR 3‐4; U=16,308; P=.007), have a major impact on health care (median 4, IQR 3‐4, vs 4, IQR 3‐4; U=16,680; P=.02) and will enable them to make more accurate diagnoses (median 4, IQR 3‐4, vs 3, IQR 3‐4; U=15,716; P<.001).

Research experience was associated with increased concern that AI will reduce patient trust in physicians (median 4, IQR 3‐4, vs 4, IQR 3‐4; U=14,072; P=.03). Detailed Mann-Whitney U test results for perceptions of AI in patient care are available in Multimedia Appendix 2.

Multivariable binary logistic regression analyses were conducted to identify predictors of positive perceptions of AI across educational and clinical domains. Prior AI usage experience significantly increased the odds of a favorable perception in multiple domains: each additional AI‐use task raised the likelihood that students would agree that AI improves learning during residency (aOR=1.43, 95% CI 1.15‐1.79, P=.002), that AI was effective in meeting medical‐school needs (aOR=1.60, 95% CI 1.34‐1.92, P<.001), and that it would help reduce medical errors (aOR=1.14, 95% CI 1.01‐1.28, P=.04).

Clinical‐year status was consistently associated with more reserved views: clinical-year students were associated with lower odds of perceiving AI as beneficial for residency learning (aOR=0.26, 95% CI 0.11‐0.64, P=.003), less likely to anticipate using future versions (aOR=0.18, 95% CI 0.07‐0.49, P<.001), and less likely to expect AI to have a major impact on health care (aOR=0.24, 95% CI 0.09‐0.65, P=.005).

Gender differences appeared: female students had 3 times the odds of insisting on verifying AI’s answers (aOR=3.12, 95% CI 1.36‐7.15, P=.007) and female students were twice as likely to worry about AI’s ethical impact on health care (aOR=2.06, 95% CI 1.19‐3.54, P=.01).

Research experience did not significantly predict any positive perception after adjustment.

This multicenter cross-sectional survey of 400 medical students and recent graduates in the Syrian Arab Republic addresses a notable gap in the literature on AI in medical education in a low-resource, conflict-affected setting. We found that students increasingly rely on AI tools for educational and clinical purposes, with the highest reported uses being examination preparation (228/400, 57%), written assignments (160/400, 40%), and digital anatomy platforms (289/400, 72.3%). Usage patterns differed significantly by academic level (higher use among clinical-year students), gender, computer access, and research experience. Although participants were generally optimistic about AI tools’ potential to enhance learning and residency training, many expressed ethical concerns and worries about reduced humanistic aspects of care; only 25.5% (102/400) of students expected direct career benefits. These findings both demonstrate the growing role of AI tools among Syrian medical trainees and underscore the need for institutional policies, targeted training, and efforts to address digital access gaps to ensure ethical and equitable integration. Thus, the summary of key findings is the high uptake of AI for study and anatomy tools; generally positive but ethically cautious perceptions; and notable subgroup differences by academic level, gender, access, and research experience.

The cross-sectional design of this study limits causal inference; therefore, interpretations about drivers of use are based on observed associations rather than proven causal relationships.

Analysis of the data reveals that the usage rate of AI tools was notably higher among males than among females across almost all fields, a pattern that aligns with findings from studies conducted in Sudan and the United Arab Emirates [35,36]. Moreover, this higher usage phenomenon may reflect differences in attitudes, access, or self-reported behaviors rather than causal effects; we therefore describe these as associations rather than causes [37]. In addition, clinical students consistently demonstrated a greater likelihood of using AI tools across many tasks compared to both graduate and prospective students, reinforcing the notion that different academic cohorts adopt AI at varying intensities. Higher AI use among clinical students may reflect greater curricular demands or exposure; qualitative work is needed to confirm underlying reasons. However, our findings stand in contrast to a study conducted in China, which reported that graduate medical students exhibited greater awareness of medical AI usage than undergraduate students [38]. To examine socioeconomic influences, we used ownership of a PC as a proxy; access to a PC was positively associated with AI usage—an observation consistent with findings from Saudi Arabia [39].

First, computer ownership likely reflects unequal access to digital infrastructure within the country; in the context of economic hardship and power instability, some students may lack reliable devices or connectivity, widen digital divides, and limit opportunities to use AI tools. Second, the association between German language study and higher AI use may reflect preparatory behavior among students aiming for postgraduate training abroad, who invest in language and digital tools to improve competitiveness. Third, intentions to specialize abroad raise concerns about potential brain drain: if those with greater access and digital literacy are also more likely to emigrate, local inequities in human resources for health could be exacerbated.

Although learning the German language was not statistically associated with most AI applications, students studying German reported slightly higher use of AI tools for language learning, particularly for completing written assignments (P=.02), and they used AI tools more frequently to assist with writing research papers than students not studying German. Previous studies have demonstrated that AI significantly impacts students’ language learning [40]. This trend may reflect culturally motivated behavior among students preparing for postgraduate training abroad and highlights AI tools’ potential role in facilitating language acquisition and international mobility.

While this study offers insights into AI tools’ perceptions and use among Syrian medical trainees, limitations include convenience sampling, reliance on self-reported data subject to recall and social desirability bias, use of proxy variables (eg, computer ownership for socioeconomic status), and lack of prestudy sample size calculation; these factors limit generalizability and preclude causal inference. We explicitly note that PC ownership was used as a proxy for income; however, this is an imperfect and 1D indicator that does not capture the multidimensional nature of socioeconomic status. The cross-sectional design and self-reported measures limit causal inference; hypotheses about drivers of AI tool use require longitudinal or qualitative validation. Further, this study involved a large number of statistical comparisons across multiple domains. Although no formal adjustment for multiple comparisons was applied, exact P values were reported throughout to support transparent interpretation of the findings, and results with marginal statistical significance should be interpreted with caution. Additionally, items referring to “during residency” capture intended future use for most respondents and do not measure observed practice among current residents.

Future research should consider longitudinal or mixed method designs to better understand evolving perceptions of AI tools throughout medical training and early clinical practice. Qualitative methodologies, including interviews and focus groups, could deepen insights into the reasoning behind student attitudes—especially regarding trust, ethics, and specialty choices. Expanding the sample to include students from private institutions, rural areas, or international campuses would improve the representativeness of findings. Policymakers and educators should consider equity-focused interventions that address device and connectivity gaps and strategies to retain trained graduates to mitigate potential brain drain.

This study reveals a predominantly optimistic outlook among Syrian medical students and residents regarding the role of AI in education and clinical care. High levels of prior usage suggest both accessibility and growing technological literacy, though notable concerns—particularly ethical and relational—underscore the need for guided integration. Differences across gender, training stage, and socioeconomic proxies reflect the nuanced landscape of AI perception in medical education. While enthusiasm is widespread, cautious appraisal of AI’s limitations and its potential to disrupt traditional human-centered care remains essential. As such, integrating AI into medical education must be performed strategically, emphasizing critical appraisal, ethical awareness, and reinforcement of compassionate clinical reasoning. These findings contribute meaningfully to the regional and global discourse on responsible AI adoption in health professions education.