ABSTRACT
Background
Large language model tools are increasingly used in higher education, offering opportunities to support self‐directed learning. In nursing education, course‐specific AI virtual tutors may provide contextualised support while addressing concerns about content accuracy and alignment; yet empirical evidence remains limited.
Objective
This study evaluated the use and perceived impact of a co‐designed AI‐powered virtual tutor embedded in a graduate‐level Master of Nursing (MN) course. We explored how students used the tutor, their perceptions of benefits and limitations, and its influence on learning and engagement.
Methods
A pilot study using a mixed‐methods explanatory sequential design was employed. The tutor was trained on course‐specific materials and integrated into the institutional learning management system. Data included anonymised usage logs and user interactions coded using Bloom's Taxonomy of Educational Objectives, post‐course surveys assessing AI self‐efficacy, usability, and learning impact, and semi‐structured interviews with students and teaching assistants (TAs). Quantitative and qualitative strands were integrated through a joint display.
Results
A total of 651 interactions by individuals within a group of ~120 MN students were logged. Interactions peaked in evenings and around assignment deadlines. Most interactions reflected lower‐order education processes, with more application and analysis later in the course. Eleven participants completed surveys; students reported high AI self‐efficacy and moderate tutor use. Perceived usefulness was mixed, but most reported the tutor enhanced both lower‐ and higher‐level learning and recommended its future use. Interviews revealed that students valued the tutor's immediacy and course‐specific accuracy, while TAs noted efficiency gains. Reported challenges included usability issues, scope limitations, privacy concerns, and risk of over‐reliance on the tool.
Conclusions
A co‐designed AI virtual tutor was feasible and valued for contextual relevance, though perceived usefulness was variable. Findings support responsible, pedagogically integrated use of AI tutors in graduate nursing education.
Keywords: artificial intelligence, bloom's taxonomy, graduate education, large language models, mixed‐methods, nursing education, virtual tutor
What Does This Paper Contribute to the Wider Global Clinical Community?
Demonstrates the feasibility and educational value of a co‐designed, course‐specific AI virtual tutor embedded in graduate nursing education, providing a scalable model for responsible AI integration in clinical training contexts.
Shows that AI tutors can enhance learners' confidence, conceptual understanding, and engagement when aligned with curricular goals and pedagogical values.
Highlights the need for ethical, privacy‐conscious, and critically engaged approaches to AI use in nursing education to prevent over‐reliance and preserve clinical reasoning skills.
1. Introduction
Artificial intelligence (AI) tools powered by large language models (LLMs) have entered higher education at an unprecedented pace since 2022 (Abdelwahab et al. 2025; Tillmanns et al. 2025). OpenAI's ChatGPT, Microsoft's Co‐Pilot, and similar systems are increasingly used by university students to retrieve information, draft text, and generate explanations to facilitate learning, among other learning activities (Labadze et al. 2023). As these tools can provide rapid access to knowledge, support self‐directed learning, and provide personalised feedback, they are perceived by students as highly useful learning aids (Schei et al. 2024; Srinivasan et al. 2024). Two recent meta‐analyses of over 50 experimental studies showed ChatGPT to be effective in improving university students learning performance, learning perception, and higher‐order thinking (Deng et al. 2025; Wang and Fan 2025).
Building on emerging evidence in higher education, nursing education offers a distinct context in which AI‐driven tools can extend beyond general academic support to actively enhance concept comprehension, clinical reasoning, and simulated practice scenarios settings. Specific transformative opportunities offered by LLMs to foster personalised and interactive learning in this context include adaptive quizzes, automated feedback, virtual tutors, and virtual patient simulators (Harrington et al. 2025; Topaz et al. 2025). Evidence from a single blind randomised controlled trial showed nursing students who used ChatGPT demonstrated improved problem‐solving skills and increased competency in the nursing process compared to those who did not use ChatGPT (Arkan et al. 2025). Students also report generative AI tools to be valuable classroom learning aids and helpful in interdisciplinary groupwork (Ruiz‐Rojas et al. 2024). More broadly, national organisations are also recognising the potential of AI to bridge gaps in teaching capacity and support diverse learner needs. For instance, the American Association of Colleges of Nursing (AACN) focused its July 2025 assembly solely on exploring how AI might innovate curricula, personalise learning, and prepare nurses for emerging workforce demands (AACN 2025).
AI integration in nursing and health professions is evolving, and the distinction between general‐purpose chatbots such as ChatGPT and specialised AI virtual tutors is being carved out (Mojjada et al. 2024). While ChatGPT draws on vast, open‐domain datasets and provides broad, conversational support, curriculum‐specific AI virtual tutors can be trained on course materials, give real‐time, targeted feedback, and provide distinct learning experiences for students (Mojjada et al. 2024).
Despite the promising educational benefits ushered in by LLMs, concerns about ethical application, including data quality, privacy, and academic integrity, exist and require attention (Hobensack et al. 2024; Sun and Hoelscher 2023). Educators have expressed uncertainty about the accuracy of AI outputs, risks of hallucination, and the challenge of designing institutional policies for responsible integration (Abdulai and Hung 2023; Athilingam and He 2024; Eke 2023). Concerns further include over‐reliance on AI outputs, reduced student–instructor interaction, and potential erosion of critical thinking and clinical reasoning skills (Sharifi Kelarijani et al. 2024). Moreover, excessive dependence on AI risks diminishing communication competencies, fostering technological addiction, and undermining the balance between human judgement and technological assistance that is essential to safe nursing practice (Sharifi Kelarijani et al. 2024). The effective and ethical use of such tools thus depends on the ability of instructors and teaching assistants (TAs) to support? thoughtful and critical use by students (Booth et al. 2021; Ronquillo et al. 2021).
In graduate‐level nursing education, students must synthesise theory, evidence, and practice to develop professional judgement skills. AI virtual tutors can offer flexible, responsive support at scale that helps students acquire theoretical knowledge, integrate evidence, and make ethical reasonings. At the Master of Nursing (MN) level, students are expected to advance beyond foundational competencies to become system‐thinking leaders, capable of translating research evidence into interprofessional practice and health policy (AACN 2011; Canadian Nurses Association 2019). Across graduate nursing training streams, students must demonstrate competencies that extend beyond technical knowledge to include advocacy, programme evaluation, and quality improvement (AACN 2011).
Co‐designed AI virtual tutors for graduate nursing students have the potential to directly address key limitations of LLMs in the classroom, including hallucinated outputs, misalignment with curricular goals, and concerns around academic integrity (Abdulai and Hung 2023; Alrazeeni et al. 2026). Developed collaboratively by education and information technology experts, such tools can be tailored to programmatic and course objectives, aligned with institutional policies, and responsive to learner needs. Applying a co‐design framework to the development of an AI virtual tutor thus offers a pathway to develop and integrate generative AI responsibly (Kilfoy et al. 2024), leveraging the technology's strengths in personalisation and immediacy while safeguarding pedagogical integrity and professional standards (Hemment et al. 2025). Further, the collaborative nature of the co‐design process embeds contextual expertise directly into the model and preserves nuance in complex learning environments (Hemment et al. 2025). However, empirical studies have yet to examine the implementation or learning impact of such a tutor in graduate nursing education. This study addresses this gap by providing the first evaluation of a co‐designed AI virtual tutor in the MN curriculum. Our specific objectives were to: (1) examine patterns of tutor use by students and TAs, including the frequency and purposes of interaction; (2) explore student and TA perceptions of the benefits and challenges to tutor use; and (3) describe overall student learning experiences and the perceived role of the tutor in supporting education.
2. Methods
2.1. Study Design
This evaluation employed a mixed‐methods explanatory sequential design to understand how graduate nursing students interacted with and perceived the AI virtual tutor (‘Program Planning Pal’) in a Program Planning and Evaluation course within the MN programme at the University of Toronto. Through this approach, quantitative data (such as usage statistics and surveys) were collected and analysed alongside qualitative semi‐structured interviews to provide a comprehensive understanding of tutor impact.
The AI tutor was built on a LLM which is an artificial intelligence system trained on extensive amounts of text to generate human‐like responses. To ensure the accuracy and relevance of the tutor's answers, prompt engineering was used, which involves crafting specific input instructions to guide the AI's responses. We adhered to established research reporting guidelines (STROBE for observational studies and COREQ for qualitative research; von Elm et al. 2007; Tong et al. 2007) (Appendix A).
2.2. Setting and Participants
Student participants were individuals from a group of ~120 Master of Nursing (MN) students enrolled across four sections (A–D) of the Winter 2025 Program Planning and Evaluation for Nurses course. All ~120 MN students had access to the AI tutor. The course introduces principles of programme development, implementation, and evaluation in clinical and health system contexts, emphasising evidence use and outcome measurement to improve practice and system performance. Students represented three graduate streams: Clinical Leadership, Health Systems Leadership and Administration, and Nurse Practitioner. The 12‐week course was followed by a final clinical practicum, where students applied programme planning concepts in real‐world settings. Although the primary study period spanned January to March, all students retained access to the virtual tutor during the practicum (April–June). Only students in the Clinical and HSLA streams were expected to apply programme planning principles at that stage; thus, usage data include interactions from both the course and practicum, reflecting continued engagement as students translated theory into practice.
Eligibility criteria for survey and interview participation were having used the AI virtual tutor at least once during the course. All seven course teaching assistants (TAs) were eligible to participate as they had supported course delivery and had access to the tutor during the study period. Recruitment was conducted via email after course completion, and participation required informed digital consent.
As this study was designed as a pilot evaluation embedded within a graduate course, a formal a priori sample size calculation was not conducted. The available sample was determined by course enrollment and reflects a pragmatic approach consistent with feasibility‐focused educational evaluations. For the qualitative component, we interviewed all students and TAs who agreed to participate.
2.3. AI Tutor Design and Deployment
Our co‐designed AI tutor leveraged a pre‐trained LLM, which we configured to prioritise course‐specific materials comprising lecture slides, readings, transcripts, and curated open‐access resources drawn directly from the assigned readings as its main information source. This was designed to mitigate common limitations of LLMs such as hallucinations and content misalignment.
Our tutor co‐design engaged course instructors (C.H.C., L.A.J., C.M.), an educational technologist (N.M.), and an information system analyst (C.C.S.). Microsoft CoPilot was employed during early testing phases to generate sample responses to common course interactions. Instructors reviewed and ranked these outputs according to accuracy, alignment with learning objectives, and pedagogical appropriateness. Instructor feedback then informed the fine‐tuning of the tutor's response patterns, enabling us to identify systematic inaccuracies and refine prompt engineering strategies. In addition, several candidate models were piloted to evaluate their relative performance on dimensions such as accuracy, coherence, and alignment with course content. Comparative testing ensured that the selected model balanced responsiveness with reliability in the educational context. This iterative process of testing and instructor‐led validation enhanced the likelihood that the tutor reflected both the substantive content of the course and the pedagogical values underpinning graduate nursing education.
We (C.H.C., L.A.J., N.M., C.C.S.) further co‐designed an onboarding process for the students in conjunction with a university specialist in generative AI pedagogy. The objective of the onboarding element is to introduce students to appropriate AI virtual tutor use and encourage its uptake. Onboarding involved a presentation on the use of the virtual tutor, its privacy protocols, and effective prompting, and a custom tutor access agreement for students. The presentation was recorded and made available as a reference for students. The access agreement highlighted expectations for responsible use, with emphasis on the student's responsibility for critical appraisal of AI‐generated content.
The tutor was deployed in the University of Toronto's secure Learning Management System (LMS) and made accessible to students through authenticated login. All student conversations with the AI tutor were logged in a secure private cloud managed by the university's Information Technology Services (ITS). Data logs contained only anonymised text data; no usernames, email addresses, or other personally identifying information were collected or stored. ITS subsequently generated a conversation dashboard for instructors, displaying aggregate metrics (e.g., total conversation count per day) and representative anonymised excerpts. This setup enabled instructors to observe usage trends and student engagement patterns without compromising individual privacy.
2.4. Data Collection
Study data were collected from three sources: (1) anonymised analytics from student conversations collected via the instructor dashboard (providing objective data on usage frequency, interaction types, and temporal usage trends); (2) REDCap‐based student surveys (providing student reports of tutor usability, benefits and challenges, and impact on engagement and understanding, as well as AI application skills per the ‘Use and Apply’ subscale of the validated Meta AI Literacy Scale (MAILS; Carolus et al. 2023)); (3) virtual semi‐structured interviews. Regarding the MAILS, item scores on the tool range from 0 = ‘not at all’ and 10 = ‘very well’. Scores of 0–3 indicate low AI literacy (limited knowledge, confidence, or ability to apply AI concepts or tools); scores of 4–6 indicate moderate AI literacy (emerging understanding and some practical competence); and scores of 7–10 indicate high AI literacy (strong conceptual understanding and confidence in use). Overall, the MAILS demonstrates acceptable‐to‐good structural validity based on confirmatory factor analysis and generally strong internal consistency across subscales (Cronbach's α > 0.70), with the ‘Use & Apply’ AI subscale showing excellent reliability (α = 0.93; Carolus et al. 2023).
All students were emailed the REDCap survey link and invited to participate in the survey and an interview after the course ended but before the end of their practicum period. Built‐in REDCap validation ensured no missing data. Individual interviews were conducted virtually once. Our instructor‐developed interview guides explored in‐depth views of the tutor's role, benefits, challenges, and learning impact. Interviews were conducted by a trained external research assistant who was unknown to participants, audio‐recorded, and transcribed verbatim. Field notes were taken. Across both surveys and interviews, participants could decline to answer questions if they felt uncomfortable.
2.5. Data Analysis
AI–student interaction records were retrieved from Dataverse and stored within the university's secure private cloud. Following data cleaning to descriptive metadata and system‐generated code, the resulting text‐based interaction dataset was analysed. All interactions entered in the tutor were coded according to Bloom's Taxonomy of Educational Objectives, which comprises six cognitive domains of learning that can be organised into a hierarchy from lower to higher order processes: remembering, understanding, applying, analysing, evaluating, and creating (Anderson et al. 2001). For the purposes of this study, each interaction was coded by NM as pertaining to one of the Taxonomy's six cognitive domains. NM subsequently coded each interaction as one of four knowledge dimensions: factual (basic element), conceptual (interrelationships among elements), procedural (how to do something), and metacognitive (awareness of thinking and strategies for learning) (Anderson et al. 2001). Quantitative survey data, including AI Literacy Tool item scores, and usage data were analysed using descriptive statistics in REDCap and Excel. Qualitative interview transcripts were analysed thematically with reference to the study objectives using Excel, following a constant comparative method. Two researchers (C.H.C. and L.A.J.) independently coded data and consensus was planned to resolve coding differences, though none occurred. Quantitative and qualitative data were integrated using an explanatory sequential approach, in which interview findings were connected to and used to explain quantitative usage patterns through linking, comparing, and generating interpretations at an in‐person meeting between C.H.C. and L.A.J. (Hwang et al. 2025). A joint display table facilitated this integration.
2.6. Ethical Considerations
The University of Toronto Research Ethics Board approved this study (#48144). Participants provided informed digital consent through REDCap (Lawrence et al. 2020), with assurances of confidentiality and voluntary participation. Data were securely stored and anonymised to protect privacy.
3. Results
3.1. Frequency of AI Tutor Usage
Interactions were made between January and June 2025, with a total of 651 historical AI–student interaction records retrieved. The highest activity levels occurred during evening hours, peaking around 19:00. Student usage of the tutor varied across course sections: Section A (44%), Section B (37%), Section C (12%), and Section D (7%). The most common question types were ‘What’ (16%) and Writing Support (5%), followed by ‘Examples’, ‘How’, and ‘Where’ questions. The heaviest usage occurred at the launch of the tutor, when 71 interactions were submitted on a single day (Figure 1). Subsequent spikes were observed over the semester, each coinciding with major assignment deadlines. Interactions tapered at the close of the course; however, an increase in tutor interactions was observed in June and corresponded with students in the Clinical and HSLA streams completing their final clinical practicums.
FIGURE 1.
Frequency and use of the AI virtual tutor over time. [Colour figure can be viewed at wileyonlinelibrary.com]
3.2. Purpose of AI Tutor Use
Coded student interactions spanned all six dimensions of Bloom's taxonomy (Table 1). Most interactions involved lower‐order cognitive processes (n = 419, 68%), namely remembering and understanding. These included factual clarifications (e.g., definitions, recall of lecture content) such as ‘Which week(s) covered logic models?’ and conceptual explanations ‘What is a fishbone (cause‐and‐effect) diagram?’. Higher‐order processes (n = 198, 32%) were also represented, especially in the applying and analysing domains. ‘Applying’ interactions were usually in the conceptual knowledge dimension and included student questions such as ‘can you give me an example of an outcome objective for a program’. ‘Analysing’ interactions were related to concepts (e.g., ‘Is a logic model same as a theory or framework in program planning?’). Interactions related to the evaluation and creation domains were less frequent but present.
TABLE 1.
Interactions to AI virtual tutor according to cognitive domain and knowledge dimension.
| Cognitive domain | Interaction n (%) | Knowledge dimension | Interaction n (%) | Example interaction a |
|---|---|---|---|---|
| Remembering | 124 (20%) | |||
| Factual | 84 (68%) | ‘What does PDSA stand for again?’ | ||
| Conceptual | 40 (32%) | ‘what 3 theories make up intervention theory’ | ||
| Procedural | 0 | |||
| Metacognitive | 0 | |||
| Understanding | 87 (13%) | |||
| Factual | 44 (51%) | ‘can you expand step 4 of the PDSA cycle’ | ||
| Conceptual | 40 (46%) | ‘Can you describe the Behaviour Change Wheel to me?’ | ||
| Procedural | 2 (2%) | ‘Hi! I am having a hard time starting the first assignment’ | ||
| Metacognitive | 1 (1%) | ''for a needs assessment ‐ we define the problem, analyze contributing factors, and identify related needs, and think about whether our assumptions/evidence/stakeholder perspectives are shaping our understanding of the problem appropriately, is this right?' | ||
| Applying | 209 (32%) | |||
| Factual | 17 (8%) | ‘Where can I find Ontario healthcare data around falls in rehabilitation setting?’ | ||
| Conceptual | 107 (51%) | ‘What is an appropriate planning framework for a community‐based program?’ | ||
| Procedural | 84 (40%) | ‘help me design a logic model for a Mental health crisis assessment pathway in the ED’ | ||
| Metacognitive | 1 (< 1%) | ‘Give me a question about the nursing’ (Sic) | ||
| Analysing | 102 (16%) | |||
| Factual | 6 (6%) | ‘What is the difference between a needs assessment and needs analysis for [X] program?’ | ||
| Conceptual | 73 (72%) | 'How do the PARIHS and CFIR frameworks differ in how they organize implementation barriers and facilitators for a mobile cancer screening program aimed at increasing screening rates in rural Ontario?' | ||
| Procedural | 23 (23%) | ‘Are these multiple implementation strategies?’ | ||
| Metacognitive | 0 | |||
| Evaluating | 87 (13%) | |||
| Factual | 0 | |||
| Conceptual | 24 (28%) | ‘how would i use a t test for pre‐and post‐surveys?’ | ||
| Procedural | 49 (56%) | ‘I am developing a 12‐week health coaching program, and implements follow‐up at 6, 9 and 12 months. When would be appropriate to do a summative evaluation for the program’ | ||
| Metacognitive | 14 (17%) | 'Can you help me evaluate whether the multiple‐choice questions I created about the theory of the problem accurately assess the key concepts? Help me identify which questions are strongest, which need revision.' | ||
| Creating | 7 (1%) | |||
| Factual | 0 | |||
| Conceptual | 0 | |||
| Procedural | 6 (86%) | ‘for example, our problem is the nurse turnover in hospitals. We define the problems, and then analysis who it affects, its context (means people, equipment, environment, policies, etc.), than next, we gonna do… Please list each steps and the contents in each steps, using the nurse turnover example. Thank you’ | ||
| Metacognitive | 1 (14%) | 'Can you help me develop a theory of the problem for inefficient prescribing practices by organizing the main contributing factors? point out whether my assumptions are evidence‐informed, and whether I have missed important system‐level factors that i need to include.' |
Student or teaching assistants entries were not edited for grammar or punctuation.
Figure 2 displays interactions by cognitive domain over time. Interactions reflecting remembering were most frequent early in the semester. Interactions coded as the higher‐order domains of ‘analysing’, ‘evaluating’, and ‘applying’ reached their highest levels in the last half of the semester. Interactions coded as creating were consistently low throughout.
FIGURE 2.
AI virtual tutor interactions by cognitive domain over time. [Colour figure can be viewed at wileyonlinelibrary.com]
3.3. Survey Insights Into Usage, Benefits, Challenges, and Learning Impact
Eleven participants completed the post‐usage survey, including nine graduate nursing students and two TAs (Table 2). All students were female, most were aged between 25 and 29 years (67%), and most were recent graduates of a Bachelor of Science in Nursing programme (56%) graduated between 2020 and 2023. Both participating TAs were doctoral nursing students with several years of clinical and teaching experience. They represented diverse perspectives, with one male and one female participant, and their reported years of nursing practice ranged from five to fifteen years.
TABLE 2.
Demographics of students participating in study surveys (N = 9).
| Characteristic | n | % |
|---|---|---|
| Age | ||
| < 30 years | 6 | 67 |
| ≥ 30 years | 3 | 33 |
| Gender | ||
| Female | 9 | 100 |
| Graduation year from Bachelor of Nursing degree | ||
| Before 2020 | 4 | 44 |
| 2020 or later | 5 | 56 |
| Years practiced | ||
| 1–5 years | 3 | 33 |
| 5–10 years | 6 | 67 |
| Programme | ||
| Clinical MN | 5 | 56 |
| NP MN | 4 | 56 |
| HSLA | 0 | 0 |
Student and TA respondents reported high AI self‐efficacy (median: 8/10 across items) and most (91%) had prior experience with generative AI use (Table 3). Reported use of the tutor was moderate, with most respondents engaging ‘every once in a while’ (55%). Perceptions of usefulness and effectiveness were mixed: 72% rated it as only ‘a little’ or ‘not useful’, while 27% found it ‘effective in supporting understanding’. Usability and accessibility ratings were adequate. Overall, 55% reported the tutor enhanced their learning. Student opinions were divided on whether they liked the tool, but most (64%) participants recommended continuing to offer an AI virtual tutor in future course offerings.
TABLE 3.
Student and teaching assistant AI virtual tutor survey responses (N = 11).
| Focus area | Item or question | Median (range) | Response option | n (%) |
|---|---|---|---|---|
| History and comfort with AI tools | I can operate AI applications a | 8 (7–10) | ||
| I can use AI applications to make my life easier a | 8 (5–10) | |||
| I can use artificial intelligence meaningfully to achieve my goals a | 8 (4–10) | |||
| I can interact with AI in a way that makes my tasks easier a | 8 (4–10) | |||
| I can work together gainfully with AI (e.g., create useful outputs) a | 8 (6–10) | |||
| I can communicate gainfully with AI (e.g., engage effectively with chatbots) a | 8 (4–10) | |||
| Have you previously used generative AI tools (i.e., virtual tutors, chatbots, or GPTs) outside of this course? | Yes | 10 (91%) | ||
| No | 1 (9%) | |||
| Comfort level using textual generative AI tools (e.g., ChatGPT, virtual tutors) | Very uncomfortable | 0 | ||
| Slightly uncomfortable | 2 (18%) | |||
| Neutral | 1 (9%) | |||
| Slightly comfortable | 5 (45%) | |||
| Very comfortable | 4 (36%) | |||
| Learning effectiveness, usability, and accessibility | Frequency of using the Program Planning Pal Virtual Tutor during the course | Only once | 2 (18%) | |
| Not much | 3 (27%) | |||
| Every once in a while | 6 (55%) | |||
| All the time | 0 | |||
| Usefulness of the Program Planning Pal Virtual Tutor | Not useful | 3 (27%) | ||
| A little useful | 5 (45%) | |||
| Moderately useful | 1 (9%) | |||
| Useful | 2 (18%) | |||
| How effective was the Virtual Tutor in supporting your understanding of the course material? | Did not support | 3 (27%) | ||
| Supported a little | 3 (27%) | |||
| Neutral | 2 (18%) | |||
| Effective | 3 (27%) | |||
| How easy was it to use the Virtual Tutor? (usability) | Very challenging | 1 (9%) | ||
| A little difficult | 1 (9%) | |||
| Neutral | 3 (27%) | |||
| Somewhat easy | 5 (45%) | |||
| Very easy | 0 | |||
| How accessible was the Virtual Tutor? | Very challenging | 0 | ||
| A little difficult | 2 (18%) | |||
| Neutral | 4 (36%) | |||
| Somewhat easy | 4 (36%) | |||
| Very easy | 1 (9%) | |||
| Did the Virtual Tutor enhance your learning experience? | No | 5 (45%) | ||
| Yes | 6 (55%) | |||
| Did the Virtual Tutor influence your engagement with the course content? | No | 8 (73%) | ||
| Yes | 3 (27%) | |||
| Did you like the Virtual Tutor? | No | 1 (9%) | ||
| Unsure | 5 (45%) | |||
| Yes | 5 (45%) | |||
| Would you recommend offering a Virtual Tutor in future courses? | No | 0 | ||
| Unsure | 4 (36%) | |||
| Yes | 7 (64%) |
Questions derived from the ‘Use and Apply’ subscale of the AI Literacy Tool on AI application skills.
3.4. In‐Depth Views on Benefits, Challenges, and Learning Impact
Three ~60‐min‐long interviews were conducted with students and two with TAs. Our analysis derived five themes from the interview data, which describe the perceived benefits, challenges, and learning impact of the AI virtual tutor.
3.4.1. Methods of Engagement
Students reported primarily using the tutor to clarify core course concepts and confirm their understanding of terminology. One student explained,
In the beginning of the semester…I used [the AI virtual tutor] mostly to supplement my learning. When I was reviewing lecture slides, if there was something confusing, I would go back to it. And then at the end of the course, when final assignments were due, I used it more so to clarify key terms, to make sure that I was using them properly in my papers. (Student 1)
The tutor was also leveraged to request simplified explanations (e.g., ‘explain Gantt charts as if I were in grade 5’) or to identify the lecture week in which specific content was addressed. TAs used the tutor as a reference point when marking, particularly to resolve ambiguous student statements or confirm their own understanding of a course concept. One TA noted, ‘When I'm marking, and I come to sticky points within the student paper…I quickly would go to the virtual tutor, refer to it, and then apply the information in the marking processes’ (TA 1).
3.4.2. Satisfaction and Perceived Benefits
Students and TAs both expressed satisfaction with the AI tutor. Students valued the immediacy and contextual accuracy of responses. The tutor's pre‐training on course‐specific materials was considered particularly beneficial, as it reduced the likelihood of irrelevant information being provided compared with other tools like ChatGPT:
I did…use an external chatbot—to ask the same prompt. It gave me a lot of information, because it's assuming that I wanted to approach it from, a business or economic standpoint. So, it would overwhelm me with information. I enjoyed that this chatbot was focused solely on this course and gave [answers in] the right context. (Student 1)
The integration within the LMS was also seen as convenient, especially when working on assignments: ‘Since I'm using it for assignments, I often have [the LMS] open so it just encourages use. It's easier to navigate that way’ (Student 2).
From the TA perspective, the tutor relieved some workload pressures by reducing the number of basic clarifications needed, allowing more time to address advanced questions. TAs also discussed improved assignment quality that they attributed to student use of the tutor. One TA summarised: ‘It makes for a more enjoyable, less cumbersome marking experience…the students are answering the questions…because they understand the material, and they're confident in their understanding’ (TA 1).
3.4.3. Challenges and Suggestions for Improvement
Students noted the interface was not as user‐friendly as those of external tools and the tutor and slower response times that limited usability. Students also found that the tutor was unhelpful when interactions extended beyond the scope of course materials. Concerns about over‐reliance on the AI tutor were also raised: ‘I did find a couple of students who relied solely on the outputs…So, I foresee a situation where students may want to totally rely on what the virtual tutor produces and not put in efforts to critically think about, what are the concepts?’ (TA 2). Trust and privacy emerged as barriers to adoption despite our onboarding process, with students mentioning some hesitancy to use the tutor lest their prompts could be linked to their names. Initial navigation difficulties were reported but described as minor and temporary. Enhancements to improve accessibility and functionality included creating a more intuitive user interface (streamlining navigation within the LMS and optimising chat layout), reducing latency in response times, embedding pre‐programmed prompts for common questions, and enabling visual assets including diagrams and tables. Regular updates to the knowledge base were also suggested to maintain alignment with evolving course content.
3.4.4. Impact on Student Learning
Students reported confidence in their ability to access high‐quality learning support and answers at any time. As one student explained, ‘The chatbot was available 24/7. If I asked my professor or TA a question, it would take time for them to respond. The chatbot was especially helpful when I was working on assignments late at night or on weekends’ (Student 1). This was particularly relevant for the students who were working shifts (e.g., night shifts) at the hospital. The virtual tutor was also considered a valuable tool to deepen student engagement with course concepts. One TA remarked: ‘Cohorts last year posed questions about very simple concepts. This year, the interactions were on more advanced concepts. You could notice the influence of the Virtual Tutor’ (TA 2).
3.4.5. Perspectives on Future Use of AI in Nursing Education
The virtual tutor was considered to reflect the univeristy's MN programme's forward‐thinking approach that embraced innovation. There were suggestions to extend its use to other graduate nursing courses, as well as future expansion to clinical placement support, simulation of patient scenarios for nurse practitioner training, and integration across the MN programme to foster cross‐course concept application. Importantly, participants stressed the role of instructors in modelling appropriate use of AI tools and fostering ongoing reflection on how such tools may shape students’ critical thinking and learning. As one student reflected: ‘My professor…she actively encouraged us to use it in an ethical way…I found that learning environment to be not restrictive, and freeing… because it is true that AI chatbots are everywhere and we should know how to use them properly’ (Student 2).
3.5. Data Integration
Integration of virtual tutor interactions, surveys, and interviews highlighted that students mainly used the tutor for just‐in‐time learning support and valued its course focus but noted usability and scope limitations. Table 4 summarises student and TA perceptions of the tutor and its burgeoning influence on learning.
TABLE 4.
Integrated AI virtual tutor quantitative and qualitative findings.
| Study objective | Quantitative findings | Qualitative findings | Integrated finding |
|---|---|---|---|
| AI virtual tutor usage |
651 interactions logged (Jan–Jun 2025), peak use at 19:00 and around assignment deadlines Most interactions lower‐order (remembering/understanding), with higher‐order (applying/analysing) rising later |
Students used the tutor to clarify terminology, confirm understanding, and locate lecture content (‘by the end of the course, I used it…to clarify key terms for my papers’—Student 1) TAs consulted it during marking to resolve ambiguities |
Tutor functioned as a just‐in‐time support for students and TAs, reinforcing knowledge early and supporting independent higher‐order application |
| Perceived benefits |
91% had prior AI experience Comfort levels generally high Self‐efficacy high (median = 8/10) 64% recommended future use |
Students valued immediacy and course‐specific pre‐training (‘focused solely on this course and gave it the right context’—Student 1) TAs noted efficiency gains and more advanced assignment work |
The tutor was valued for relevance and accessibility, contributing to efficiency for TAs and confidence for students, even if not universally ‘useful’ |
| Perceived challenges |
Usability rated adequate 72% rated the tutor as only ‘a little’ or ‘not useful’ |
Students described interface issues, slow response times, scope limits, and privacy concerns TAs raised risk of over‐reliance (‘students may want to totally rely on what the tutor produces’—TA 2) |
Mixed perceptions reflect strengths in convenience and context but limitations in usability and trust |
| Influence on learning and engagement |
55% reported enhanced learning 27% reported greater engagement |
Students described reassurance and confidence in applying concepts (‘helpful when working late at night or weekends’—Student 1) TAs observed a shift towards more advanced questions compared to prior cohorts. |
Tutor primarily supported knowledge checking and consolidation, which enhanced TA‐rated assignment quality. |
| Future potential | 64% recommended continuing to offer |
Suggestions included expanding to other MN courses, clinical placements, and simulation training Instructors must play key role in modelling ethical AI use |
Support to build upon and extend use, with considerations around responsible use and usability improvements |
4. Discussion
This mixed‐methods evaluation demonstrates that a co‐designed, course‐specific AI virtual tutor is feasible to implement and actively used by graduate nursing students, with 651 logged interactions. Students valued the tutor's immediacy and close alignment to course‐specific concepts but identified problematic issues with its usability and scope limits and had some privacy hesitations. Overall, students and TAs identified improvements in learning confidence and conceptual consolidation and recommended continued refinement of the tutor for future course iterations.
Students mainly sought just‐in‐time clarification of core lower‐order learning concepts early in the term, with a later shift towards higher‐order learning usages including application and analysis. Combined with TA observations of improved student deep thinking with tutor use, this suggests students may use the tool to develop a foundation of knowledge on which they build the independent advanced application skills, required in their nursing careers (Chu et al. 2022). Application of course concepts to complex clinical, leadership, and policy problem is a cornerstone of graduate education. These findings align with emerging higher‐education research showing that LLM‐based tools may bolster learning efficiency and self‐directed study (Deng et al. 2025; Labadze et al. 2023; Peláez‐Sánchez et al. 2024; Schei et al. 2024; Wang and Fan 2025; Wu and Yu 2023). In nursing education specifically, improvements in student problem‐solving and attitudes towards AI co‐occur with the purposeful use of AI technology (Arkan et al. 2025; Harrington et al. 2025; Song et al. 2025). The present study extends this literature by showing how a course‐specific tutor, trained on course materials and embedded within a university LMS, can reduce the number of irrelevant information delivered to students to offer contextually appropriate guidance. This course‐specific solution therefore addresses many known concerns about AI hallucinations and curricular misalignment noted for general‐purpose chatbots. At the same time, our qualitative data underscore some risks of student over‐reliance on the tool and limited critical appraisal skill‐building. This reinforces the need for the use of pedagogical frameworks and assessment strategies that value the learning process as well as student outputs.
Based on these findings, practical implications for graduate nursing curricula include a need to explicitly embed teaching about appropriate and responsible AI use into courses and to model how to scrutinise, confirm, and appropriately cite AI outputs. Providing students with structured opportunities to develop high‐quality prompting strategies (O'Connor et al. 2024) and critically appraise LLM‐generated responses to queries can support them in using LLMs appropriately as opposed to as a shortcut that fosters over‐reliance (Blomquist et al. 2025). A second practical implication is the requirement to update AI virtual tutor interfaces and the resources on which they are trained to ensure they remain relevnt and reflect the evolving knowledge. In nursing education, this includes updating AI virtual tutors to reflect updates in entry‐to‐practice nursing informatics competencies (e.g., Chu et al. 2025; Canadian Association of Schools of Nursing 2025), and other new frameworks or professional standards. Training instructors and TAs on appropriate and ethical tutor use and application is also required to enhance their confidence in leveraging such tools to teach. These implications are consistent with institutional guidance, including that from our own university which states that curricula should incorporate responsible AI use, training in critical appraisal of AI outputs and prompting should be provided, AI resources ought to be refreshed regularly, and educators should be empowered and exposed to AI tools to strengthen confidence in AI‐enhanced teaching practices (University of Toronto 2025).
Importantly, the implications of AI virtual tutors extend beyond academic learning environments to clinical nursing practice. Graduate nursing education prepares clinicians to integrate evidence, apply clinical reasoning, and evaluate practice innovations within complex health systems. AI‐supported learning tools may help develop these competencies by supporting knowledge consolidation, structured reasoning, and evidence‐informed decision processes. Skills reinforced through interaction with AI tutors, including clarifying concepts and applying theoretical frameworks, mirror cognitive tasks required in clinical contexts. However, consistent with concerns raised in the present study, safe integration of AI into clinical contexts requires careful attention to transparency, verification of outputs, professional accountability, and critical thinking. As health systems increasingly incorporate AI‐enabled decision supports and digital clinical tools, familiarity with responsible AI use will increasingly represent an emerging competency for nurses (AACN 2011).
Having embedded our AI virtual tutor in a secure institutional LMS with authenticated access and a usage agreement mitigated privacy risks and promoted transparency of expectations with students. Still, students expressed some hesitancy with engaging with the tutor and it is possible that more education related to anonymity and institutional AI academic misconduct rules is required. Previous research has also suggested that factors including disclosing data handling practices and anonymising interactions can enhance trust in online learning modalities (Aliakbari et al. 2025; Luo et al. 2025). Our future efforts will expand our co‐designed approach to include students in developing the tutor onboarding programme, with the goal of creating materials that directly address student privacy concerns.
Nursing instructors must consider how to embed AI tutors into teaching in ways that preserve students' critical engagement with course materials—a pedagogical requirement that is particularly important in graduate nursing education (AACN 2011). Mechanisms that encourage critical thinking include LLM‐embedded citations to source passages, open discussions around critically challenging model outputs, and student meta‐reflection or a ‘thinking about thinking’ (Merkebu et al. 2024). Given the potential for over‐reliance on AI identified in our study, instructors should also design assignments that require process documentation, such as rationales, drafts, reflections, to assess and strengthen students' critical appraisal skills (Moorhouse et al. 2023).
A key strength of this study lies in the co‐design of the virtual tutor with nursing instructors and technologists, ensuring alignment with curricular goals and pedagogical values. Unlike general‐purpose chatbots, the AI virtual tutor was pre‐trained exclusively on course‐specific content and embedded in our institutional LMS, enhancing relevance and user privacy. The use of a mixed‐methods design, incorporating Bloom's taxonomy, provides a robust method to capture patterns of use and explore user perceptions.
Considering limitations, while all interaction content was analysed, the anonymised nature of these data prevented linking interactions to individual participants and precluded any description of patterns between student characteristics and tutor use. Survey and interview data were collected after course completion. This likely constrained our sample size and introduced a potential response bias. There is also the possibility that we did not achieve adequate information power within our qualitative interviews as the sample size was determined by the number of students and TAs willing to participate, though our results were robust and addressed our aims in an in‐depth way. Additional limitations are the study's single‐course context, which constrains generalisability of our findings, its descriptive design, which precludes causal inferences about learning outcomes, and that we did not conduct member‐checking to validate our qualitative findings and results.
Guided by these findings, our future research in this area would include: (1) re‐platform the tutor to a software with greater usability and streamlined LMS integration; (2) train the re‐platformed tutor on updated course materials and frameworks; (3) consider visual asset capabilities and scaffolded prompts to promote higher‐order cognition, and (4) conduct large‐scale evaluations using quasi‐experimental comparisons, where feasible. We also plan to examine equity impacts across learner subgroups, and develop tailored instructor education on how to introduce virtual tutors in an ethical and effective manner that supports active learning.
A co‐designed, course‐specific AI tutor can serve as a scalable, just‐in‐time learning support companion that consolidates foundational knowledge and supports progression to higher‐order learning tasks, particularly during high‐demand periods. In graduate nursing education where synthesis, judgment, and ethics are paramount, AI tutors are promising adjuncts to traditional learning methods. To translate these advantages into consistent learning gains, programmes should pair AI tutors with robust privacy, transparency, and ethics resources, usable interfaces, and scope that supports degree‐wide learning.
Funding
This project was supported through the University of Toronto Centre for Teaching Support and Innovation's Artificial Intelligence Virtual Tutor Initiative, which provided software and personnel support to build and evaluate the tutor.
Conflicts of Interest
The authors declare no conflicts of interest.
Appendix A. Completed STROBE and CONSORT Reporting Guidelines
| STROBE item no. | STROBE recommendation | Page no. (from paper) |
|---|---|---|
| Title and abstract | ||
| 1a | Indicate the study's design with a commonly used term in the title or abstract | Title & Abstract, p. 1 |
| 1b | Provide an informative and balanced summary of what was done and what was found | Abstract, p. 1 |
| Introduction | ||
| 2 | Explain the scientific background and rationale for the investigation being reported | Introduction, pp. 3–4 |
| 3 | State specific objectives, including any prespecified hypotheses | End of Introduction, p. 4 |
| Methods | ||
| 4 | Present key elements of study design early in the paper | Methods, p. 5 (‘mixed‐methods explanatory sequential design’) |
| 5 | Describe the setting, locations, and relevant dates, including recruitment and data collection periods | Methods 2.2, pp. 5–6 |
| 6a | Give eligibility criteria, and the sources and methods of selection of participants | Methods 2.2 (Setting and Participants), p. 5 |
| 7 | Clearly define all outcomes, exposures, predictors, potential confounders, and effect modifiers | Methods 2.4 (Data Collection), p. 6 where applicable |
| 8 | For each variable, give sources of data and methods of assessment | Methods 2.4 (Data Collection), pp. 6–7 |
| 9 | Describe efforts to address potential sources of bias | Ethics and Data Handling (p. 7); privacy and anonymisation described |
| 10 | Explain how the study size was arrived at | Methods 2.2 (Setting and Participants), p. 5 (‘Participants were approximately 120 Master of Nursing (MN) students enrolled’) and Methods 2.4 (Data Collection), p. 7 (‘All students were emailed the REDCap survey link and invited to participate in the survey’) |
| 11 | Explain how quantitative variables were handled in the analyses | Methods 2.5 (Data Analysis), p. 7 |
| 12a | Describe all statistical methods, including those used to control for confounding | Methods 2.5, p. 7 (descriptive only; no confounder control not applicable) |
| 12b | Describe any methods used to examine subgroups and interactions | Not applicable |
| 12c | Explain how missing data were addressed | Methods 2.4 (Data Collection), p. 7 (‘REDCap validation ensured no missing data’) |
| 12d | Describe analytical methods taking account of sampling strategy | Not applicable |
| 12e | Describe any sensitivity analyses | Not applicable |
| Results | ||
| 13a | Report numbers of individuals at each study stage | Results 3.3 (Surveys n = 11; Interviews n = 5), pp. 8–9 |
| 13b | Give reasons for non‐participation at each stage | Discussion (‘Survey and interview data were collected after course completion. This likely constrained our sample size’) |
| 13c | Consider use of a flow diagram | Not applicable |
| 14a | Give characteristics of study participants | Results 3.3 p. 8 and Table 2 |
| 14b | Indicate number of participants with missing data for each variable | Methods 2.4 (Data Collection), p. 7 (‘REDCap validation ensured no missing data’) |
| 15 | Report numbers of outcome events or summary measures | Tables 1 and 3 |
| 16a | Give unadjusted and adjusted estimates and their precision | Not applicable |
| 16b | Report category boundaries when continuous variables were categorised | Not applicable |
| 16c | Translate estimates of relative risk into absolute risk, if relevant | Not applicable |
| 17 | Report other analyses done (subgroups, sensitivity) | Not applicable |
| Discussion | ||
| 18 | Summarise key results with reference to objectives | Discussion, pp. 11–13 |
| 19 | Discuss study limitations and potential bias | Discussion, p. 13 |
| 20 | Give cautious interpretation of results considering objectives and evidence | Discussion, pp. 11–13 |
| 21 | Discuss the generalisability of study results | Discussion, p. 13 |
| Other information | ||
| 22 | Give the source of funding and the role of funders | Title Page, p. 1 |
| COREQ domain/Item | COREQ Description | Page no. (from paper) |
| Domain 1: Research team and reflexivity | ||
| 1. Interviewer/facilitator | Identifies who conducted the interviews or focus groups | p. 7 (‘interviews were conducted by a trained external research assistant with who was unknown to participants’) |
| 2. Credentials | Reports researcher credentials (e.g., PhD, RN) | p. 1 |
| 3. Occupation | Describes researcher's occupation at time of study | p. 1 |
| 4. Gender | Reports interviewer's gender | Not reported |
| 5. Experience and training | Describes interviewer experience or training | p. 7 (‘conducted by a trained external research assistant’) |
| 6. Relationship established | Notes any prior relationship with participants | p. 7 (‘interviews were conducted by a trained external research assistant with who was unknown to participants’) |
| 7. Participant knowledge of interviewer | Describes what participants knew about interviewer | p. 7 (‘interviews were conducted by a trained external research assistant with who was unknown to participants’) |
| 8. Interviewer characteristics | Notes interviewer bias, assumptions, or interests | p. 7 (‘interviews were conducted by a trained external research assistant’) |
| Domain 2: Study design | ||
| 9. Methodological orientation/theory | Identifies methodological framework (e.g., grounded theory, phenomenology, content analysis) | p. 5 (‘mixed‐methods explanatory sequential design’) |
| 10. Sampling | Describes sampling method (e.g., purposive, convenience) | p. 5 (‘all students and TAs who agreed to participate’) |
| 11. Method of approach | Describes how participants were approached | p. 5 (‘recruited via email’) |
| 12. Sample size | Reports number of participants | p. 9 (‘interviews n = 5’) |
| 13. Non‐participation | Reports refusals or dropouts | p.13 (‘Survey and interview data were collected after course completion. This likely constrained our sample size’) |
| 14. Setting of data collection | Describes interview location | p. 7 (‘interviews conducted virtually’) |
| 15. Presence of non‐participants | Notes who else was present during interviews | p. 7 (‘Individual interviews were conducted’) |
| 16. Description of sample | Describes key sample characteristics | Table 2 |
| 17. Interview guide | Mentions use of interview guide | p. 7 (‘instructor‐developed interview guides explored in‐depth views…’) |
| 18. Repeat interviews | Reports whether repeat interviews were carried out | p. 7 (‘once’) |
| 19. Audio/visual recording | Reports recording methods | p. 7 (‘audio‐recorded and transcribed verbatim’) |
| 20. Field notes | Indicates if field notes were made | p. 7 (‘Field notes were taken’) |
| 21. Duration | Reports average interview length | p. 9 (‘~30 min‐long’) |
| 22. Data saturation | Describes data saturation or stopping criteria | Not applicable |
| 23. Transcripts returned | Notes if transcripts were returned to participants for feedback | p. 13 (‘we did not conduct member‐checking to validate our qualitative findings and results’) |
| Domain 3: Analysis and findings | ||
| 24. Number of data coders | Reports number of analysts/coders | p. 7 (‘Two researchers (C.H.C. and L.A.J.) independently coded data’) |
| 25. Description of coding tree | Describes themes or coding structure | p. 7 (‘transcripts were analysed thematically with reference to the study objectives’) |
| 26. Derivation of themes | Explains how themes were derived | p. 7 (‘transcripts were analysed thematically with reference to the study objectives’) |
| 27. Software | Reports software used | p. 7 (‘Excel’) |
| 28. Participant checking | Describes whether participants checked findings | p. 13 (‘we did not conduct member‐checking to validate our qualitative findings and results’) |
| 29. Quotations presented | Provides participant quotations to illustrate themes | pp. 8–11 (quotes embedded in Results) |
| 30. Data and findings consistent | Confirms consistency between data and findings | p. 13 (Discussion references triangulation and coherence) |
| 31. Clarity of major themes | Major themes clearly presented | pp. 8–11 (Results section) |
| 32. Clarity of minor themes | Describes diverse cases or minor themes | pp. 8–11 (Results section) |
Data Availability Statement
The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.
References
- Abdelwahab, S. I. , Taha M. M. E., Farasani A., et al. 2025. “Artificial Intelligence in Nursing Education: A Bibliometric Analysis of Trends, Challenges, and Future Directions.” Teaching and Learning in Nursing 20, no. 2: e356–e367. 10.1016/j.teln.2024.11.018. [DOI] [Google Scholar]
- Abdulai, A. F. , and Hung L.. 2023. “Will ChatGPT Undermine Ethical Values in Nursing Education, Research, and Practice?” Nursing Inquiry 30, no. 3: e12556. [DOI] [PubMed] [Google Scholar]
- Aliakbari, M. , Barzan P., and Maadikhah M. M.. 2025. “Privacy Issues in Online Learning: A Review of Literature and Suggestions for Further Research.” Language, Identity and the Digital Realm 1, no. 1: 1–17. [Google Scholar]
- Alrazeeni, D. M. , Alharrasi M., Khan Rony M. K., et al. 2026. “Transforming Nursing Education With Artificial Intelligence: A Systematic Review (2010–2025).” SAGE Open Nursing 12: 23779608261424597. 10.1177/23779608261424597. [DOI] [PMC free article] [PubMed] [Google Scholar]
- American Association of Colleges of Nursing . 2011. The Essentials of Master's Education in Nursing. https://www.aacnnursing.org/Portals/0/PDFs/Publications/MastersEssentials11.pdf. [Google Scholar]
- American Association of Colleges of Nursing . 2025. 2025 Thought Leaders Assembly: Examining the Potential of AI to Transform Nursing Education [Summary Report]. https://www.aacnnursing.org/Portals/0/PDFs/Reports/Thought‐Leadership/AACN‐2025‐Thought‐Leaders‐Assembly‐Summary.pdf?utm_source=chatgpt.com. [Google Scholar]
- Anderson, L. W. , Krathwohl D. R., Airasian P. W., et al. 2001. A Taxonomy for Learning, Teaching, and Assessing: A Revision of Bloom's Taxonomy of Educational Objectives. Longman. [Google Scholar]
- Arkan, B. , Dallı Ö. E., and Varol B.. 2025. “The Impact of ChatGPT Training in the Nursing Process on Nursing Students' Problem‐Solving Skills, Attitudes Towards Artificial Intelligence, Competency, and Satisfaction Levels: Single‐Blind Randomized Controlled Study.” Nurse Education Today 152: 106765. [DOI] [PubMed] [Google Scholar]
- Athilingam, P. , and He H. G.. 2024. “ChatGPT in Nursing Education: Opportunities and Challenges.” Teaching and Learning in Nursing 19, no. 1: 97–101. [Google Scholar]
- Blomquist, J. , Llewellyn S., Alderden J., and Connor K.. 2025. “Empowering Faculty to Incorporate Large Language Models in Nursing Education Using a Delegation Framework.” Nursing Education Perspectives 46, no. 2: 126–128. [DOI] [PubMed] [Google Scholar]
- Booth, R. G. , Strudwick G., McBride S., O'Connor S., and Solano López A. L.. 2021. “How the Nursing Profession Should Adapt for a Digital Future.” BMJ 373: n1190. [Google Scholar]
- Canadian Association of the Schools of Nursing . 2025. “Nursing Informatics Entry‐to‐Practice Competencies for Registered Nurses (Second Edition).” https://www.casn.ca/Assets/CASN/Documents/Resources/Informatics‐Entry‐to‐Practice‐Competencies_2025_EN.pdf.
- Canadian Nurses Association . 2019. Advanced Practice Nursing: A Pan‐Canadian Framework. https://hl‐prod‐ca‐oc‐download.s3‐ca‐central‐1.amazonaws.com/CNA/2f975e7e‐4a40‐45ca‐863c‐5ebf0a138d5e/UploadedImages/documents/nursing/Advanced_Practice_Nursing_framework_e.pdf. [Google Scholar]
- Carolus, A. , Koch M. J., Straka S., Latoschick M. E., and Wienrich C.. 2023. “MAILS—Meta AI Literacy Scale: Development and Testing of an AI Literacy Questionnaire Based on Well‐Founded Competency Models and Psychological Change‐ and Meta‐Competencies.” Computers in Human Behavior: Artificial Humans 1, no. 2: 100014. [Google Scholar]
- Chu, C. H. , Conway A., Jibb L., and Ronquillo C. E.. 2022. “The Impact of Digital Technologies, Data Analytics and AI on Nursing Informatics: The New Skills and Knowledge Nurses Need for the 21st Century.” In Nursing and Informatics for the 21st Century—Embracing a Digital World, edited by Saba V. K. and McCormick K. A., 3rd ed., 22–32. Productivity Press. [Google Scholar]
- Chu, C. H. , Siganevich Y., Donato‐Woodger S., et al. 2025. “Teaching Informatics Competencies Within Nursing Education: A Scoping Review of Teaching Strategies.” Applied Clinical Informatics 16, no. 04: 911–929. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Deng, R. , Jiang M., Yu X., Lu Y., and Liu S.. 2025. “Does ChatGPT Enhance Student Learning? A Systematic Review and Meta‐Analysis of Experimental Studies.” Computers & Education 227: 1–35. [Google Scholar]
- Eke, D. O. 2023. “ChatGPT and the Rise of Generative AI: Threat to Academic Integrity?” Journal of Responsible Technology 13: 100060. [Google Scholar]
- Harrington, J. , Booth R. G., and Jackson K. T.. 2025. “Large Language Models in Nursing Education: Concept Analysis.” JMIR Nursing 8: e77948. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Hemment, D. , Kommers C., Ahnert R., et al. 2025. Doing AI Differently: Rethinking the Foundations of AI via the Humanities. Alan Turing Institute. 10.5281/zenodo.1642129. [DOI] [Google Scholar]
- Hobensack, M. , von Gerich H., Vyas P., et al. 2024. “A Rapid Review on Current and Potential Uses of Large Language Models in Nursing.” International Journal of Nursing Studies 154: 104753. [DOI] [PubMed] [Google Scholar]
- Hwang, Y. , Knobf M. T., and Sadler L. S.. 2025. “Integration in Mixed‐Methods Research With an Exemplar Explanatory Sequential Study.” Nursing Research 74, no. 2: 144–149. [DOI] [PubMed] [Google Scholar]
- Kilfoy, A. , Hsu T. C., Stockton‐Powdrell C., Whelan P., Chu C. H., and Jibb L.. 2024. “An Umbrella Review on How Digital Health Intervention Co‐Design Is Conducted and Described.” Npj Digital Medicine 7, no. 1: 374. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Labadze, L. , Grigolia M., and Machaidze L.. 2023. “Role of AI Chatbots in Education: Systematic Literature Review.” International Journal of Educational Technology in Higher Education 20: 56. [Google Scholar]
- Lawrence, C. E. , Dunkel L., McEver M., et al. 2020. “A REDCap‐Based Model for Electronic Consent (eConsent): Moving Toward a More Personalized Consent.” Journal of Clinical and Translational Science 4, no. 4: 345–353. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Luo, J. , Zheng C., Yin J., and Teo H. H.. 2025. “Design and Assessment of AI‐Based Learning Tools in Higher Education: A Systematic Review.” International Journal of Educational Technology in Higher Education 22: 42. [Google Scholar]
- Merkebu, J. , Veen M., Hosseini S., and Varpio L.. 2024. “The Case for Metacognitive Reflection: A Theory Integrative Review With Implications for Medical Education.” Advances in Health Sciences Education: Theory and Practice 29, no. 4: 1481–1500. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Mojjada, H. , Chand N. G., Priyanka A. L., and Sravya T.. 2024. “The Evolution of AI Virtual Tutors in Modern Higher Education.” International Journal of Innovative Research in Technology 11, no. 7: 2430–2435. [Google Scholar]
- Moorhouse, B. L. , Yeo M. A., and Wan Y.. 2023. “Generative AI Tools and Assessment: Guidelines of the World's Top‐Ranking Universities.” Computers and Education Open 5, no. 15: 100090. [Google Scholar]
- O'Connor, S. , Peltonen L. M., Topaz M., et al. 2024. “Prompt Engineering When Using Generative AI in Nursing Education.” Nurse Education in Practice 74: 103825. [DOI] [PubMed] [Google Scholar]
- Peláez‐Sánchez, C. , Camaqui D. V., and Glasserman‐Morales L. D.. 2024. “The Impact of Large Language Models on Higher Education: Exploring the Connection Between AI and Education 4.0.” Frontiers in Education 9: 1392091. [Google Scholar]
- Ronquillo, C. E. , Peltonen L. M., Pruinelli L., et al. 2021. “Artificial Intelligence in Nursing: Priorities and Opportunities From an International Invitational Think‐Tank of the Nursing and Artificial Intelligence Leadership Collaborative.” Journal of Advanced Nursing 77, no. 9: 3707–3717. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Ruiz‐Rojas, L. I. , Salvador‐Ullauri L., and Acosta‐Vargas P.. 2024. “Collaborative Working and Critical Thinking: Adoption of Generative Artificial Intelligence Tools in Higher Education.” Sustainability 16, no. 13: 5367. [Google Scholar]
- Schei, O. M. , Møgelvang A., and Ludvigsen K.. 2024. “Perceptions and Use of AI Chatbots Among Students in Higher Education: A Scoping Review of Empirical Studies.” Education Sciences 14, no. 8: 922. [Google Scholar]
- Sharifi Kelarijani, A. , Safdari A., and Golitaleb M.. 2024. “Every Coin Has Two Sides: ChatGPT Poses a Potential Threat to Nursing Students' Education.” Frontiers in Medicine 11: 1415067. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Song, D. , Zhang P., Zhu Y., et al. 2025. “Effects of Generative Artificial Intelligence on Higher‐Order Thinking Skills in Nursing Education.” Nurse Education in Practice 88: 104549. [DOI] [PubMed] [Google Scholar]
- Srinivasan, M. , Venugopal A., Venkatesan L., and Kumar R.. 2024. “Navigating the Pedagogical Landscape: Exploring the Implications of AI and Chatbots in Nursing Education.” JMIR Nursing 7: e52105. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Sun, G. H. , and Hoelscher S. H.. 2023. “The ChatGPT Storm and What Faculty Can Do.” Nurse Educator 48, no. 3: 119–124. [DOI] [PubMed] [Google Scholar]
- Tillmanns, T. , Salomão Filho A., Rudra S., et al. 2025. “Mapping Tomorrow's Teaching and Learning Spaces: A Systematic Review on GenAI in Higher Education.” Trends in Higher Education 4, no. 1: 2. 10.3390/higheredu4010002. [DOI] [Google Scholar]
- Tong, A. , Sainsbury P., and Craig J.. 2007. “Consolidated Criteria for Reporting Qualitative Research (COREQ): A 32‐Item Checklist for Interviews and Focus Groups.” International Journal for Quality in Health Care 19, no. 6: 349–357. [DOI] [PubMed] [Google Scholar]
- Topaz, M. , Peltonen L. M., Michalowski M., et al. 2025. “The ChatGPT Effect: Nursing Education and Generative Artificial Intelligence.” Journal of Nursing Education 64, no. 6: e40–e43. [DOI] [PubMed] [Google Scholar]
- University of Toronto . 2025. Toward an AI‐Ready University. https://ai.utoronto.ca. [Google Scholar]
- von Elm, E. , Altman D. G., Egger M., et al. 2007. “The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) Statement: Guidelines for Reporting Observational Studies.” Lancet 370, no. 9596: 1453–1457. [DOI] [PubMed] [Google Scholar]
- Wang, J. , and Fan W.. 2025. “The Effect of ChatGPT on Students' Learning Performance, Learning Perception, and Higher‐Order Thinking: Insights From a Meta‐Analysis.” Humanities and Social Sciences Communications 12, no. 1: 621. [Google Scholar]
- Wu, R. , and Yu Z.. 2023. “Do AI Chatbots Improve Students Learning Outcomes? Evidence From a Meta‐Analysis.” British Journal of Educational Technology 55, no. 1: 10–33. [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Data Availability Statement
The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.