ChatGPT is quietly reshaping medical school, but most students are using it wrong

Two-thirds of future doctors, pharmacists, and dentists are already using ChatGPT, but a new study from the United Arab Emirates suggests they are mostly using it as a glorified search engine rather than as the powerful learning tool it could be. The research, published in BMC Medical Education, surveyed 304 healthcare profession students at the University of Sharjah and found that 67.4 percent had used ChatGPT in the past three months. Yet the most impactful applications, such as practice testing and personalized feedback on assignments, remain strikingly underutilized. The findings carry a warning: students with lower AI literacy risk being left behind in an education system increasingly driven by artificial intelligence.

The study is among the first to examine ChatGPT adoption among medical students in the UAE, where usage had previously gone unexamined. The researchers enrolled students from four colleges: health sciences, medicine, pharmacy, and dentistry. Nearly 79 percent of participants were female, reflecting the composition of the surveyed cohort. Data were collected through a self-administered questionnaire incorporating validated instruments. The TAME-ChatGPT tool measured student perceptions of the benefits and risks of the technology, while the Meta AI Literacy Score assessed how well students understood AI systems. Analysis was performed in SPSS 22 using descriptive statistics, correlation, and nonparametric tests, with a significance level of 5 percent.

The headline result is that ChatGPT use was significantly associated with positive attitudes toward integrating the technology into the curriculum, with a p-value below 0.001. Students who used ChatGPT also tended to have higher GPAs, a link significant at p equal to 0.003, and use varied by college degree at p equal to 0.006. Perhaps most intriguingly, students who deployed ChatGPT for practice tests, assignment feedback, and written work reported lower academic stress, a relationship significant at p below 0.01. This suggests the chatbot is not merely a shortcut for the anxious; in certain configurations, it may actually function as a stress-reduction mechanism by providing low-stakes opportunities to test knowledge and refine work before submission.

But the pattern of use reveals a gap between what ChatGPT can do and what students actually ask of it. The majority, 64.1 percent, used the model primarily as an information source, essentially treating a generative AI system as a conversational encyclopedia. Among the 32.6 percent who did not use ChatGPT at all, the main reasons were a perceived lack of usefulness, cited by 49.4 percent, and lack of knowledge about the tool, cited by 40 percent. In other words, nonuse was driven less by principled objections and more by indifference and unfamiliarity, a combination that hints at how quickly informal adoption can outpace structured guidance.

AI literacy emerged as the study’s central variable. Students scored a moderate 6.06 out of 10 on the Meta AI Literacy Scale, with an interquartile range spanning 2.77 to 7.89, meaning the middle half of students fell within that band. Literacy scores were strongly linked to ChatGPT use at p below 0.001. Higher AI literacy correlated with reduced anxiety about the technology, greater perceived usefulness, and more positive behavioral attitudes toward it. The researchers frame this as an equity issue: students with limited AI skills risk being disadvantaged in AI-driven education, not because they lack ability but because they lack the fluency to extract value from the tools their peers are already exploiting.

Technically, ChatGPT belongs to a class of systems known as large language models, which use generative AI to produce human-like, context-aware responses. These models are trained on vast text corpora and generate output by predicting likely continuations of a prompt, which explains both their fluency and their capacity for confident error. In an educational setting, this dual nature is exactly why the researchers distinguish between superficial uses, such as looking up facts, and deeper uses, such as requesting feedback on a draft or generating practice questions that force active recall. Learning science has long shown that retrieval practice and timely feedback are among the most effective study techniques; ChatGPT can deliver both on demand, but only if students know to ask.

The study’s authors, a team led by Riham Hafiz Mohamed Elbashir, Farah Khaled Ismail Youssef, and Sanar Ghassan Abdelrahman Al-Qaimari, all contributing equally, conducted the work as part of the Community-Based Research Program at the University of Sharjah. The research received ethics approval from the university’s Research Ethics Committee and was carried out in accordance with the Declaration of Helsinki. Participation was voluntary and anonymous, with implied informed consent taken from proceeding with the online questionnaire. The article was published as open access under a Creative Commons Attribution 4.0 license and is being shared early to provide faster access to peer-reviewed, accepted research, with a final Version of Record to follow.

What makes the findings resonate beyond the UAE is the portrait they paint of informal AI adoption across health professions education. Students are not waiting for their institutions to decide how ChatGPT should be used; they are experimenting on their own, guided by curiosity and peer influence rather than pedagogy. The result is a patchwork in which the students most comfortable with AI extract the greatest benefit, while others either dismiss the technology as useless or avoid it because they do not understand it. The study’s conclusion that structured AI literacy programs should be integrated into curricula is a direct response to this patchwork, aimed at ensuring equitable access to the benefits while mitigating the risks.

The researchers argue that empowering all learners to use ChatGPT more effectively and safely would facilitate smoother integration of AI into medical education. Their prescription is not prohibition but fluency: teaching students how large language models work, where they fail, and how to prompt them for the uses with the strongest evidence of benefit. Given that the students most likely to benefit from ChatGPT’s stress-reducing, feedback-generating capabilities are precisely those least likely to discover them on their own, the case for structured instruction is difficult to ignore. As AI-driven education accelerates, the gap between students who understand these systems and those who do not may become one of the defining equity challenges of the coming decade.

For now, the study stands as a snapshot of a profession in transition. Two-thirds of the next generation of healthcare providers are already consulting a chatbot about the material they will one day use to treat patients. Whether that consultation becomes a superficial fact-lookup or a rigorous practice partner depends largely on what educators do next. The evidence from Sharjah suggests that the technology is already in the classroom; the question is whether the classroom will teach students how to use it.

Subject of Research: ChatGPT in medical education: usage, perceptions, and the role of AI literacy among healthcare profession students

Article Title: ChatGPT in medical education: usage, perceptions, and the role of AI literacy among healthcare profession students

Article References: Elbashir, R. H. M., Youssef, F. K. I., Al-Qaimari, S. G. A., Al Hajji, A., Almualla, A. A. N. A., Aldhanhani, M. A. R. A., Al-Salman, A. M. A., & Hussein, A. (2026). ChatGPT in medical education: usage, perceptions, and the role of AI literacy among healthcare profession students. BMC Medical Education. https://doi.org/10.1186/s12909-026-10148-3

Image Credits: AI Generated

DOI: 10.1186/s12909-026-10148-3

Keywords: ChatGPT, medical, education, usage, perceptions, role, literacy, healthcare, profession, students, scientific research