Artificial Intelligence Integration in Radiation Therapy Education: A Multi‐Modal Approach

ABSTRACT

Introduction

Critical workforce shortages in radiation oncology have led tertiary institutions to rapidly expand their radiation therapy (RT) student cohorts. The increase in students entering university created the challenge of scaling educational delivery while preserving the quantity and quality of learning essential for developing well‐prepared, competent clinicians. Artificial Intelligence (AI) applications have been recognised as valuable tools in medical education; however, limited research has addressed their use in RT education. This study aimed to evaluate three AI educational innovations used in the RT degree at the University of Newcastle.

Methods

A cross‐sectional study design was implemented to investigate the perceptions of students toward the integration of three AI educational innovations, which were embedded in the RT program curriculum across Years 1, 2 and 3. The three innovations included: (1) delivery innovation (AI video lectures), (2) assessment innovation (AI‐assisted assessment feedback), and (3) content innovation (AI‐simulated communication tasks). Descriptive statistics were calculated for quantitative survey responses. Open‐ended responses were analysed to find recurring themes.

Results

A total of 62 students participated in the study across three cohorts (Years 1 (n = 33), 2 (n = 13) and 3 (n = 16)), with a mean age of 20.6 years. In Year 1, n = 24 (73%) of students reported being ‘satisfied’ or ‘very satisfied’ with the AI video format. Among Year 2 students, n = 7 (54%) wanted AI feedback on future assessments, while n = 6 (46%) were unsure or opposed to future AI feedback. In Year 3, n = 12 (75%) felt ‘more’ or ‘much more comfortable’ practising with the AI patient than with peers.

Conclusion

This research revealed varied outcomes related to AI innovations among year groups. The integration of AI was perceived positively; however, participants favoured AI tools for formative learning over those used for assessments.

Three artificial intelligence (AI) innovations were implemented in a radiation therapy degree to address a 130% enrolment increase. Students (n = 62) evaluated AI video lectures, assessment feedback, and communication simulations. Students strongly preferred AI for practice‐based learning over assessment, with communication simulations receiving 100% recommendation.

1. Introduction

Globally, the radiation therapy (RT) workforce faces critical shortages. In Australia, demand projections for radiation therapists indicate significant gaps between the supply of practitioners needed and the number of qualified practitioners [1]. These shortages increase the workload in radiation oncology departments, resulting in slower patient throughput. This causes longer wait lists for cancer treatments, with each month of delay raising the patient's risk of death by approximately 10% [2]. Tertiary institutions have responded by working to increase student enrolments where possible, producing more graduates to fill vacancies urgently. The University of Newcastle's (UON) RT degree exemplifies this trend, expanding from typical cohorts of approximately 50 students annually (before 2025) to 115 first‐year enrolments in 2025, representing a 130% increase.

The Medical Radiation Practice Board of Australia (MRPBA) outlines key capabilities that registered medical radiation practitioners must demonstrate, including the application of anatomical, physiological, and pathological knowledge; effective use of clinical information systems; and the ability to deliver safe, patient‐centred care across various imaging and treatment procedures [3]. To ensure RT students are trained in accordance with the MRPBA capability requirements, the medical radiation science—RT degree undergoes a thorough accreditation process every 3 years [4]. The rapid cohort expansion presents multifaceted challenges in delivering the UON 4‐year Honours degree. These include maintaining individualised attention, providing authentic professional skill practice, delivering timely formative feedback, and managing increasingly complex curricula reflecting advancing treatment technologies. These challenges have necessitated immediate and agile responses in transforming teaching through innovative approaches [5, 6].

Artificial Intelligence (AI) is the development of computer systems that can perform tasks typically requiring human intelligence [7]. AI systems are continually emerging, becoming part of everyday life, from ‘smart voice’ assistance applications (such as Siri or Alexa) to robotics (such as self‐deploying vacuum cleaners) [8, 9]. RT is a health profession that utilises AI for various clinical tasks, including image registration, organ/target guidance contouring, dose optimisation, and quality assurance [10]. As a result, although AI is considered progressive in many other professions, RT academics and students are well‐versed in utilising AI and its features within the clinical RT workflow.

As discussed by Alenezi et al. integrating digital methods into higher education (through various delivery approaches) can result in improved student outcomes. This can include ‘adaptive, blended, personalised and virtual learning environments’ [11]. Furthermore, a systematic review of teachers' views on technology integration identified advantages, including heightened learner engagement, universal access to resources, and positive correlations with academic success. Conversely, barriers included insufficient institutional infrastructure and gaps in educator technology training [12].

In recent years, AI applications in medical education have gained prominence, with emerging evidence supporting their use in various contexts. According to a scoping review conducted by Nagi et al., AI applications have been studied in multiple health education disciplines, leading to better patient results by providing healthcare professionals with enhanced capabilities and understanding [13]. Additionally, in a systematic review of the impact of ChatGPT and AI chatbots on higher education institutions by Dempere et al., various benefits are reported. These include personalised feedback to students, automated grading assistance, and the generation of course materials and practice questions. However, the authors' key recommendation is that while AI integration in higher education is inevitable and beneficial, institutions must perform a strategic, well‐regulated implementation that balances innovation while mitigating risks [14]. This is further supported by Issa et al. who recommend that standardised AI training programs be designed and taught by AI experts collaborating with healthcare educators to effectively prepare students for AI‐enhanced clinical practice [15].

Despite AI becoming a prominent tool in health education and being adopted in the discipline of RT, limited research addresses AI implementation specifically within RT education or its application across multiple educational functions simultaneously. Therefore, this study evaluates the implementation and effectiveness of three distinct AI‐enhanced educational innovations within the UON, RT degree. The educational innovations included the delivery of lectures, the provision of feedback and practical content delivery, which are explained in more detail throughout this manuscript. The innovations were designed to address scalability challenges while preserving educational quality and human educator presence.

2. Methods

2.1. Setting

Three AI educational innovations were implemented within the RT program at UON, College of Health, Medicine, and Wellbeing, School of Health Sciences, as part of the Radiation Oncology Collaborative Network, Australia.

2.2. Participants

To be eligible for participation, students were to be enrolled in either Year 1, Year 2, or Year 3 of the RT degree at the time of the study and must have completed the relevant AI‐enhanced educational activities as part of their regular class curriculum. A participation information statement, alongside a link to the survey, was provided to students via an announcement on the UON learning management system (Canvas: https://canvas.newcastle.edu.au). Consent was implied upon completing the survey. Each survey took approximately 5 min to complete.

2.3. Artificial Intelligence Educational Innovations

The following AI educational innovations were developed and implemented:

1. Delivery Innovation—traditional lecture delivery (Year 1): Traditional lecture content was transformed using AI‐assisted scriptwriting (Claude AI: Claude Sonnet 4.5, Anthropic) combined with text‐to‐video technology (Pictory AI) [16, 17]. Previous lecture materials were converted into video formats with accompanying AI‐generated formative assessment in the form of retention quiz questions.

2. Assessment Innovation—provision of feedback (Year 2): Claude AI was prompted to deliver automated feedback, providing detailed, rubric‐aligned feedback on individually designed research questions, with academic oversight maintaining quality standards. Comments included detailed population, intervention, comparator, outcome ‘PICO’ analysis, keyword suggestions, or multiple refined question options [16].

3. Content Innovation—clinical communication skills (Year 3): Students interacted with Claude AI configured through specific prompts to simulate various patient scenarios (anxious first‐time patients, elderly patients with comprehension difficulties, reluctant young adults). Three scenario prompts were made available via Canvas as PDF documents, with students required to complete a minimum of one scenario in their own time as a mandatory formative task. Each prompt provided students with instructions to copy/paste into Claude AI, which configured the AI to simulate a specific patient persona and communication challenge. Students engaged in text‐based, asynchronous conversations with the AI patient, typing their responses and receiving typed patient replies. Upon completing their interaction, students used a second prompt to trigger AI‐generated feedback analysing their communication techniques. Students were encouraged to submit a copy of their conversation transcript, though submission was voluntary. The task was not graded, serving as formative practice for clinical communication skills [16].

The selection of AI tools for these innovations was determined through an iterative evaluation process that prioritised quality, efficiency, and accuracy. Both the Claude AI and Pictory AI tools were tested against alternative platforms during the development phase, with the final selection based on output quality, ease of integration into existing curricula, and reliability in producing content suitable for the medical radiation sciences context [16, 17]. Detailed development workflows for each innovation, including time investments, quality control procedures, and technical specifications, are provided as Supporting Information.

2.4. Data Collection and Analysis

Data were collected via online surveys administered through QuestionPro (www.questionpro.com) to students following the completion of each AI‐enhanced educational activity. Surveys were designed specifically for each intervention to assess relevant outcomes. All surveys included demographic questions (age, gender, and prior relevant experience) and comparative questions that asked students to rate the AI‐enhanced approach against traditional methods they had previously experienced.

Descriptive statistics (frequencies, percentages, means, and standard deviations) were calculated for quantitative survey responses. Open‐ended responses were reviewed to identify common themes.

As each approach addressed different educational functions with distinct survey instruments, comparative analyses focus on conceptually similar metrics (e.g., recommendation likelihood, perceived helpfulness) to illustrate patterns across innovations rather than to establish statistical superiority of one innovation over another. Table 1 provides an overview of the topics covered in each survey, along with the corresponding number of questions for each topic.

TABLE 1.

Survey overview (Years 1, 2 and 3).

Section/topic	Survey 1: Year 1 AI‐enhanced videos	Survey 2: Year 2 AI assessment feedback	Survey 3: Year 3 AI communication simulation
Demographics	3 questions Year level Age Gender	3 questions Year level Age Gender	3 questions Year level Age Gender
Prior experience	1 question Previous online learning experience	1 question Previous experience with automated feedback	1 question Previous patient communication training
Usage/participation	2 questions Which AI videos watched Previous traditional lecture attendance	1 question Speed of feedback received	1 question Most valuable patient scenario
Comparative Evaluation	3 questions Enjoyability vs. traditional Ease of following vs. traditional Impact on learning vs. traditional	2 questions Detail vs. previous feedback Quality vs. staff feedback	2 questions Comfort vs. peer role‐play Comparison to other communication training
Quality assessment	2 questions Helpfulness of quiz questions Overall satisfaction	4 questions Helpfulness for improvement Clarity of comments Alignment with rubric Understanding mistakes	3 questions Realism of scenarios Helpfulness of feedback Preparation for real interactions
Confidence/learning impact	N/A	N/A	2 questions Confidence before simulation Confidence after simulation
Advantages	1 question (multiple select) Own pace, rewatch, timing, visuals, quizzes	N/A	1 question (multiple select) No judgement, practice, feedback, scenarios, availability
Disadvantages/limitations	1 question (multiple select) Less interactive, technical issues, missed interaction, unclear	N/A	1 question (multiple select) Not real, inappropriate responses, technical issues, no non‐verbal cues
Future recommendation	1 question Recommend continuing approach	2 questions Overall satisfaction Like AI feedback on future assignments	1 question Recommend for future students
Open‐ended questions	4 questions What was liked most What to improve Additional comments Overall final comment	6 questions Most useful aspect Aspects to improve Timing comparison Impact of timing Concerns about AI vs. human Overall final comment	3 questions Most valuable aspect Improvements needed Overall final comment
Total	17 questions	19 questions	17 questions

3. Results

3.1. Participants Characteristics

A total of 62 students across three‐year levels participated in the study (Year 1: n = 33; Year 2: n = 13; Year 3: n = 16). Response rates by year included: Year 1: 29%, Year 2: 33% and Year 3: 50%. Demographic characteristics are presented in Table 2.

TABLE 2.

Participant characteristics across three AI educational innovations.

Characteristic	Year 1 (n = 33)	Year 2 (n = 13)	Year 3 (n = 16)
Age, mean (SD)	20.5 (2.8)	20.4 (1.7)	21.0 (0.9)
Age range, years	18–29	19–25	20–23
Female, n (%)	27 (82)	10 (77)	14 (87)
Male, n (%)	6 (18)	3 (23)	2 (12)
Prior relevant experience	Moderate/extensive online learning: 79%	Prior automated feedback: 14%	Moderate/extensive communication training: 94%

The cohorts were comparable in terms of age and gender distribution. Mean ages ranged from 20.5 to 21.0 years across the three cohorts. Most participants across all cohorts were female (n = 51).

3.2. Delivery Innovation: AI‐Enhanced Video Lectures (Year 1)

3.2.1. Student Engagement and Satisfaction

Among the 33 Year 1 students surveyed, the majority (92%) had prior experience with traditional lectures. Student satisfaction with AI‐generated video lectures was high, with 73% reporting being satisfied or very satisfied. Most students found the videos more enjoyable (73%) and easier to follow (71%) than traditional lectures and rated the integrated quiz questions as ‘very’ to ‘extremely’ helpful (64%).

3.2.2. Perceived Learning Impact, Advantages and Disadvantages

When asked about the impact on their learning, 73% reported that the video format had a better or significantly better impact compared to traditional lectures. As shown in Figure 1, 73% of students would recommend continuing this approach for future cohorts.

FIGURE 1.

Student perceptions of AI educational innovations across three metrics: Recommendation likelihood, perceived helpfulness, and comparison to previous methods (Years 1, 2 and 3). While survey instruments varied for each Year level, comparable metrics were selected to illustrate general patterns.

Students identified several key advantages of the AI‐generated videos. The most frequently cited advantages were the ability to watch at one's own pace (26% of all responses), the ability to rewatch sections (20%), more convenient timing (21%), and integrated quizzes for retention (19%). The main disadvantages identified were less interactive than live lectures (32% of disadvantage responses), missing face‐to‐face interaction (24%), and unclear content (14%). Several students specifically noted accessibility concerns, with three students commenting on the lack of closed captions for students with hearing difficulties or those who benefit from visual text support.

3.3. Assessment Innovation: AI‐Assisted Feedback (Year 2)

3.3.1. Feedback Delivery, Quality and Timeliness

Feedback was delivered promptly, with most students receiving it within 1 week of submission. The combined AI generation and academic oversight process required only 40 min for the entire cohort, with the timing of feedback release determined by course scheduling rather than marking completion time. The majority found the AI‐generated feedback detailed (64%) and clear (73%), though approximately one‐quarter reported concerns about detail or clarity.

3.3.2. Perceived Helpfulness and Learning Impact

Students had mixed perceptions of the AI feedback's value. While 63% found it helpful for improving their research question, responses regarding future use were divided: 54% wanted AI feedback on future assignments, while 46% were unsure or opposed (Figure 2).

FIGURE 2.

Student satisfaction levels with AI‐enhanced video lectures (Year 1) and AI‐assisted assessment feedback (Year 2). While survey instruments varied for each Year level, comparable metrics were selected to illustrate general patterns. Year 3 data are not included as the survey instrument used different satisfaction metrics (recommendation likelihood and comfort levels) that were not directly comparable to the satisfaction scales used in Years 1 and 2.

3.3.3. Student Concerns

Open‐ended responses revealed several concerns about AI‐generated feedback. Multiple students emphasised the importance of human oversight, with one noting: “I feel like things may be missed and the feedback may not be as accurate as it should be.” (Y2P1) Several students specifically mentioned preferring AI‐assisted marking (where AI helps the human marker) rather than fully automated feedback. One student expressed concern about the perceived inconsistency: “We're constantly warned that using AI in our assignments could lead to serious consequences… yet the university is using AI tools to give us feedback.” (Y2P2) Students also requested more specific, actionable feedback and noted that some AI‐generated language “did not seem natural.”

3.4. Content Innovation: AI Communication Simulation (Year 3)

3.4.1. Realism and Comfort Level

Of the 16 Year 3 students, most (94%) had moderate to extensive prior communication training. The majority (81%) reported the AI patient scenarios ‘realistic’ or ‘very realistic’. Compared to traditional peer role‐playing, 75% felt more comfortable practising with the AI patient.

3.4.2. Feedback Quality and Helpfulness

The immediate AI feedback on communication techniques was highly valued, with 81% (n = 13) rating it as ‘very’ or ‘extremely helpful’. This immediate feedback was consistently highlighted in open responses, with one student noting: “The most valuable aspect was definitely the immediate feedback and having to think on your feet when the ‘patient’ replied with something you didn't expect.” (Y3P1).

3.4.3. Key Advantages and Limitations

The main advantages identified by students were immediate feedback (23%), ability to practice multiple times (20%), available anytime (20%), no judgement from peers or staff (19%), and variety of scenarios (17%). One student captured the accessibility benefit: “Non‐judgmental environment, easy to do, can practice anytime.” (Y3P2) Another emphasised the learning value: “You don't know what they could respond with, just like with humans—you don't know what that person will say to you.” (Y3P3) The primary limitations identified were not like real human interaction (30%), lack of non‐verbal cues (26%), and AI responses sometimes inappropriate (15%). Notably, 11% (n = 3) of students indicated no limitations (‘none of the above’). One student noted a unique concern: “Have too much time to facilitate answer, which isn't like real life” (Y3P4).

3.4.4. Student Recommendations and Satisfaction

All students (100%) indicated they would ‘probably’ or ‘definitely’ recommend this AI communication training for future students. When comparing this experience to other communication training they had received, 50% (n = 8) rated it as ‘better’ or ‘much better’, 37% rated it as ‘the same’, and only 6% rated it as ‘worse’ (Figure 2).

3.5. Educator Efficiency Gains Across Innovations

The three AI innovations demonstrated distinct efficiency advantages in addressing cohort scalability challenges. Year 1 video lectures eliminated recurring educator time investment, with the 24 h per semester previously required for live lecture delivery (2 h × 12 weeks) reduced to zero once videos were created. Videos remained accessible for subsequent cohorts regardless of enrolment size. Student engagement time was also reduced, with traditional 2‐h lectures condensed to an average of 20 min of video content (90% reduction).

Year 2 AI‐assisted feedback reduced marking time from 6.7 h to 40 min per assignment cycle, representing a 90% time reduction. This 40‐min timeframe included both AI generation and mandatory academic review to maintain quality standards, with feedback subsequently released to students within one week.

Year 3 communication simulations enabled unlimited student practice without additional institutional resources. Unlike traditional peer role‐play limited by timetabling, staffing, and physical space constraints, AI simulations allowed students to practice clinical communication scenarios repeatedly in their own time, addressing the scalability challenge of providing adequate skills practice to growing cohorts. While these innovations required substantial upfront time investment for development and pilot testing (detailed in Supporting Information), the recurring time savings and scalability benefits justify the initial resource allocation, particularly given ongoing cohort expansion.

3.6. Comparative Analysis Across Innovations

Figure 1 presents a comparison of satisfaction and recommendation rates across the three AI educational innovations. The Year 3 communication simulation demonstrated the highest recommendation rate (100%), followed by Year 1 video lectures (76%) and Year 2 AI feedback (38%).

4. Discussion

The three AI educational innovations demonstrated varying levels of student acceptance and perceived effectiveness. The AI communication simulation showed the strongest student endorsement (100% recommendation rate), likely due to its provision of safe, judgement‐free practice with immediate feedback. The AI‐enhanced video lectures received a positive response (73% satisfaction and recommendation), particularly for the flexibility and self‐paced learning. The AI‐assisted assessment feedback showed more mixed results (54% would recommend), with students expressing concerns about accuracy, consistency, and the appropriateness of AI in summative assessment contexts. These findings suggest that AI applications may be most successful when they enhance practice and formative learning rather than when they replace human judgement in assessment.

4.1. Educational Impact

The educational impact of our AI innovations demonstrates similarities with Singh et al. review of AI's impact on higher education teaching and learning [18]. The notable findings that AI improves instructional efficiency, enables personalisation, and enhances student learning were confirmed across our AI innovations. Particularly in the 73% reporting improved learning impact (Year 1) and 81% finding AI feedback extremely helpful (Year 3). In line with this, a longitudinal study of AI integration and its impact on learning in higher education by Hardini et al. report that AI‐enabled personalised learning systems (‘machine learning algorithms adapt instructional materials according to a student's progress’) are strongly associated with improved academic performance [19].

However, our multi‐modal approach reveals that educational impact is not uniform across AI applications. The range of acceptance rates between the years (Year 3: 100% recommendation; Year 1: 73%; Year 2: 54%) suggests that AI's educational effectiveness is based on its pedagogical purpose. Applications supporting practice and formative learning achieved the outcomes Singh et al. describe, while applications involving summative assessment encountered resistance and quality concerns [18]. This suggests that increasing AI's educational impact requires careful alignment between the AI capabilities and pedagogical purposes, rather than broad implementation across all educational functions.

While the educational impact findings demonstrate AI's potential to enhance learning outcomes, our results suggest this impact is linked to how well human elements are preserved in the educational process. The differences in acceptance across the AI innovations (formative applications achieving higher satisfaction than summative) suggest that students do not evaluate AI exclusively on technical capabilities or learning improvements. Students considered the importance of keeping human educators involved in the learning process.

4.2. Preservation of Human Presence

A study of a conceptual ethical framework to preserve natural human presence in the use of AI systems in education, by Isop et al. reports that AI in education must preserve human presence by maintaining clear role distinctions, with AI operating in supporting roles rather than replacing human educators [20]. It is suggested that this includes role clarity and synchronisation between humans and AI. Furthermore, Roe et al. establish that 282 students across three Asian universities did not support AI‐generated feedback, though combined AI and instructor feedback was better accepted [21].

The clear acceptance of Year 3's communication simulation (100%) could be due to it being a formative learning task (rather than a summative assessment), aligning with Roe et al. finding that students and staff were more comfortable with AI systems for knowledge checking and participation monitoring than for assessment marking, suggesting greater acceptance of AI in formative contexts [21]. Year 2's mixed responses (54% would recommend) validate Roe et al. finding that students resist AI‐generated feedback in assessment contexts. Students' requests for AI‐assisted rather than fully automated feedback support this.

Advancing technology and AI implementation are revolutionising healthcare delivery. However, developing and maintaining face‐to‐face communication skills remains essential in all aspects of patient care. The therapeutic relationship that develops between a patient and clinician is dependent on empathy, trust, and understanding. These qualities emerge through genuine human connection. A narrative review on guiding radiation therapists in health literacy by Kelly et al. emphasised that radiation therapist communication is crucial for providing person‐centred care, noting that effective information exchange either facilitates patient empowerment or creates barriers to understanding [22]. Research demonstrates that healthcare professionals' communication styles directly impact patient outcomes, either facilitating or creating barriers to information exchange and patient empowerment [23]. Importantly, radiation therapists rely on observing non‐verbal cues (such as facial expressions, body language, and emotional responses) to assess patient understanding and identify those who may require additional support. These observations enable practitioners to adjust their communication approach in real‐time, employing techniques such as ‘chunk and check’ or ‘teach‐back’ methods that require interactive dialogue [22, 24].

As the complexity of healthcare technology has increased, clinicians are at risk of losing the interpersonal skills vital to daily practice [25]. The use of AI technologies in healthcare education to enhance, rather than replace, skill development has the potential to deliver more focused training for practising simulated patient encounters tailored to the challenges faced in a clinical setting [26]. However, the disadvantage of these simulated experiences, as noted by the Year 3 responses, is the lack of authenticity in these interactions. Although AI‐generated materials can supplement learning, the interpersonal skills essential to patient care (such as reading non‐verbal cues and adapting communication in real time) are best developed through direct, face‐to‐face clinical experiences.

4.3. Equity and Accessibility

The implementation of AI in educational settings presents a complex equity contradiction; while these technologies can enhance accessibility for some learners, they simultaneously create barriers for others.

The Year 1 video lectures, despite achieving high satisfaction (73%), inadvertently excluded students with hearing difficulties. Three students noted the absence of closed captions. Morris et al. determined that 99% of students reported captions as helpful for clarification, comprehension, and note‐taking [27]. The 10% of students reporting a lack of closed captions in our study reflects the AI equity divide. Furthermore, Dumitru et al. found that AI‐generated captions often misinterpret, creating accessibility barriers, and therefore, students with disabilities face a particular risk from AI bias [28].

Conversely, the Year 3 communication simulation (100% recommendation) actively reduced barriers by fostering a judgement‐free environment. The opportunity to practice patient‐based communication without the pressure of a peer or educator presence was well received. This is in line with Zhang et al. who found that AI‐based interventions showed significant positive effects for students with disabilities, and various AI tools are used to engage in learning tasks to enhance learning outcomes [29].

4.4. Scalability and Sustainability

This study addressed a critical institutional challenge: a 130% increase in enrolment in the RT program, expanding from approximately 50 to 115 first‐year students. The three AI innovations demonstrated substantial efficiency gains while preserving educational quality. Year 1 video lectures eliminated 24 h per semester of live delivery while reducing student engagement time by 90%. Year 2 AI‐assisted feedback reduced marking time by 90% (6.7 h to 40 min). Year 3 simulations enabled unlimited practice without resource constraints.

However, efficiency gains did not translate uniformly into student acceptance (Year 3: 100%; Year 1: 73%; Year 2: 54%). Liang et al. report that AI effectively shares instructors' tutoring tasks and reduces workload while providing immediate feedback [30]. The differential acceptance suggests pedagogical purpose matters more than efficiency alone; formative applications achieved higher satisfaction than summative assessment, with students expressing concerns about accuracy and emphasising the need for human oversight in assessment contexts.

The reduction in educator time enables redirection toward individualised consultation, curriculum development, and practical supervision, activities that cannot be scaled through automation. Strielkowski et al. highlight that long‐term sustainability requires regular updates to maintain alignment with evolving educational standards and technological advancements, highlighting the need for ongoing investment alongside immediate efficiency gains [31]. Our findings suggest AI can successfully address scalability when implementation prioritises formative learning, maintains human oversight, and complements rather than replaces educator expertise.

4.5. Efficiency Gains and Educational Trade‐Offs

The reported efficiency gains warrant examination of their impact on educational quality. The 90% reduction in face‐to‐face time during Year 1 lectures resulted from removing both redundant material and extended discussions, condensing 2‐h lectures into focused 20‐min videos with integrated quizzes. All students passed the final examination covering this material, suggesting learning outcomes were preserved. While this eliminates spontaneous questions during delivery, students gain reduced cognitive fatigue and increased time for self‐directed study, which is particularly valuable when managing clinical placements alongside coursework [11, 32].

Year 2 research question marking time reduction (6.7 h to 40 min for 39 students) reflects efficiency in feedback delivery: traditional marking averaged 10 min per student, while AI‐assisted marking required approximately 1 min per student, including mandatory academic oversight to ensure quality and accuracy [5].

Reclaimed educator time has been reinvested into individualised student consultations, curriculum development, clinical supervision for expanding cohorts, and research mentorship, high‐value activities requiring human expertise that AI cannot replicate [18]. Future evaluation should track academic performance, clinical placement outcomes, and graduate success to ensure these innovations maintain educational quality while addressing scalability challenges.

4.6. Future Directions

This study demonstrates that students perceive AI‐integrated education innovations as valuable, particularly for formative‐based learning tasks. In view of the effective implementation of AI innovations for specific year groups, future applications could include all year groups utilising the video lectures and communication simulations, scaffolded to the abilities of each year. Furthermore, as the innovations are not specifically RT‐centric, they could be adopted by various disciplines of health education.

4.7. Limitations

The study's small cohort size limits the generalisability of results. Additionally, using different survey instruments across innovations restricts direct statistical comparisons; however, conceptually similar measurements facilitated the identification of descriptive patterns. Lastly, the survey design relied on self‐reporting, which could produce social desirability bias; participants may select responses they think would be more acceptable rather than selecting responses based on their true opinions.

5. Conclusion

This study revealed mixed results between year groups and AI innovations. Overall, the integration of AI was perceived as beneficial; however, the Year 2 AI‐assisted assessment feedback revealed concerns regarding AI accuracy. The results have highlighted that formative learning tasks are more accepted by students than assessment‐based AI innovations. In addition, participants suggested areas for improvement, including the addition of closed captions on video lectures. As a preliminary implementation of AI in the RT‐education space at UON, the results suggest AI is a satisfactory way to assist the teaching of growing cohort sizes when used in formative learning settings.

Disclosure

AI Use Declaration: During the preparation of this work, the author(s) used Claude AI to assist in the copy editing and proofreading. After using this tool/service, the author(s) reviewed and edited the content as needed and take full responsibility for the content of the published article.

Ethics Statement

The study was approved by the UON Human Ethics Advisory Panel (approval number: H‐2025‐0263).

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Data S1: jmrs70073‐sup‐0001‐Supinfo.pdf.

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.