'Co-pilot won't teach you to be a nurse': nursing academics' perspectives on GenAI use in undergraduate education

Synthesis: Dabkowski and colleagues interviewed 22 nursing academics from universities across Australia and New Zealand in February 2025 and analyzed the transcripts with reflexive thematic analysis, asking how they perceive and experience GenAI in undergraduate nursing education. Three themes emerged with nine minor themes beneath them: Navigating the Unknown: Ambiguity in GenAI Use, GenAI Challenging Nursing's Core Values, and Developing Ethical Nurses in a Digital Age. Participants reported inconsistent policy, divergent collegial attitudes, a drift back to invigilated and oral assessment, and fear that AI-mediated coursework lets students pass without the reasoning that nursing practice depends on. The authors argue GenAI troubles more than Academic Integrity procedure: it disturbs the basis on which educators judge whether a student is safe to practice.

Key Findings

1. Policy was absent, late, or not written for the people implementing it. Participants described "no clear, timely and practical guidance," with one saying "There was no proper information regarding AI use in our university... I think that's the first time something proper has come out" [P3]. Another questioned the provenance of the rules: "Sometimes I think the people who are in charge of policies are not on the ground dealing with the outcome of the policy" [P18], and a third that "the IT is probably evolving quicker than we can keep up with at the moment... I just don't think we're there at the moment" [P15]. Inconsistent practice followed: "Some academics absolutely ignore the use of AI or Grammarly in their course. But then the next one is absolutely stringent, so there needs to be some consistency" [P19].
2. Academic attitudes diverged sharply, from skepticism to acceptance. Some rejected the technology's creative value: "I think artificial intelligence is a bit of a misnomer. It's not original thought. It's just a reflection of what already exists" [P11]. Others saw resistance as futile: "We need to accept that it's here and we need to accept that students are going to use it. You can't put the genie back into the bottle" [P16], and "Maybe we need to stop fighting it and look at ways to use it productively to help our students" [P2]. One participant described tools as "an enthusiastic but somewhat unreliable assistant" [P11]; another worried about colleagues who tell students "You must not use it – if you use it, you will fail" [P21].
3. The line participants drew was replacement, not use. Editing and writing support were tolerated; generation of whole submissions was not. "It's purely when it replaces the student learning that I have an issue... they shouldn't be allowed to use it to replace learning" [P14]. Another framed the aim as critical engagement: "It's about how we can use it authentically, how we can use it to enable us to think critically, not take information for granted" [P1].
4. Assessment design shifted toward invigilation, orals, and process evidence. "We can't rely on a 2000 word essay anymore" [P4]; "eventually we are going to have to go back to written exams or... one-on-one so that people have to actually tell their knowledge" [P22]. Practical measures appeared alongside: rubrics "written in a way that you can identify AI" [P20], "five marks for showing us a screenshot of their literature search terms from a database" [P4], and oral vivas with several scenarios because "you can actually look at their clinical reasoning. It's that student at that time, that's their work" [P19]. Counter-evidence came too: "If students choose to use AI all the way through for their assessments and they get to the exam and know nothing, it's like 'what did you think was gonna happen?'" [P14].
5. Core values, not just marks, were seen as at stake. Participants positioned themselves as "the guardians of the profession" [P9] and as "the voice of the patient," arguing "If we can't do that and have nursing students that are honest and caring, then we're not doing our job" [P9]. One weighed reputation against care directly: "I would rather the students who learn from underneath me go out there and be phenomenal nurses who have no idea about AI than have a nurse go out there who's like, 'I'm an expert in AI' but have no idea how to care for a person" [P14]. Another tied the qualification to safety: "It needs to be legitimate because we need to consider patients' safety first and foremost" [P15].
6. Misconduct was read as a rehearsal for unsafe practice. "If a student's willing to breach AI then what shortcuts will they take in the clinical area?" [P2]; "those that get away with academic misconduct at the university will have issues with professionalism when they get onto the real world" [P9]. One participant linked dishonesty to documentation: "I have often seen nurses manipulate those results to not result in a clinical review" [P6], and another quoted students who had used AI undetected [P1]. On hallucinated content one was unequivocal: "If I think about the kind of nurse that we need in a healthcare setting, it's certainly not someone that sources fake references for information, because that's quite ethically and morally wrong" [P17].

How the study was conducted

The paper reports one analytic strand of a broader qualitative study of academic integrity in undergraduate nursing education. A qualitative descriptive design was implemented through semi-structured interviews conducted over Microsoft Teams, reported against the COREQ checklist, with reflexive thematic analysis following Braun and Clarke (2021). Three researchers coded inductively, then grouped codes into themes that the team reviewed and refined iteratively; the authors ran a side-by-side analysis of the academic integrity and GenAI strands rather than one combined account.

Recruitment was purposive, distributed through the Australasian Academic Integrity Network mailing list, a network of more than 1100 staff at approximately 140 institutions in Australia and New Zealand. Eligibility required current teaching or coordination in an undergraduate nursing program; permanent and sessional staff were both eligible. Twenty-two academics responded and all were interviewed, between 4 and 28 February 2025. The average interview was 39.05 minutes (range 19.53 to 59.45). The majority were female with five identifying as male; average academic experience was 6.4 years (range 1 to 20); nine held doctoral qualifications and several held academic integrity leadership roles. Data collection stopped once the team judged information power sufficient, having pre-estimated that 15 to 20 interviews would likely suffice (Malterud et al. 2016). Participants were offered the chance to check their transcripts, and all data were de-identified at transcription with numeric codes. All five researchers held Academic Integrity Officer roles at the time of analysis, four of them nursing academics, and the paper names this positionality as a possible influence on interpretation.

Knowledge translation and clinical reasoning at risk

The second theme moves furthest from integrity policing. The objection here is about knowledge translation rather than authorship: what happens to reasoning that is never practiced. "If using AI in assessment tasks to generate answers for caring for a deteriorating patient, how will the student develop skills?" [P13], and more bluntly, "The biggest issue that I find is that it's inhibiting their critical thinking" [P17]. The worry was explicitly longitudinal: "in three or five years' time, we're going to probably have a big cohort of students that are just not going to be fit for practice" [P9], reinforced by concern that "students potentially not thinking that there's base knowledge that they need to know, especially around patient assessment" [P15].

Under the third theme, participants argued for neither prohibition nor permissive access. They called for deliberate teaching of ethical, transparent use, staff development, and unit redesign, with one describing "lots of professional development around writing new units to incorporate GenAI" [P10], and rejecting avoidance because "it would be unhelpful to actually not allow our nursing students to learn how to use it efficiently, ethically, effectively, when others will" [P21]. Several described using GenAI's fallibility as a teaching device: AI-generated bots role-playing a graduate coordinator for interview practice, or deliberately inaccurate outputs posted for discussion so students "understand that not everything we get from it is accurate" [P21]. The paper's closing quote gives its title: "Copilot won't teach you to be a nurse" [P11].

What this means for practice

• Instructors. Review assessment design continually so tasks still evaluate a student's own reasoning and judgment, giving more weight to authentic and process-oriented formats; the paper follows Dawson et al. (2024) in treating validity as mattering more than cheating.
• Instructors. Embed GenAI literacy and ethical reasoning across the curriculum (Simms, 2025), and give nursing its own guidance aligned with the profession's ethical codes and standards instead of relying on generic institutional AI policy.
• Medical educators. Work through assessment redesign with a structured framework: the authors name TWO PRISMS as one way to do it (Winchester 2025).
• Curriculum designers. Pair redesign with staff development and co-design with staff, students and Stakeholders, which the authors say is needed for any of this to hold.

Limitations

• Recruitment ran through a single professional network — the Australasian Academic Integrity Network mailing list of more than 1100 staff at approximately 140 institutions — and 22 academics from Australia and New Zealand responded. Views of nursing academics outside that network and outside those two countries are not represented.
• The evidence is self-reported interview accounts about a sensitive topic. Participants described their own and colleagues' AI practice, policy experience and misconduct concerns, with no observation of teaching, marking or assessment behavior to corroborate them, so candor may have been constrained.
• All five researchers held Academic Integrity Officer roles while analyzing the data and four were nursing academics; the paper names that positionality as a possible influence on which questions were asked and how accounts were read.
• Data collection stopped at 22 interviews once information power was judged sufficient, against a pre-estimate that 15 to 20 would likely suffice, and the authors note that GenAI tools and institutional responses are changing fast enough that the findings describe a particular moment.

Citation

Dabkowski, E., Missen, K., Allen, L., Whitehead, D., & Worn, R. (2026). 'Co-pilot won't teach you to be a nurse': Nursing academics' perspectives on GenAI use in undergraduate education. International Journal for Educational Integrity, 22(11).