What do Saudi health students report about AI—and what does the survey not prove?
In a survey of 310 University of Bisha students, 279 said they used ChatGPT while 82.6% reported no formal AI instruction. Frequent use was associated with higher self-reported GPA, but the cross-sectional design cannot show that AI improved learning or grades.
The short version
- What happened
- The cross-sectional study surveyed 310 health-science students through interviewer-administered questionnaires using a stratified two-stage cluster design; it measured reports and associations, not learning gains caused by AI.
- Who it affects
- Students are already using AI in health-related study while most surveyed students report no formal preparation; unclear rules can leave responsible use dependent on individual confidence and access.
- What remains unproven
- Cross-sectional survey at one Saudi university; it cannot establish temporal order or causation.
Keep in mind
Cross-sectional survey at one Saudi university; it cannot establish temporal order or causation.
The denominator was 310 health-science students, not a national or international sample.
AI use, perceived effects and GPA were self-reported rather than verified through system logs, institutional records or common assessments.
The survey did not measure durable learning, clinical performance, patient outcomes or the accuracy of AI-generated material.
Statistical differences by gender and programme do not explain their cause or practical size.
The authors reported no financial support and no competing interests; they disclosed using AI-assisted tools for language editing and formatting, with human review.
Source analysis · 1 linked source. Primary analysed source date: 3 November 2026.
Keep following the evidence
Join The Impact Brief founding list for selected AI reports, their effect on people and the evidence limits. Read a sample first.
Read the full report · 9 min The evidence, methods, implications and limitations
The clearest finding is a training gap, not an AI learning effect
A peer-reviewed survey at the University of Bisha in Saudi Arabia found that student use of generative AI was already widespread while formal preparation lagged behind. Of 310 health-science students, 279—90%—said they used ChatGPT. At the same time, 82.6% reported receiving no formal instruction in AI. The mean reported familiarity score was 3.08. For educators, the mismatch between widespread use and limited formal instruction is more actionable than any claim that the tools improved achievement.
The paper also reports that more frequent AI use was associated with higher self-reported grade-point average, with an analysis-of-variance statistic of F=15.56. That is not evidence that ChatGPT raised grades. The survey captured exposure and outcomes at one point in time. Students who already perform well may be more likely to use new tools, programmes may differ in both grading and adoption, and motivation or digital access may affect both. Because GPA and academic effects were self-reported, reporting and recall can add further error.[1]
How the study was designed
The researchers used a stratified two-stage cluster sampling design to recruit 310 students across several health-related curricula. Data came from interviewer-administered questionnaires rather than learning-platform records, verified assessment results or direct observation of AI use. The study asked about familiarity, tools, frequency, perceived effects, academic performance and ethical concerns. That approach can capture attitudes and practices across programmes efficiently, but it measures what participants say they do and experience.
The sample and setting define the denominator. These were students connected to one Saudi university, not a national sample of medical, nursing or allied-health education. Cluster sampling can improve coverage inside the institution, yet students within the same programme or cohort may share teaching and assessment conditions. The published result therefore describes this university at this point in its adoption, with uncertain transfer to other Saudi institutions, other countries or programmes that have different access policies and assessment designs.[1]
The distinction that matters
What the survey measured—and what it cannot establish
The same numbers support a strong case for training and only a weak case about academic benefit.
Observed reports
Survey 310 students at one university
279 used ChatGPT; 82.6% reported no formal AI instruction
Association
Compare use frequency with self-reported GPA
Groups differed, but direction and cause remain unknown
Concern
Ask about ethical issues
61% reported concerns, with variation by gender and programme
Needed next
Test defined training across multiple universities
Measure verified learning, safety and independent reasoning over time
Use was common, but frequency is not the same as quality
Reporting that 90% used ChatGPT establishes reach, not what the students did with it. A tool may be used to explain a concept, translate terminology, draft a study plan, generate practice questions or produce work submitted for assessment. Those uses have different implications for learning and integrity. The survey reports perceived improvements in efficiency and comprehension, but it did not administer a common knowledge test, compare an AI-supported assignment with a control assignment or inspect whether generated answers were accurate.
Frequency can also hide exposure time and dependence. Occasional checking and daily delegation may fall on the same ordinal scale without revealing how much thinking the learner retained. Health-professions education makes this distinction consequential: an apparently fluent answer may omit a contraindication or invent a reference, while repeated reliance can make it harder to tell whether a student can reason independently. Training needs to address verification, source use, uncertainty and when not to use a general-purpose model, rather than treating adoption alone as digital competence.[1]
Why the GPA association needs restraint
An association between use frequency and GPA can arise in several directions. AI might help some students organise material or receive rapid explanations. Equally, higher-performing students may be better at prompting, checking and integrating outputs. Programme, year of study, language confidence, internet access, prior academic ability and instructor policy could influence both use and grades. A cross-sectional comparison cannot separate these routes, even when a statistical test finds that the groups differ.
The paper reports the F statistic but the practical magnitude matters as much as significance. Without a randomised or longitudinal comparator, the result cannot tell an educator how much learning changed, whether a difference persisted or whether the association reflects grading practices. A verified GPA from institutional records would reduce one source of measurement error but still would not establish causality. Stronger evidence would follow students over time, record actual use and compare performance on validated assessments designed to test durable knowledge and clinical reasoning.[1]
Ethical concern was common and unevenly distributed
Sixty-one per cent of respondents reported ethical concerns. The level varied by gender, with p=0.0015, and by academic programme, with p=0.0114. Those values indicate differences unlikely to arise from sampling noise under the study's statistical assumptions; they do not explain why the groups differed or how large and educationally important each difference was. The finding should prompt investigation, not stereotypes about which students are more ethical or more ready for AI.
Ethical literacy in health education covers more than plagiarism. Students may enter identifiable patient information into external systems, accept biased or unsafe recommendations, cite fabricated sources or become unclear about who is accountable for an answer. Programmes can also create unfairness when some students have paid tools, better connectivity or more prior experience. A useful curriculum should combine privacy, evidence checking, disclosure and assessment rules with realistic practice. A prohibition alone may leave widespread use invisible rather than making it safer.[1]
What universities can act on without overstating the research
The immediate case for instruction does not depend on proving a grade benefit. When nine in ten surveyed students report using a tool and more than four in five report no formal training, educators have a governance and curriculum problem. Programmes can publish permitted-use rules, teach how to check a claim against primary clinical evidence, require disclosure when AI materially contributes, protect patient information and design assignments that reveal reasoning. Staff need parallel support so expectations are consistent across courses.
Assessment should distinguish practice from certification. Generative tools can help a student rehearse explanations or generate cases for critique, but a high-stakes examination may need to establish what the learner can do without assistance. Oral defence, supervised practical work, source annotation and comparison of model answers with guidelines can expose reasoning while still allowing appropriate tools in preparation. These are design responses to widespread use; the Bisha survey does not demonstrate that any particular intervention improves competence or reduces misconduct.[1]
What would change the assessment
A stronger next study would recruit several universities, verify academic outcomes, describe exact tool use and follow students over time. It should compare a defined training intervention with usual teaching, measure knowledge and clinical reasoning before and after, and report whether any gain persists. Outcomes should include accuracy, source verification, privacy behaviour and independent performance, not only satisfaction or perceived efficiency. Prespecified subgroup analysis could test whether training closes rather than widens differences in access and confidence.
Qualitative interviews could explain the gender and programme differences without reducing them to p-values. Direct review of anonymised student workflows could show where hallucinations, over-reliance or privacy risks occur. The assessment would become more favourable if structured teaching improved verified learning and safe practice without weakening independent reasoning. It would become less favourable if apparent efficiency reflected copied errors, concealed assistance or performance that disappeared when the tool was removed. For now, the evidence supports education about AI use, not a claim that AI education tools improve health-professions outcomes.[1]
Watch the background, then check the evidence
Illustrative photo · not a video frame
UNESCO · 8 min 30 sec · Background explainer
Teacher to Teacher: Is AI reshaping education for the better?
YouTube loads only when you choose; playback starts when you press play.
A conversation about how teachers can guide AI use while preserving pupils’ judgement. Watch for the difference between helping a learner think and supplying an answer.
Evidence check: Teachers’ perspectives provide context. They do not establish that a particular AI tool improves learning outcomes.
Published 2025-01-23. Source checked 14 October 2026. Check captions on the publisher’s video page.
Read the education evidencePhoto: Vitaly Gariev / Unsplash · Licence
Continue the story
Read next
Continue exploring assessment: compare the finding and its evidence limits.
Seven in ten US students said AI made take-home work unfair—what does the survey prove?
A weighted survey of 1,003 US college students found widespread concern about take-home assessment and self-taught AI skills. It measures beliefs and self-reports, not cheating rates or learning outcomes.
Read the next reportExplore the full source and evidence record
What this means for people
- Students are already using AI in health-related study while most surveyed students report no formal preparation; unclear rules can leave responsible use dependent on individual confidence and access.
- Teachers need assessment and disclosure practices that reveal reasoning and protect patient information without pretending that widespread use can be managed by silence.
- Patients are not direct participants in this study, but weak verification habits learned during training could matter later when graduates interpret clinical information.
Global context
The study adds current evidence from Saudi Arabia to a literature often dominated by North American and European institutions. Health curricula, language, assessment and access differ across regions, so neither the 90% use rate nor the reported associations should be treated as universal. The general policy question—how to teach verification and accountability when student adoption outruns formal instruction—travels more readily than the exact percentages.
What the evidence does not yet show
- Cross-sectional survey at one Saudi university; it cannot establish temporal order or causation.
- The denominator was 310 health-science students, not a national or international sample.
- AI use, perceived effects and GPA were self-reported rather than verified through system logs, institutional records or common assessments.
- The survey did not measure durable learning, clinical performance, patient outcomes or the accuracy of AI-generated material.
- Statistical differences by gender and programme do not explain their cause or practical size.
- The authors reported no financial support and no competing interests; they disclosed using AI-assisted tools for language editing and formatting, with human review.
What to watch next
- Multicentre Saudi and regional studies with verified academic outcomes
- Prospective trials of defined AI-literacy teaching rather than adoption surveys
- Performance on durable knowledge, clinical reasoning and source-verification tasks
- Privacy behaviour and the handling of patient information in student workflows
- Whether assessment redesign reduces hidden use without widening access gaps
Living evidence record
Impact record IAI-1C9QC0J
Evidence stage
Studied
Confidence
Supported
Reporting basis
Source analysis
Independent or research support
Present
Record status
Monitoring
Last checked
3 November 2026
Source trail
1 direct source across 1 source type.
People impact
Documented in this record.
Uncertainty
Limits and next checks are explicit.
Stages describe the evidence available—not whether a technology is good or bad. See the public method.
Single-source reporting disclosure
This record analyses one direct source. It can establish what BMC Medical Education published or reported, but it is not independent corroboration of every performance claim or predicted outcome. The confidence label will change only when broader evidence is added.
Evidence trail
Sources used for this report
Links checked 3 November 2026
Research question, topics and regions
How health-science students at the University of Bisha report using AI, what training they receive and how use relates to self-reported academic outcomes and ethical concerns
Topics: Health professions education · Student AI use · AI literacy · Assessment · Academic integrity
Regions: Saudi Arabia · Middle East
This report is labelled source analysis. We summarise and analyse source material in our own words; company statements remain attributed claims until independently supported. Translated summaries preserve the meaning of the original source and link back to it. Read our editorial standards.
Reader commentary
What does this mean where you are?
Tell us what you have seen at work, in your community or in everyday life. A question is welcome too. No account is needed; we review notes before publishing them.
Explore commentary across the portal →Published reader notes
0No published reader notes yet. You can start the evidence-led discussion above.
Prefer a private correction or response? Contact the newsroom.