The purpose of this study is to explore how generative artificial intelligence (GenAI) can support instructors in designing pedagogically complex teaching materials, specifically customizable case studies – a largely underexplored area in current educational research, which has predominantly focused on student-facing applications.
Adopting a design-based research approach, this study implemented a mixed-methods exploratory case design grounded in the Technology Acceptance Model and Cognitive Load Theory. Data sources included system usage analytics, service interaction logs and qualitative analysis of AI-generated instructional content.
Instructors demonstrated increasing engagement with the GenAI-powered tool, suggesting positive perceptions of its usefulness and ease of use. The tool also helped reduce instructors’ cognitive load by automating case structure and aligning outputs with teaching objectives. Emergent themes highlighted efficiency, customization and pedagogical integration.
Several limitations must be acknowledged. First, while usage data is rich in behavioral detail, it lacks user sentiment and rationale, which could be captured through future interviews or surveys. Second, this study focused on early-stage adoption, and longitudinal data would be valuable in evaluating sustained usage and integration. Third, the pilot involved a self-selected sample of early adopters; generalizing to broader populations requires replication in diverse institutional settings.
Future research should explore hybrid data collection models that combine log data with instructor reflections and student outcome measures. Comparative studies of different GenAI tools, as well as domain-specific adaptations (e.g. for health or engineering education), would also provide valuable insights. Additionally, as GenAI tools evolve to include multimodal and multilingual capabilities, future work should examine how these affect cognitive load and usability.
This study shifts the focus from student-facing AI tools to instructor-centered applications, offering novel insights into how GenAI can serve as a co-design partner in curriculum development. This study contributes to theory by applying Technology Acceptance Model and Cognitive Load Theory to instructor-facing GenAI use in higher education and highlights key ethical and institutional considerations for responsibly scaling AI-driven instructional design.
1. Introduction
The rapid rise of GenAI is transforming instructional design in higher education, offering new opportunities to automate content creation, personalize learning and improve efficiency. While existing literature largely emphasizes student-facing applications, such as AI-supported writing tools, proctoring systems or tutoring bots, less attention has been paid to instructor-centered uses of GenAI, particularly in the development of complex instructional materials (Zawacki-Richter et al., 2019; Holmes et al., 2019).
This gap is especially relevant in disciplines like business, health care and engineering, where instructors frequently rely on narrative-driven case studies to teach critical thinking and decision-making. Developing such materials is cognitively demanding and time-intensive, requiring realism, contextualization and alignment with learning outcomes.
This study explores the use of a custom-built GenAI-powered tool designed to assist instructors in generating discipline-specific case studies. Rather than replacing educators, the tool supports their creative and instructional processes by automating structural components, offering narrative templates and enabling rapid prototyping.
The analysis is guided by two complementary frameworks: the Technology Acceptance Model (Davis, 1989), which explores instructors’ perceptions of GenAI in terms of usefulness and ease of use, and Cognitive Load Theory (CLT) (Sweller, 2011), which assesses the extent to which the tool reduces the cognitive burden associated with complex case design.
Using a design-based research (DBR) approach within a mixed-methods exploratory intervention framework, this study draws on system usage data, service-level analytics and content analysis of AI-generated cases. It investigates instructor interaction patterns, preferences for case elements and the alignment between tool outputs and pedagogical goals.
Findings have implications for both educational researchers and institutional leaders seeking to integrate GenAI responsibly. As higher education faces growing demands for innovation and scalability, instructor-focused GenAI tools offer a promising, yet underexplored, avenue. By examining real-world usage in an instructional context, this study contributes to a practical and theoretically grounded understanding of GenAI’s emerging role in course design.
2. Literature review
2.1 Introduction
The rapid evolution of generative artificial intelligence (GenAI), particularly large language models (LLMs) like ChatGPT, has introduced new avenues for instructional design. While much existing research emphasizes student-facing applications (AI tutoring, essay assistance and feedback systems), there is growing interest in how GenAI supports instructors in course preparation. In particular, the use of LLMs to co-design customized case studies presents a promising avenue for reducing instructor workload and enhancing content relevance (McDonald et al., 2025).
Instructors often face barriers in developing case-based materials, including time limitations and challenges generating domain-specific scenarios. GenAI can function as a creative collaborator, assisting with structural formatting, narrative scaffolding and rapid content prototyping. This evolving role for AI prompts a shift in pedagogical agency, where instructors curate and refine AI-generated outputs rather than build cases from scratch. Such a shift warrants theoretical exploration and empirical scrutiny, particularly in disciplines like business and engineering where applied learning is crucial.
Despite GenAI’s increasing integration in education, the literature remains largely focused on learners. Recent discourse has begun to position GenAI tools as instruments for democratizing innovation in higher education, making it possible for instructors with limited time or institutional resources to still develop sophisticated teaching materials (Crowther and Hamdan, 2024). This study addresses the underexplored role of GenAI in faculty workflows, specifically in case-based pedagogy. It is grounded in two complementary frameworks: Technology Acceptance Model (TAM) and CLT, which together inform how GenAI tools are perceived, adopted and cognitively processed by educators.
2.2 Technology acceptance model and instructor adoption of GenAI tools
TAM offers a foundational framework to understand how individuals, especially instructors, adopt emerging technologies in educational settings. According to TAM, technology acceptance is primarily influenced by two constructs: perceived usefulness (PU) and perceived ease of use (PEOU) (Davis, 1989). These determine the user’s attitude toward using, intention to use and, ultimately, actual system use. These constructs have been consistently validated in studies examining technology adoption across education and professional domains (Granić and Marangunić, 2019).
In the context of AI-assisted instructional design, such as case study generation, PU is reflected in whether instructors believe GenAI tools improve teaching effectiveness or instructional productivity. For instance, AI-powered case generators can reduce hours of manual effort by producing complex, domain-relevant and engaging materials. Studies show that tools offering instructional alignment, context specificity and pedagogical relevance significantly enhance PU (Mastour et al., 2025; Barz et al., 2024).
PEOU is equally crucial. For GenAI systems to be adopted by instructors with varying levels of digital fluency, interfaces must be intuitive and outputs predictable. In higher education, PEOU has been shown to be a significant predictor of behavioral intention to adopt e-learning systems, especially when systems minimize friction and cognitive barriers (Panicker, 2020). With GenAI platforms, ease of prompt construction, ability to revise drafts and integration with familiar learning management systems environments all influence instructor confidence and willingness to experiment.
Recent studies suggest expanding TAM to include constructs like technological self-efficacy, social influence and perceived risk, particularly when ethical concerns or content accuracy are involved (McDonald et al., 2025; Kajiwara and Kawabata, 2024; Ibrahim et al., 2025). Instructors, especially in disciplines requiring factual accuracy and sensitivity, may hesitate to use LLMs unless the system’s transparency, output validation and institutional policy support are clear. Mastour et al. (2025) found that institutional culture, including training support and peer modeling, significantly mediates TAM constructs, especially among late adopters.
In the AI context, trust and perceived control are emerging as pivotal to TAM extensions. As explored by Ibrahim et al. (2025), when instructors feel they retain authorship and critical oversight over GenAI-generated cases, they are more willing to adopt the technology as a co-creative partner rather than a black-box solution.
Ultimately, assuming that GenAI tools enhance instructional outcomes, reduce planning time and maintain pedagogical integrity, instructors are more likely to integrate them into their practice. These dynamics reaffirm the model’s enduring value while highlighting the need to adapt TAM frameworks for emerging, generative technologies in education.
2.3 Cognitive load theory and case development tasks
While TAM addresses motivational and behavioral factors, CLT provides a powerful lens to analyze the mental demands instructors face when engaging with GenAI tools for instructional design, particularly in the cognitively intensive task of developing rich, pedagogically sound case studies. According to Sweller (2011), cognitive load comprises three types:
intrinsic load (task complexity);
extraneous load (inefficiencies in task execution); and
germane load (constructive mental effort for schema development).
Case design is inherently high in intrinsic load: instructors must balance realism with learning goals, manage complex narrative structures and tailor content to specific disciplines and cohorts. GenAI offers support by scaffolding this complexity, producing initial drafts, simulating stakeholder perspectives and embedding realistic dilemmas. In doing so, GenAI mitigates the intrinsic load associated with starting from a blank slate (Sweller, 2020).
Critically, GenAI platforms can reduce extraneous load by automating non-pedagogical tasks such as formatting, basic stakeholder role definition or contextual descriptions. As Ouwehand et al. (2025) observed, when interfaces are intuitive and structured, instructors spend less time navigating the system and more time engaging with pedagogical refinement. Similarly, in clinical training contexts, Tabatabaee et al. (2024) showed that reducing extraneous load enhances instructional engagement and design quality.
Perhaps most importantly, GenAI enhances germane load by promoting reflective practice. When instructors critically revise AI-generated content, deciding what to keep, what to refine and what to discard, they engage in a cognitively enriching process of schema construction. This aligns with research by Si (2024), who found that instructor adaptation of AI-suggested content promotes deeper expertise and improves long-term instructional capability.
GenAI tools that allow layered engagement, such as toggling complexity, editing narratives or introducing alternate decision points, further support schema development. They empower instructors to build more dynamic cases over time, moving beyond one-size-fits-all formats. Importantly, when instructors see AI as a collaborator in cognitive effort rather than a replacement for expertise, they are more likely to sustain meaningful engagement with these tools.
Thus, CLT reinforces the idea that GenAI systems must be designed for not only ease of use but also cognitive empowerment. Instructors are not just users; they are designers, analysts and evaluators. AI systems that reduce unnecessary load while fostering deeper instructional insight stand to significantly transform pedagogical practice. By offloading preliminary drafting and structural organization to GenAI, instructors can manage their overall instructional workload more sustainably (Crowther and Hamdan, 2024).
2.4 Artificial intelligence-enabled pedagogical co-design and entrepreneurship
In traditional course design models, instructors are responsible for developing educational content, including the particularly labor-intensive process of crafting case studies. The rise of GenAI tools marks a shift toward pedagogical co-design, where AI systems suggest drafts, scenarios and content variations, and educators exercise critical judgment to refine them. This collaborative dynamic resonates strongly with entrepreneurial practices, which are characterized by iterative prototyping, rapid experimentation and the strategic use of emerging technologies to create value.
In fields where case-based teaching is central, such as business, health care and engineering, GenAI facilitates adaptive instructional content development. Instructors can simulate different stakeholder perspectives, real-time dilemmas or contextual shifts, allowing for more customized and immersive learning scenarios. As Giabbanelli (2023) illustrates, LLM-based systems support dynamic simulation design, empowering educators to manipulate case parameters and explore multiple narrative arcs. This iterative experimentation reflects the entrepreneurial process of refining a product through successive feedback loops and opportunity recognition.
The integration of GenAI into educational practice is not merely an enhancement of instructional efficiency; it represents a form of academic entrepreneurship. Educators who experiment with these tools are engaging in a type of intrapreneurship, driving innovation from within established institutions by leveraging new technologies for pedagogical advancement. Kotturi et al. (2024), for example, explore how human-centered design approaches can scaffold GenAI use among entrepreneurs and educators, demonstrating the reciprocal learning that occurs when digital tools and teaching goals converge. This entrepreneurial lens allows us to reframe the role of instructors as not simply content deliverers but also innovators who architect new pathways for engagement and learning.
Moreover, GenAI has broader implications for education as an industry. As Mollick (2024) points out, the integration of generative AI into entrepreneurial practice is reshaping workflows, ideation processes and even the way business education itself is structured. AI’s capacity to generate structured content, simulate real-world complexity and personalize learning journeys positions it as a disruptive force, one that requires educators to adopt entrepreneurial mindsets to remain responsive to changing educational demands (Siegel and Wright, 2015). Similarly, Wang (2025) argues that when GenAI tools are embedded in business classrooms, they scaffold not just student creativity but also faculty innovation, effectively transforming the instructional landscape.
Case-based co-design also fosters reflexive practice. Exposure to AI-generated narrative patterns or alternative decision structures can encourage instructors to question their assumptions about what constitutes an effective or engaging case. In this way, GenAI tools do not just generate content but also act as intellectual provocateurs, challenging traditional pedagogical models and prompting educators to reimagine their teaching strategies. Zhu and Luo (2025) highlight how AI-powered scaffold systems in entrepreneurship education can support this transformation by guiding instructors and learners alike through complex, iterative planning and design processes, much like startup incubation models.
In summary, GenAI-supported case customization represents more than a technological upgrade, it embodies a shift toward an entrepreneurial approach to education. By embracing the co-design of instructional materials, educators are participating in innovation ecosystems that mirror those found in entrepreneurial ventures. This evolution has the potential to reshape both how instructors engage with curriculum development and how institutions position themselves in a rapidly transforming educational marketplace.
2.5 Ethical and institutional dimensions
The integration of GenAI into teaching introduces ethical and professional complexities. Questions of authorship, transparency and ownership arise when AI is involved in content generation. Institutions must clarify policies around attribution, disclosure and intellectual property.
Content reliability also presents concerns. LLMs can produce biased or hallucinated information, posing risks for pedagogically sensitive materials (Blodgett et al., 2020). Instructors must evaluate outputs critically, particularly in high-stakes or culturally nuanced contexts. Such concerns resonate with emerging frameworks like ARCHED, which argues for transparency, accountability and human-centered oversight in AI-supported instructional design (Li et al., 2025).
Institutional policies have largely focused on student use of GenAI, with less emphasis on supporting instructors. McDonald et al. (2025) found that while many universities offer guidance for learners, few provide resources tailored to faculty needs. Supporting experimentation, professional development and community-building is essential to foster responsible innovation and reduce misuse or hesitation.
While this study identifies recurring concerns around authorship, content reliability and institutional policy gaps, it does not claim to provide new frameworks for ethical practice. Instead, it synthesizes existing challenges to underscore the importance of institutional support and critical instructor oversight in scaling GenAI tools responsibly.
2.6 Gaps in the literature and relevance to the present study
Although the literature on GenAI in education is expanding rapidly, it mostly focuses on student-facing applications like writing support, AI tutors and plagiarism detection (Bai et al., 2023; Kajiwara and Kawabata, 2024). Comparatively little research investigates how instructors use GenAI for curriculum development, particularly in case-based pedagogies, which demand domain-specific narrative structures, realism and relevance.
Even fewer studies examine the intersection of GenAI and academic entrepreneurship, even though many instructors are independently developing, piloting and iterating on AI-supported tools. This entrepreneurial behavior reflects a growing shift in higher education, where instructors act as not just content deliverers but also innovators and system designers responding to evolving learner needs and institutional constraints (Kotturi et al., 2024; Wang, 2025).
Recent frameworks also provide structured approaches to integrating GenAI into instructional design. For example, the GAIDE framework (Dickey and Bejarano, 2024) emphasizes the use of generative AI in course content development, offering practical steps for aligning AI outputs with pedagogical objectives. Similarly, the ARCHED framework (Li et al., 2025) highlights human-centered, transparent and collaborative design principles that address ethical and institutional concerns. While this study does not directly adopt these frameworks, it complements them by empirically exploring how instructors interact with a GenAI case generator tool, grounding theoretical insights in usage data.
This study addresses these gaps by analyzing real-world instructor usage data from a custom-built GenAI case generator, applying TAM and CLT to understand behavioral intention, design cognition and the affordances of co-creation. By embedding this analysis within both instructional theory and innovation practice, the study contributes to a more nuanced understanding of how faculty navigate GenAI adoption.
In doing so, this research not only expands the empirical foundation of instructor-facing GenAI applications but also foregrounds the entrepreneurial potential of educators in shaping the future of teaching and learning. It offers practical and theoretical insights for institutions, policymakers and researchers seeking to support sustainable, ethical and high-impact AI integration in higher education.
3. Methodology
This study adopts a design-based research framework, complemented by a mixed-methods exploratory case study approach, to investigate how a GenAI-powered platform (Case Generator – CG) supports instructor-led case creation in higher education. DBR emphasizes iterative tool development in authentic settings (Anderson and Shattuck, 2012), while the case study perspective (Yin, 2009) provides in-depth insights into real-world deployment. Mixed methods integrate user engagement metrics with qualitative analysis of system-generated instructional artifacts (Zawacki-Richter et al., 2019).
The intervention involved the development and testing of a Web-based GenAI platform designed to support faculty in generating customized case studies. Though not experimental in the strict sense, the study included implementation of an AI-based tool in a naturalistic setting and data-driven evaluation of its use and outputs (Sandoval, 2014).
The platform integrated GPT-3.5 through API and WordPress, offering two services: ready-to-download AI-generated cases and customized case prompts. Service A provides a cost-effective alternative to commercial case libraries, while Service B enables discipline-specific case generation tailored to instructors’ needs. The site, referred to as Case Generator (CG), is accessible to users but anonymized in this paper to prevent conflicts of interest.
3.1 Data sources
Three primary data sources were analyzed:
User engagement metrics
It is captured through Google Analytics, including session counts, user types, page views, engagement time and geographic distribution. These metrics provide insight into how users navigate and interact with CG:
Service data
Collected through WooCommerce integration, this includes download frequency, selected case attributes (topic, type and level) and anonymized user information. These data reflect user preferences and behavioral patterns related to case acquisition:
Generated case study content
It is analyzed from Service B, which uses structured prompts (∼2,000 words) embedded into a form-based interface. User entries (e.g. topic, level and category) are parsed through custom logic to produce personalized prompts. All user inputs and generated content were stored in anonymized logs for qualitative analysis.
3.2 Data analysis
A qualitative, multi-method analysis was applied to explore emerging patterns:
Content analysis (Hsieh and Shannon, 2005) was used to identify prevalent themes and case attributes from downloaded and generated content.
Thematic analysis (Braun and Clarke, 2006) examined instructional objectives and pedagogical patterns across AI-generated cases.
Triangulation (Denzin, 2017) cross-validated themes by comparing user behavior metrics, service logs and generated artifacts.
This approach aligns with Yin’s (2009) model of analytical generalization, drawing conceptual insights rather than statistical inferences. The study is theoretically grounded in TAM and CLT. TAM provided insights into PU and ease of use (Davis, 1989), while CLT shaped evaluations of how CG minimized intrinsic and extraneous cognitive load through the automation of case structure and formatting (Sweller, 2011; Chandler and Sweller, 1991).
The method responds to recent calls to explore AI tools for instructors that facilitate co-design and innovation (Chatterjee and Bhattacharjee, 2020; Holmes et al., 2019), prioritizing objective usage statistics rather than self-reported data (Luckin et al., 2016).
While this study draws on unobtrusive system data, it does not incorporate direct instructor perspectives through surveys or interviews. As a result, constructs such as PU and ease of use are inferred indirectly from behavioral data rather than measured explicitly. This design choice reflects the exploratory scope of the study but also introduces limitations in interpretive depth.
3.3 Ethical considerations
Only anonymized system-level data and generated content were analyzed. No personal feedback or identifiable data were retained. This study followed institutional and national ethical standards for educational research.
3.4 Tool development and implementation
The Case Generator platform was developed using WordPress because of its flexibility and extensibility. WooCommerce enabled digital case downloads, while custom code plugins allowed unique algorithms to support prompt generation and AI integration. These design choices ensured the platform could be iteratively improved in alignment with usage data, reflecting the principles of design-based research.
The platform was integrated with OpenAI’s GPT 3.5 model through a custom-coded plugin, enabling communication between user-facing forms and the language model. This allowed instructors to input case attributes (e.g. category, topic and type) and receive tailored outputs. The integration emphasized ease of use, minimizing technical barriers for instructors, which aligned with the TAM construct of PEOU.
Prompt engineering was a critical component of tool development. Through iterative refinement, structured templates were designed to ensure AI-generated cases aligned with instructional needs and avoided incoherent outputs. This process underscored the importance of human input quality, as AI models interpret prompts literally. The refinement of prompts exemplified the DBR approach of iterative prototyping, ensuring that the system produced usable cases for real-world teaching contexts.
4. Findings
The findings are organized into three subsections reflecting the three core data sets collected:
user engagement metrics;
service data; and
case study content analysis.
These insights illustrate how instructors used the GenAI-powered CG platform, what types of content were requested or downloaded and what themes emerged from the AI-generated instructional materials.
4.1 Website user analysis
“Total users” is a telling metric for website growth over a standard quarter, although it may experience seasonal ups and downs. “New users” is important, as it represents the total number of new visitors accessing the website. Generally, the “New Users” metric makes up a significant portion of the “Total Users” visiting the site. This percentage varies depending on the website’s content, topic and type, generally ranging between 75% and 90% of Total Users. The remaining users are typically “returning users.” As displayed in Figure 1, CG has reached a total of 2.8 K users since its launch, with an average engagement time of 1 min and 56 s.
The chart plots total users on the vertical axis and months from July 2023 to April 2024 on the horizontal axis. The data shows an initial decrease from July to September 2023, followed by gradual growth through the year, with a significant rise starting in January 2024 and peaking at around 700 users in April 2024. The total user count displayed on the chart is 2.8 thousand, indicating consistent audience expansion over time.CG’s total users and average engagement time
Source: Authors’ own work
The chart plots total users on the vertical axis and months from July 2023 to April 2024 on the horizontal axis. The data shows an initial decrease from July to September 2023, followed by gradual growth through the year, with a significant rise starting in January 2024 and peaking at around 700 users in April 2024. The total user count displayed on the chart is 2.8 thousand, indicating consistent audience expansion over time.CG’s total users and average engagement time
Source: Authors’ own work
“Average engagement time” is a website-specific metric; however, it does provide insights as to how the website content is used and if it is engaging or not. For example, CG is mainly a source of AI-generated case downloads (Service A) and AI-assisted case writing (Service B). Both services have different levels of user engagement. Service A has an average engagement time of 39 s, based on how the user typically browses until a download action. Meanwhile, Service B has an average engagement time of 1 min and 43 s. Given that its generator tool requires more steps as a service and allows more decision room for users, a higher average was initially expectedly (Table 1).
Average engagement time comparison by service – Service A vs Service B
| Service | Average engagement time | Period |
|---|---|---|
| Service A | 39 s | July 2023–April 2024 |
| Service B | 1 min 43 s | July 2023–April 2024 |
These differences in engagement times between Service A (short browsing) and Service B (longer interaction) can be interpreted through CLT: Service A minimizes extraneous load by providing ready-made content, while Service B encourages germane load by requiring instructors to actively refine prompts. Similarly, the high proportion of returning users suggests that instructors perceived ongoing usefulness in the platform, reflecting TAM’s construct of PU.
As for “user traffic sources,” most CG’s new users come from direct traffic, as indicated by the high number of users accessing the platform directly via URL entries or bookmarks (Figure 2). Additionally, organic search represents a significant acquisition channel. Geographically, CG’s users are mostly from the USA, Canada and India (Figure 3). This composition might reflect regional content preferences and language relevance.
The figure categorises new users by traffic sources. Direct traffic accounts for 38.18 percent of total users, followed by organic search at 34.88 percent, and organic video at 19 percent. Organic social, referral, unassigned, and email sources contribute smaller shares at 5.87, 1.43, 0.43, and 0.21 percent respectively. The total user count is 2.8 thousand, demonstrating that most user engagement originates from direct access and search-based discovery.CG’s traffic ranked by channels
Source: Authors’ own work
The figure categorises new users by traffic sources. Direct traffic accounts for 38.18 percent of total users, followed by organic search at 34.88 percent, and organic video at 19 percent. Organic social, referral, unassigned, and email sources contribute smaller shares at 5.87, 1.43, 0.43, and 0.21 percent respectively. The total user count is 2.8 thousand, demonstrating that most user engagement originates from direct access and search-based discovery.CG’s traffic ranked by channels
Source: Authors’ own work
The chart shows geographic distribution of total users. The United States leads with 28.5 percent, followed by Canada at 20.3 percent, and India at 14.9 percent. The Philippines contributes 6.5 percent, the United Kingdom 5.0 percent, China 2.1 percent, and Pakistan 2.0 percent. The remaining 105 countries collectively make up the rest of the user base. The total number of users is 2.8 thousand, indicating broad international engagement with higher concentration in North America and South Asia.CG’s traffic ranked by country of origin
Source: Authors’ own work
The chart shows geographic distribution of total users. The United States leads with 28.5 percent, followed by Canada at 20.3 percent, and India at 14.9 percent. The Philippines contributes 6.5 percent, the United Kingdom 5.0 percent, China 2.1 percent, and Pakistan 2.0 percent. The remaining 105 countries collectively make up the rest of the user base. The total number of users is 2.8 thousand, indicating broad international engagement with higher concentration in North America and South Asia.CG’s traffic ranked by country of origin
Source: Authors’ own work
Based on user activity trends (Figure 4), a significant upward trend can be observed after January 2024, which coincides with the adoption of social media content strategies on LinkedIn and YouTube. While activity is volatile in the short term (seven-day active users), a steady growth of activity for the long term is noticed (30-day active users).
The chart tracks user activity from July to April, displaying data for one-day, seven-day, and thirty-day active users. The thirty-day users rise steeply to 747 by April, seven-day users reach 127, and one-day users 21. Activity remains low and stable until early 2024, after which a sharp upward trend appears, suggesting significant user growth and higher engagement in recent months.CG’s user activity trends for 1, 7 and 30 days
Source: Authors’ own work
The chart tracks user activity from July to April, displaying data for one-day, seven-day, and thirty-day active users. The thirty-day users rise steeply to 747 by April, seven-day users reach 127, and one-day users 21. Activity remains low and stable until early 2024, after which a sharp upward trend appears, suggesting significant user growth and higher engagement in recent months.CG’s user activity trends for 1, 7 and 30 days
Source: Authors’ own work
From user engagement events, Page Views and User Engagement are by far the most predominant events, where users spend a significant amount of time on the platform. In such cases, high interactions with the platform usually translates to good user retention. Meanwhile, events with lower frequency like “Form Start” and “Form Submit,” though less frequent, are equally important and are probably the ones that signify the most critical and important user interactions for the platform such as sign-ups or active participation in platform activities that ultimately convert to its users and improve user retention (Figure 5).
The figure lists various tracked events with event counts and corresponding user totals. The total event count is 45,358 across 2,796 users. Page views account for 16,410 events involving 2,789 users, while user engagement records 10,872 events across 1,881 users. Other key actions include scroll (5,316 events, 639 users), session start (3,986 events, 2,772 users), form start (2,213 events, 868 users), first visit (2,772 events, 2,769 users), and form submission (2,678 events, 571 users). The data highlights that most users interact through page views and engagements, with fewer completing forms.CG’s user engagement events ranked
Source: Authors’ own work
The figure lists various tracked events with event counts and corresponding user totals. The total event count is 45,358 across 2,796 users. Page views account for 16,410 events involving 2,789 users, while user engagement records 10,872 events across 1,881 users. Other key actions include scroll (5,316 events, 639 users), session start (3,986 events, 2,772 users), form start (2,213 events, 868 users), first visit (2,772 events, 2,769 users), and form submission (2,678 events, 571 users). The data highlights that most users interact through page views and engagements, with fewer completing forms.CG’s user engagement events ranked
Source: Authors’ own work
The higher proportion of returning users suggests that instructors perceived recurring value in the tool, aligning with TAM’s construct of PU. Similarly, the longer engagement times observed for Service B versus Service A may reflect deeper cognitive involvement, resonating with CLT’s distinction between extraneous and germane load. These behavioral indicators suggest perceived utility and active engagement, though they remain indirect proxies.
4.2 Service A
Enabling Service A’s provision and data collection was an extensive process. Each case had to be initially planned, developed by the AI using the correct prompts, including topics, categories, themes and disciplines, then reviewed before made available. The cases were formatted in a user-friendly manner, similar to existing alternatives in the market (Table 2). The case repository comprising Service A offers a total of 68 AI-generated cases, divided into 17 case categories, with each case being categorized in more than one category, at times (Table 3).
List of existing case repositories used to model Case Generator’s service a user interface
| Repository name | Focus area | Website |
|---|---|---|
| Darden Business Publishing | Business, ethics and finance | Dardenbusinesspublishing.com |
| Harvard Business Publishing Education | Business and management | Hbsp.harvard.edu |
| INSEAD Case Publishing | Leadership, innovation and entrepreneurship | Insead.edu |
| Ivey Publishing | International business and strategy | Iveypublising.ca |
| Kellog School of Management | Marketing and growth strategies | Kellogg.northwestern.edu |
| MIT Sloan Teaching Innovation Resources | Operations and information technology | Mitsloan.mit.edu |
| Stanford Graduate School of Business | Business, leadership and entrepreneurship | Gsb.stanford.edu |
| The Case Centre | Business, management and marketing | Thecasecentre.org |
| Thunderbird School of Global Management | Global business strategies | Thunderbird.asu.edu |
List of case categories available in Service A
| Case categories | |
|---|---|
| Accounting | International business |
| Communications | Leadership |
| Economics and public policy | Legal |
| Entrepreneurship | Marketing |
| Ethics | Operations management |
| Finance | Organizational behavior |
| General management | Strategy |
| Human resources management | Sustainability |
| Information systems | |
When users visit CG’s case repository and select a case for download, they are directed to a page displaying the case, via WooCommerce (Figure 6). Each case provides various details such as discipline, case ID, license number, number of pages and whether a case study analysis document or teaching notes are included. Additionally, categories aligned to the case are listed for instructors to preview its applicability to their intended courses. During the checkout process, information such as user’s first name, last name, associated institution and email address are collected. Once the checkout process is completed, the user can download cases without any monetary cost.
The image features a close-up view of a stylish car wheel with a prominent black rim and an eye-catching red brake caliper. Adjacent to the wheel, there is a piece of paper that appears to contain handwritten notes or details. The design of the wheel displays a multi-spoke pattern, highlighting its modern aesthetic. The overall composition emphasizes both the mechanical and artistic aspects of automotive design, capturing a moment of attention to detail.Service A’s case checkout cart
Source: Authors’ own work
The image features a close-up view of a stylish car wheel with a prominent black rim and an eye-catching red brake caliper. Adjacent to the wheel, there is a piece of paper that appears to contain handwritten notes or details. The design of the wheel displays a multi-spoke pattern, highlighting its modern aesthetic. The overall composition emphasizes both the mechanical and artistic aspects of automotive design, capturing a moment of attention to detail.Service A’s case checkout cart
Source: Authors’ own work
A total of 67 cases were downloaded between July 2023 and April 2024 (Figure 7). Among these, the top five most popular case categories were Business, International Business, General Management/Strategy, Organizational Behavior and Marketing.
This line graph illustrates the total downloads from July 2023 through April 2024. The x-axis represents the months, while the y-axis indicates the number of downloads, ranging from zero to fifteen. Monthly data points appear, with several spikes, notably reaching a peak of ten downloads in January 2024 and another peak of six downloads in November 2023. Data points are plotted for each month, with numerous values of one and two scattered throughout, suggesting fluctuating download activity. The graph is organized chronologically, moving from July to April, allowing for an easy trace of download patterns over time.Service A download trends from July 2023 to April 2024
Source: Authors’ own work
This line graph illustrates the total downloads from July 2023 through April 2024. The x-axis represents the months, while the y-axis indicates the number of downloads, ranging from zero to fifteen. Monthly data points appear, with several spikes, notably reaching a peak of ten downloads in January 2024 and another peak of six downloads in November 2023. Data points are plotted for each month, with numerous values of one and two scattered throughout, suggesting fluctuating download activity. The graph is organized chronologically, moving from July to April, allowing for an easy trace of download patterns over time.Service A download trends from July 2023 to April 2024
Source: Authors’ own work
4.3 Service B
Service B allows users to turn their case idea into an advanced prompt. This prompt generator (Figure 8) requires the input of case category, case topic and case type so that the generator can output the required prompt, based on the users’ inputs. Finally, the generator will provide the prompt for the user to copy.
The interface is divided into two sections. On the left, the user inputs the case category, topic, type, and email before generating a prompt. On the right, the generated case prompt summary is shown. It includes the case category as Biotechnology, the topic as Covid-19 and Human Populations, and the style as Research Case. Below, a detailed generated prompt provides guidance for writing an elaborate and realistic case study with emphasis on understanding context, preparing data, and creating believable business examples.Service B case prompt generator
Source: Authors’ own work
The interface is divided into two sections. On the left, the user inputs the case category, topic, type, and email before generating a prompt. On the right, the generated case prompt summary is shown. It includes the case category as Biotechnology, the topic as Covid-19 and Human Populations, and the style as Research Case. Below, a detailed generated prompt provides guidance for writing an elaborate and realistic case study with emphasis on understanding context, preparing data, and creating believable business examples.Service B case prompt generator
Source: Authors’ own work
Users must first select one of 38 predetermined case categories. The available themes span from accounting to AI, marketing, strategy and human resources management, among others in the business/management field. Then, users must manually type their topic of interest. Topics can vary between single-word choices (e.g. marketing) to more detailed topic choices such as the “four P’s of marketing” or their “relevance to university-level students.” The next step requires users to select among four options of case type. Each case category contains unique and specific definitions of the types of cases that fall within each category. This might appear to overly restrict the specificity of the cases that the prompt generator can create, but this design hinged on research and an understanding of the common types of cases that users are interested in creating and was crafted to help users, not to restrict them.
After the user inputs all three pieces of information and their email address, they must click on a “Generate Prompt” button. The generator combines their inputs in a pre-written prompt template. At the bottom, the generator includes a short output that summarizes the inputs: the case category, topic and style, chosen by the user. Below this, the generator includes a prompt that is tailored to the inputs of the user. To make it easier to use, a “Copy to Clipboard” button was created, allowing users to copy the prompt and paste it in their AI interface, so they can benefit by having used the case prompt.
At each use of Service B, data inputs are collected: case category, topic and style, as well as the user’s email address, the date and time the generator is used. All inputs are recorded in a text file over the website’s database, programmed to automatically update (with no limit to how many times a user can input data) with new inputs (Figure 9). This process gives CG’s website greater insight into user habits and preferences, allowing for more informed and targeted content and prompts.
The image displays several entries of email correspondence, formatted in a consistent layout. Each entry includes sections for email address, date and time of the email, theme, topic, and type of case. The dates occur between the nineteenth and twentieth of December in twenty twenty-three, with recorded times reflecting various hours. Themes include Economics and Public Policy, Entrepreneurship, and Operations Management, while the topics address issues such as staffing problems in small businesses and production strategy. The type of case alternates between Research Case and Business Case, providing a structured view of the email data related to these themes.Excerpt from Service B’s text inputs from collected data
Source: Authors’ own work
The image displays several entries of email correspondence, formatted in a consistent layout. Each entry includes sections for email address, date and time of the email, theme, topic, and type of case. The dates occur between the nineteenth and twentieth of December in twenty twenty-three, with recorded times reflecting various hours. Themes include Economics and Public Policy, Entrepreneurship, and Operations Management, while the topics address issues such as staffing problems in small businesses and production strategy. The type of case alternates between Research Case and Business Case, providing a structured view of the email data related to these themes.Excerpt from Service B’s text inputs from collected data
Source: Authors’ own work
Over the course of this study, Service B was used 982 times by 486 unique users (Table 4). While the increase seen in the first quarter of 2024 can be attributed to improvements made in the service itself, the use of social media channels and profiles for sharing CG services must also be factored in, given the increase in organic traffic in this period. The preferred case themes and types are detailed by month on Table 5.
Users and generated prompts metrics for Service B
| Metric | 2023 | 2024 | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| July | Aug | Sept | Oct | Nov | Dec | Jan | Feb | Mar | Apr | |
| Total prompts generated per month | 11 | 51 | 90 | 50 | 82 | 55 | 220 | 93 | 155 | 175 |
| Total unique users per month | 11 | 50 | 41 | 28 | 44 | 29 | 83 | 48 | 93 | 118 |
| Total prompts generated per year | 339 | 643 | ||||||||
| Total unique users per year | 144 | 342 | ||||||||
Most popular case themes and topics selected in Service B
| Year | Month | Most popular case themes | (% of total cases) per month |
|---|---|---|---|
| 2023 | July | Marketing | 27.27 |
| Aug | General management/strategy | 13.73 | |
| Sept | HR management | 23.33 | |
| Oct | HR management | 16.00 | |
| Nov | Marketing | 18.29 | |
| Dec | Marketing | 14.55 | |
| 2024 | Jan | International business | 13.18 |
| Feb | Marketing | 34.41 | |
| Mar | Marketing | 12.26 | |
| Apr | Marketing | 6.86 | |
| Year | Month | Most popular case type | (% of total cases) per month |
| 2023 | July | Business case | 45.45 |
| Aug | Business case | 49.02 | |
| Sept | Business case | 42.22 | |
| Oct | Research case | 38 | |
| Nov | Business case | 35.37 | |
| Dec | Research case | 41.82 | |
| 2024 | Jan | Business case | 68.18 |
| Feb | Business case | 59.14 | |
| Mar | Business case | 38.71 | |
| Apr | Business case | 42.29 |
The uptake of Service B indicates that instructors valued the opportunity to co-design case materials, rather than relying solely on pre-generated cases. This aligns with TAM’s PEOU, as the interface reduced technical barriers, while also reinforcing PU by providing context-specific, customizable outputs.
This preference for tailored prompts over ready-made cases highlights instructors’ need for context-specific materials, which supports the view that GenAI functions not merely as a content generator but also as a co-design partner, reinforcing TAM’s emphasis on PU in professional tasks.
5. Discussion
This study examined instructors’ engagement with a GenAI-powered platform designed to generate customized teaching case studies. The analysis of usage metrics, content preferences and generated artifacts supports a growing recognition that AI tools can serve not only students but also educators in higher education (Zawacki-Richter et al., 2019; Holmes et al., 2019).
Usage data illustrated in Figure 1 and Table 1 reveals patterns of engagement indicative of perceived utility and relevance. The predominance of repeat visits and high interaction with the case customization service (Service B) supports the TAM proposition that tools with clear, practical value are more likely to be adopted (Davis, 1989; Granić and Marangunić, 2019). Moreover, the longer interaction times with customized prompts suggest that instructors were willing to invest cognitive effort when outputs aligned with their teaching objectives, resonating with CLT’s distinction between extraneous and germane load. In other words, instructors tolerated higher engagement time when it translated into more pedagogically relevant outcomes. These interpretations remain proxies derived from behavioral data, but they highlight how TAM and CLT provide explanatory lenses for understanding observed usage patterns.
Instructors appeared particularly responsive to the tool’s time-saving capabilities and topic specificity, as reflected in the sustained engagement metrics and the range of use cases observed (Table 1 and Figure 5).
Moreover, the widespread selection of tailored case categories, especially in entrepreneurship, marketing and management, points to perceived alignment between the tool’s outputs and instructors’ pedagogical needs (Table 4). This mirrors TAM’s emphasis on PU as a key determinant of acceptance and highlights how context-relevant outputs drive engagement.
From the lens of CLT (Sweller, 2011; de Jong, 2010), the tool appears to reduce extraneous load by offering structured templates and streamlined input interfaces. The analysis of generated case outputs (Table 4) demonstrates the system’s ability to support instructors in developing scenario-based narratives with appropriate complexity, including decision dilemmas, stakeholder roles and quantitative data points. These components support instructional goals by allowing instructors to redirect cognitive energy from structural creation to pedagogical refinement.
The overall volume and consistency of generated case content, as previously described, reflect meaningful engagement and underscore the viability of AI-supported co-design. These findings echo prior arguments that intelligent systems can scaffold creative processes and support innovation in instructional design (Roll and Wylie, 2016; Luckin et al., 2016).
At the same time, these findings must be interpreted cautiously. Usage metrics provide suggestive evidence of usefulness and engagement but cannot, on their own, establish pedagogical effectiveness, without qualitative feedback or student learning outcomes, our claims remain exploratory, What the present analysis demonstrates is the potential alignment of behavioral patterns with TAM and CLT constructs, not definitive proof of adoption or cognitive benefit.
At an institutional level, the uneven availability of support structures, highlighted by prior research on GenAI policy adoption in higher education (McDonald et al., 2025), emphasizes the need for formalized frameworks to encourage instructor experimentation. The clear demand for tools like the CG platform suggests that faculty may be willing to engage with GenAI tools when such engagement is framed as professionally relevant, technically accessible and ethically supported.
Finally, the findings also reveal an important design consideration: instructors are not passive adopters but active co-designers. Their interactions with the tool, visible in both the customization data and the diversity of case structures generated (Table 5), suggest that GenAI platforms are most effective when they enhance, rather than automate, instructor creativity. The tool’s affordances, its modular prompts, decision points and embedded support, enabled educators to produce cases that matched their instructional objectives, validating prior claims about the transformative potential of co-designed AI systems (Chatterjee and Bhattacharjee, 2020).
5.1 Implications for research
The insights from this study contribute to both educational practice and scholarly discourse on AI in teaching. Practically, the tool shows that GenAI platforms can serve as viable co-design partners for instructors, alleviating workload and enhancing pedagogical experimentation. Institutions seeking to support faculty in course preparation could adopt similar models, provided they also address ethical, usability and integration issues highlighted in recent literature (McDonald et al., 2025; Crowther and Hamdan, 2024). In particular, the pattern of returning users is noteworthy. Repeat engagement suggests that instructors found sustained value in the tool, which has practical implications for institutions considering adoption. While not equivalent to formal measures of effectiveness, returning user patterns provide an important behavioral proxy for PU and should be examined more closely in future research. Future research should consider direct measures of TAM constructs (e.g. surveys of PU/PEOU) and structured assessments of cognitive load to more robustly validate the relationships that our behavioral proxies only tentatively suggest.
This work calls the attention of researchers to the importance of system-generated usage data as an underexplored source of insight into how instructors interact with AI tools. Rather than relying solely on surveys or interviews, usage logs, input prompts and content outputs provide rich, unobtrusive data that reflect real user behavior. This supports the growing recognition of learning analytics in educational technology evaluation (Luckin et al., 2016).
This study also reinforces the need to shift AI in education research beyond learner-facing applications. As our literature review emphasized, most studies to date focus on GenAI’s effects on student learning, tutoring or academic integrity (Bai et al., 2023; Kajiwara and Kawabata, 2024). This study uniquely demonstrates the instructional design affordances of GenAI and presents a methodology that can be replicated in future exploratory interventions.
5.2 Limitations and future research
Several limitations must be acknowledged, the most critical one being this study’s reliance on system-generated data without complementary instructor feedback. Constructs such as PU, PEOU and cognitive load dimensions are inferred indirectly from behavioral proxies (e.g. return visits and engagement times) rather than measured explicitly. This reliance on analytics alone means that conclusions about instructor motivation, perceptions or cognitive processes remain speculative. Future studies should integrate complementary methods such as interviews, focus groups or perception surveys to more directly assess TAM constructs and cognitive load dimensions, providing a fuller account of instructor experiences.
Second, this study focused on early-stage adoption, and longitudinal data would be valuable in evaluating sustained usage and integration. Third, the pilot involved a self-selected sample of early adopters; generalizing to broader populations requires replication in diverse institutional settings.
Because this study relies primarily on system-generated usage data, its results should not be read as rigorous metrics of effectiveness. While repeat visits, engagement time and service preferences are suggestive of adoption, they do not capture actual improvements in teaching outcomes or reductions in instructor workload. Future studies should triangulate behavioral analytics with direct instructor feedback and measures of case quality to more convincingly demonstrate effectiveness.
Comparative studies of different GenAI tools, as well as domain-specific adaptations (e.g. for health or engineering education), would also provide valuable insights. Additionally, as GenAI tools evolve to include multimodal and multilingual capabilities, future work should examine how these affect cognitive load and usability.
6. Conclusion
This study adds to a growing but still nascent literature on AI-enabled instructional design by offering empirical insights into how educators engage with a GenAI-powered tool to generate teaching case studies. By grounding the research in TAM and CLT and implementing a design-based research methodology, this study offers both theoretical and practical contributions.
The findings suggest that such tools can reduce cognitive effort, support pedagogical creativity and align well with instructors’ expectations of relevance and efficiency. However, the responsible adoption of GenAI in education also demands institutional policies that balance innovation with quality assurance and ethical accountability. This study’s contribution lies in surfacing and contextualizing these ethical and institutional challenges rather than proposing entirely novel solutions, offering a foundation for future empirical and policy-focused research.
Future research should expand beyond usage analytics to incorporate instructor perspectives through interviews, surveys and longitudinal designs. Comparative studies across disciplines and contexts could also illuminate differences in adoption patterns and instructional needs. As the education sector continues to navigate the rapid rise of GenAI, studies like this one underscore the need to center the instructor’s role, as not merely a user but also a co-designer of educational futures.
