Bridging technology and pedagogy: a cross-sectional analysis of AI use, self-reported academic effects, and ethical concerns among students across diverse health science curricula

Synthesis: A cross-sectional survey of 310 health science students at a Saudi university documents the pattern that now defines generative AI in professional education: near-universal LLM use without formal preparation. Ninety percent (n = 279) had used ChatGPT, yet 82.6% had received no formal instruction in AI and mean self-rated knowledge was 3.08 ± 1.06, a gap between use and instruction that is the paper's most policy-relevant result. Usage frequency was associated with higher GPA (ANOVA F = 15.56), but 61% reported ethical concerns, with rates varying by gender (p = 0.0015) and program (p = 0.0114). The authors read this as a case for professional training that pairs technical instruction with AI literacy and ethical literacy — including automation bias — and for academic integrity rules students can actually apply. For nursing and medical programs, adoption has outrun the curriculum, and self-reported study gains are not evidence of learning.

Key Findings

1. The study surveyed 310 students across health-related programs using a stratified two-stage cluster sample and interviewer-administered questionnaires at the University of Bisha in Saudi Arabia.
2. Participants reported only moderate familiarity with AI: the mean knowledge score was 3.08 ± 1.06, despite widespread hands-on exposure to the tools.
3. Ninety percent (n = 279) of students had used ChatGPT, while 82.6% reported receiving no formal instruction on AI of any kind.
4. AI usage frequency was significantly associated with higher GPA (ANOVA, F = 15.56), presented by the authors as association rather than evidence of causation.
5. Ethical concerns were prevalent, reported by 61% of respondents, and varied significantly by gender (p = 0.0015) and by academic program (p = 0.0114).
6. Students self-reported better study efficiency and comprehension, and the authors recommend "proactive" training that addresses automation bias and ethical literacy for every student.

Summary

The paper starts from a plain mismatch: generative AI has become ordinary in health professions education, while evidence on its academic effects and on students' ethical worries remains thin. The authors describe three things at once — how students use AI, what they believe it does for their academic work, and which ethical concerns they hold — across programs that differ in content and clinical demand. The setting is a single Saudi university, and the design is a cross-sectional survey of 310 students recruited through a stratified two-stage cluster sample and questioned in person. The headline is adoption without instruction: 90% of students used ChatGPT, 82.6% had never been taught about AI, and mean self-rated knowledge was 3.08 on the scale used. The authors conclude that unequal formal training and prevalent ethical concern together justify a structured AI curriculum, with ethical risk named explicitly rather than left to be absorbed informally.

Study Design and Method

The study used stratified two-stage cluster sampling to reach 310 students enrolled in various health-related programs, collecting data through interviewer-administered questionnaires rather than an online form. That choice matters for interpretation: an interviewer can clarify items, but students may also under-report behavior their program discourages. The instrument captured familiarity with AI on a scored scale, the frequency and tools of use, self-reported academic effects, and a set of ethical concerns. Analysis compared usage frequency against GPA and tested whether ethical concern varied by gender and program. The paper's own conclusion states the central constraint plainly: the cross-sectional design does not establish causality between AI use and academic achievement. The sampling frame is confined to one institution, so the estimates describe Bisha's health science students rather than Saudi health professions education as a whole.

Adoption Ahead of Instruction

The two prevalence figures do most of the analytical work. If 90% of students have used ChatGPT while 82.6% have had no formal instruction, then whatever competence exists was acquired informally — from peers, from trial and error, or from the tool's interface. That is a familiar pattern in professional training, where placements and study demands create immediate incentives to use whatever is available. A mean knowledge score of 3.08 ± 1.06 suggests students know they are working with something they do not fully understand: a reasonable starting point for teaching, a poor basis for self-directed use in high-stakes coursework. The authors' emphasis on automation bias is well placed, because a student who trusts a fluent answer has no independent check on it, and in health curricula the cost of an unchecked error is not only a grade. Adoption also distributes unevenly across programs.

Academic Effects and Ethical Concerns

The GPA result is the finding most likely to be misread. Usage frequency was significantly associated with higher GPA (F = 15.56), and students reported better study efficiency and comprehension, but a cross-sectional comparison cannot say whether AI use raised grades, whether stronger students use AI more, or whether both reflect a third factor such as available study time. The authors are careful here. The ethical picture is more differentiated than one prevalence figure suggests: 61% reported concerns, and the rate varied by gender (p = 0.0015) and program (p = 0.0114), which implies concern is learned somewhere — a curriculum, a clinical culture, a peer group — rather than evenly distributed. For academic integrity, students without instruction have no shared rule to follow, so the same behavior can look like efficient studying to one student and misconduct to a lecturer.

What this means for practice

• Instructors. Teach AI use inside existing courses rather than assuming students acquired good habits elsewhere, because 82.6% reported no formal instruction and knowledge scores stayed moderate.
• Faculty developers. Build sessions around automation bias and verification — how to check a fluent but wrong answer — the risk the authors single out in their practice recommendation.
• Faculty developers. Make the integrity rule concrete for each program, since ethical concern varied by program (p = 0.0114) and untaught students have no shared standard.
• Administrators. Fund a structured AI curriculum across health programs rather than optional workshops, given that 90% of students already used ChatGPT.

Limitations

• The design is cross-sectional, so the GPA association (F = 15.56) cannot establish that AI use improves achievement; the authors state this limitation directly.
• The sample of 310 students comes from one university in Bisha, limiting generalization to other Saudi or regional health science programs.
• AI knowledge was a self-rated score (mean 3.08 ± 1.06), not tested competence, so it reflects familiarity rather than verified skill.
• With 82.6% lacking formal instruction, reported use describes self-taught practice, and ethical concern is reported as aggregate prevalence rather than by type.

Citation

Alshahrani, A. M., Ibrahim, I. A. E., Jibo, A. M., et al. (2026). Bridging technology and pedagogy: a cross-sectional analysis of AI use, self-reported academic effects, and ethical concerns among students across diverse health science curricula. BMC Medical Education.