Recommendations for Training Faculty in Generative AI Use: Crafting Higher‐Order Application Exercises in Team‐Based Learning

ABSTRACT

Application exercises are the most critical components of the sequenced steps of team‐based learning (TBL) that should be designed as complex, real‐world problems to elicit deep learning and foster student engagement. The process of crafting these exercises presents several challenges for educators—time constraints, alignment with learning objectives, scaffolding and authentic simulation of real‐world problem solving. The use of artificial intelligence (AI) cuts down time and effort for design and allows focus on refinement and error correction. In addition, it allows use of an iterative process to generate higher order application exercises that provide the context and competency level required for students to make learning gains.

We are sharing our knowledge on using AI to design application exercises (AEs) garnered from our international workshops on TBL. We empowered educators to write effective prompts using the Task–Role–Audience–Create–Intent (TRACI) framework to create engaging AEs that meet the 4S principles, i.e., significant problem, same problem, specific choice, simultaneous reporting, of TBL AEs, aligned to higher order Bloom's taxonomy objectives for critical thinking and learner engagement. We then demonstrated how we used Anthropic's Claude 4 Sonnet AI tool to design the exercises using an iterative process. The workshop concluded with a reflection exercise and feedback to promote intentionality.

Our collective experience facilitating this workshop in interdisciplinary and interprofessional settings, serving as TBL facilitators at our respective institutions and informal participant feedback at the workshops informed the development of this series of best practice recommendations for faculty development. This foundational yet critical skill is necessary in the rapidly evolving landscape of AI skill‐building competencies for health professions educators. We describe our best practice recommendations here within the paradigm of course design for TBL.

1. Introduction

The explosion of artificial intelligence (AI) in education means health profession educators have had to develop adaptive expertise in a novel field at a breathtaking pace. This presents a tremendous opportunity to innovate in the service of deeper learning. Educators who have honed their craft without AI for years now face the challenge of developing a new knowledge base in technology and being expected to incorporate it into their daily practice. This creates frustration and is a complex problem. It therefore calls for a nuanced approach to faculty development efforts in use of educational technologies [1].

While the revised Bloom's taxonomy remains critical to pedagogy, educators are reframing it to signal the pedagogic shift necessary to facilitate critical thinking in learners [2]. Effective application exercises in TBL are traditionally geared toward higher order thinking skills levels of analyse, evaluate and create. Our worked example demonstrates this in Supplementary File S1. Clark proposes shifting from ‘analyse’ to ‘compare’ and ‘validate’, ‘evaluate’ to ‘challenge and reflect’, ‘create’ to “co‐create’ and introduces a new level ‘transform’ to indicate preparation of learners for real‐world impact [3]. Applying this new paradigm shift to TBL means that it becomes incumbent on educators to design effective application exercises that not only assess knowledge acquisition but equip learners to examine authenticity of information, apply ethical reasoning, co‐create responsibly and drive real‐world change.

The value of TBL in interprofessional education (IPE) is evidenced by the study conducted by Black et al. (2016) [4]. Their study involved learners from 10 healthcare professions. When the team included one or more exceptional team members, the team performed higher on the assessment. Considering that interprofessional collaboration requires an understanding of the unique contributions of each field, this study demonstrates that TBL can foster teamwork within an IPE program. Viewed through the lens of IPE, application exercises must be deliberately constructed to draw out discipline‐specific expertise. Herein, the use of AI tools for activity design, guided by trained faculty, is well‐positioned to support at scale. Burgess and McGregor in their systematic review identify several challenges to the implementation of TBL in IPE settings, two of which include the design of patient cases that suit multiple disciplines and alignment of topics within the curricula of multiple disciplines [5]. A faculty development workshop focused on using AI to generate and refine application exercises could directly address this barrier.

In their narrative review of TBL in graduate medical education, Poppelman et al. report that the time investment required to develop effective TBL curricula is considerable [6]. This underscores the need for systematic faculty support in leveraging AI for crafting application exercises that move beyond simple knowledge recall toward clinical reasoning and problem‐solving. The combined flipped classroom and TBL approach demonstrated by Shuai et al. reports that when clinical internship students in orthopaedics were guided through well‐structured in‐class problem exchanges, they achieved significantly higher scores in clinical application domains such as imaging analysis, fracture classification and complication management [7]. Herein, faculty development focused on designing such exercises using AI could meaningfully enhance the clinical teaching competencies that traditional lecture‐based formats fail to develop.

We designed a faculty development workshop that focused on crafting higher order application exercises for TBL using AI. It was delivered in a TBL format (see Figure 1) to a group of international medical educators at the TBLC annual conference in California in March 2025.

FIGURE 1.

Main phases of team‐based learning (TBL).

Our intended learning outcomes for the workshop were as follows:

After participating in this workshop, learners will be able to:

1. Recognize the key elements of an effective prompt based on the TRACI framework

2. Create an effective prompt on an AI large language model (LLM) for generation of an application exercise that promotes higher‐order thinking skills.

3. Evaluate what elements make certain prompts more effective in generating higher‐order thinking exercises.

The design of this workshop (see Figure 2) was underpinned by the analysis‐design‐development‐implementation‐evaluation (ADDIE) model and team‐based learning (TBL) to promote active learning, collaboration and experiential learning [8, 9].

FIGURE 2.

Schematic illustration of the workshop format.

Based on our own experience, we propose the following recommendations for implementation.

Tip #1. Offer Pre‐Work as a Priming Tool to Ensure Readiness to Participate

Applying the ‘analyse’ phase of the ADDIE model, we recognized the need for participating faculty to develop baseline knowledge for effective prompting using a framework that would deliver the desired output [8]. Because this requires participants to iterate prompts using trial and error, we provided guided examples and instructions for signing up for a free account on a LLM of their choice. While there are several examples of prompt frameworks, we chose TRACI for its simplicity. We provided a concise pre‐work document (see Supplementary File S2) to serve as a scaffold, which was shared with all registered participants prior to coming to the workshop. This served a dual purpose: more time for hands‐on activities during the workshop and ensuring all faculty are provided with a level playing field for participation. This aligns with the constructivist learning theory, which espouses that learning occurs through the interaction of prior existing knowledge of the learner, the thinking process of the individual learner and the type of learning activity [10].

The pre‐work document was focused on TRACI—a prompt framework that provides critical context to guide the LLM in producing more accurate and relevant responses [11]. TRACI is an acronym for Task–Role–Audience–Create–Intent. A task is defined as the general activity the LLM is being asked to perform, role is the identity to be adopted while producing the output, audience is the recipient of the output, create is the format of the response and intent is the objective of the output generated.

Tip #2. Design the Workshop Using Evidence‐Based Best Practices for Effective Learning

Continuing with the ‘design’ phase of the ADDIE model, we defined the learning outcomes (as stated above) and determined that the TBL format would be most amenable to meeting those outcomes [9]. Since faculty will be designing for TBL, it makes sense to immerse them in the activity using the same format.

The TBL activity was delivered on InteDashboard, an online TBL instruction platform [12]. This aligns with evidence from literature for medical students, which recommends using technology as a scaffold to support learning, without being weighed down by the tool per se, keeping the focus on learning outcomes, and being reminded that the tool exists to enhance design of more effective learning experiences [13]. We began the workshop with a brief overview of the TRACI framework that included a think‐pair‐share activity, as we were cognizant that time constraints may have prevented participants from reading the pre‐work material (see Supplementary File S1).

The TBL format was a sound approach because it ensured all participants were at the same level throughout the individual readiness assurance test (IRAT) phase and team readiness assurance test (TRAT) phase. These phases triggered prior knowledge of the TRACI framework, higher order cognitive processing tasks per Bloom's taxonomy and the ‘4S’ principles of effective application exercises, i.e., significant problem, same problem, specific choice, simultaneous reporting (see Supplementary File S1). The single‐best response items for the IRAT and TRAT were identical, first taken individually followed by teamwork using the Intedashboard platform, which did not have the inbuilt AI function. Test items were written by the authors at the recall–understand–apply level of Bloom's revised taxonomy. Any ambiguous concepts were clarified by the facilitators during the debrief.

To create effective application exercises aligned with Bloom's higher order cognitive processing tasks, participants were tasked with designing an integrated application exercise for second year undergraduate medical students in a cardiovascular module. They were required to demonstrate use of the TRACI framework, iterate on the prompts at least once and document changes and rationale for these. The conception of this type of task aligns with another evidence‐based recommendation by Lidolf and Pasco, melding the faculty‐as‐learner and faculty‐as‐designer approach while developing education technology related professional development programs for faculty [14]. Because faculty would be involved in designing TBL activities at their home institutions, the focus of this workshop was to equip them with the skills to design application exercises that involve higher order thinking skills using AI. Thus, it made perfect sense to immerse faculty in this design task as part of the workshop. In the faculty‐as‐learner paradigm, we acknowledge that self‐directed learning and providing participants with the opportunity to explore and problem‐solve is essential rather than mere transmission of knowledge. Participants were allowed to use any AI LLM of their choice to ensure equity and meet individual institutional requirements and in support of the fact that technology proficiency mediates learners' interaction with AI. The transition from the use of the technologic tool to practical learning is dependent on learners' comfort with the tool [15]. No information was systematically collected regarding specific LLMs used. However, the share‐out following the application activity involved a comparison of the output from various LLMs. Most participants used free versions of ChatGPT 3.5 and Copilot.

Delivery of this workshop in a TBL format was based on the ‘modelling’ premise, providing an authentic context for faculty [16]. Working collaboratively in small groups and allowing them to experience as learners the kind of understanding that can be acquired through effective use of AI is a very relevant example of the learning by design approach [17]. Not only were faculty learning to use effective prompts but they were simultaneously enhancing their pedagogic understanding of how to align higher order Bloom's taxonomy objectives—analyse, evaluate, create with the 4S principle of TBL application exercises. This effective approach of amalgamating a technologic tool and learning activity design with pedagogical awareness, elevates participants from being mere consumers of technology to designers who can harness technology for specific goals.

This effective approach of amalgamating a technologic tool and learning activity design with pedagogical awareness elevates participants from being mere consumers of technology to designers who can harness technology for specific goals.

Thus, the three phases of the TBL‐format workshop aligned with the ‘develop’ and ‘implement’ phases of the ADDIE model. Tips 3 and 4 are aligned with the ‘evaluate’ phase of the ADDIE model to see if the workshop achieved the intended learning outcomes.

Tip #3. Facilitate Engagement in the Learning Process Through Opportunities to Provide and Receive Feedback

The application exercise was followed by an e‐gallery walk with each group receiving and providing feedback on their prompts and generated output (application exercise), to consolidate learning. Pre‐determined criteria for feedback were as follows:

• effectiveness of prompts in generating suitable exercises

• alignment of exercises with higher order thinking skills

• relevance and engagement factor of exercises for learners.

Continuing in the spirit of modelling, the facilitators demonstrated their own prompt, the generated application exercise and the revised iteration of the prompt. This demonstration of a worked example (see Supplementary File S1) provides another source of immediate feedback for further deepening participants' understanding of what elements make certain prompts more effective in generating higher order thinking exercises: a stated learning outcome of the workshop.

Tip #4. Provide Opportunities to Reflect on the Process of Learning That Has Occurred

The final phase of the workshop entailed a structured large group facilitated reflection exercise that asked participants to critically review the following:

• What elements made certain prompts more effective in generating higher‐order thinking exercises?

• What challenges did you encounter when crafting prompts for AI, and how did you overcome them?

• How might you apply these prompt‐crafting skills to other areas of your course design and teaching?

This aligns with the third essential element of active learning—critical reflection, aside from intentional engagement and purposeful observation [18]. Viewed through the lens of Schon's theory of reflective practice, this phase of the workshop allowed participants to practice’ reflection on action’‐ defined as consciously thinking about an experience after it has occurred and deciding to do things differently or in a similar pattern in the future [19]. This reflective practice may not always occur spontaneously and therefore faculty developers should create a space for intentional reflection during such professional development activities. This needs to be guided by facilitators by asking the right prompts that promote critical reflection of learning that just occurred. Participants appreciated the use of a framework serving as a guardrail for generating desired output. It was recognized that the level of detail specified during prompting heavily influenced the kind of output generated by the LLM. Engaging in iterative prompting and being intentional about the level of Bloom's taxonomy that the application exercise was targeted to were cited as ways to overcome challenges. Participants outlined additional applications of prompting such as designing multiple choice question assessments, lesson plans, problem‐based learning exercises, grading rubrics for patient notes among others.

Finally, we solicited feedback from participants by administering a brief questionnaire. This helped not only gauge faculty reactions but also improve future iterations of the workshop.

2. Key Takeaways

The key takeaways for health profession educators interested in designing faculty development efforts in this area are summarized in Table 1.

TABLE 1.

Key takeaways.

	Best practice recommendations	Benefit to participants	Benefit to facilitators
1	Provide pre‐work ahead of the workshop	Develop baseline knowledge of topic	Frees up time for hands‐on activities during the workshop
2	Design the workshop in TBL format	Active learning approach ensures engagement	IRAT and TRAT phases even out the learning ground so everyone can participate fully in the application phase where the ‘learning by doing’ occurs
3	Harness learning technologies while training faculty in the use of technology	Facilitates ease of use Can view the work of other teams	Ensures smooth flow of tasks during the workshop as well as debrief
4	Allocate time for teams to provide and receive feedback while specifying pre‐determined criteria	Opportunity to give and receive feedback improves learning	Setting pre‐determined criteria signals the ‘intended outcome’
5	Provide worked examples after the teams have completed the task	Solidifies learning	Opportunity to showcase what the ‘ideal’ outcome looks like by modelling it
6	Create reflection discussions at the end	Facilitates critical thinking about actions	The pause creates room for eliciting lessons learned and new insights

3. Conclusion

This toolbox offers suggestions in facilitating the design of continuing professional development opportunities for AI skill building for health profession educators in an effective manner. Viewed through the lens of the International Advisory Committee for Artificial Intelligence (IACAI) recommendations for integrating AI for health profession educators, the format of our workshop aligns with domains II (AI foundation skills), V (AI for instruction and academic tasks) and VI (AI to enhance clinical skills and training) [20]. This has implications that extend into other areas of health professions education—assessments, integration of clinical and basic sciences, curriculum design, lesson planning, to name a few.

We used the design of higher‐order application exercises in TBL as a template for how health profession educators can be effectively trained to integrate AI into course design. Higher‐order application exercises should balance the tightrope walk between complexity level and cognitive load without overwhelming learners. Therefore, educators should themselves be willing to experiment iteratively with prompting and intentionally reflect on the appropriateness of the application exercises generated. AI‐generated content should always be verified by the subject matter expert to retain the ‘human in the loop’ in the collaborative interaction between AI and humans.

Author Contributions

Deborah Dalmeida: conceptualization, visualization, writing – original draft, writing – review and editing. Nahla Gomaa: conceptualization, supervision, writing – review and editing. Elizabeth Prabhakar: conceptualization, supervision, writing – review and editing.

Funding

The authors have nothing to report.

Ethics Statement

This manuscript has been written as the following article type: The Clinical Teacher's toolbox. It does not contain research data and does not require access to a data repository. Two appendices have been provided as separate downloadable files. The authors agree that by submitting a manuscript to the Clinical Teacher, our name, email address and affiliation, and other contact details the publication might require will be used for the regular operations of the publication. All authors have agreed to the final submitted version. No generative AI was used in the creation of this manuscript and all sources whose work has been referenced have been cited appropriately in the manuscript.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Data S1: IRAT_TRAT_application exercise_worked example.

Data S2: Pre‐workshop reading.

Dalmeida D., Gomaa N., and Prabhakar E., “Recommendations for Training Faculty in Generative AI Use: Crafting Higher‐Order Application Exercises in Team‐Based Learning,” The Clinical Teacher 23, no. 5 (2026): e70489, 10.1111/tct.70489.

Data Availability Statement

The data that supports the findings of this study are available in the Supporting Information of this article.