Executive Overview
As higher education grapples with the accelerating integration of artificial intelligence, a central anxiety has gripped faculty lounges and academic committees: When students are given a choice, will they abandon traditional, instructor-created materials in favor of slick, AI-generated alternatives?
Fears of academic disintermediation—the complete bypass of the professor in favor of generative tools—have dominated conversations about the future of teaching. However, a recent classroom case study conducted by Jess Kahlow, an adjunct assistant professor in the College of Education at the University of Texas at Arlington, suggests that these fears may be misplaced.
Rather than viewing artificial intelligence as a substitute for human instruction, students are utilizing AI-generated resources as complementary pathways for learning. When given transparent, side-by-side choices between an instructor-led audio podcast and an AI-condensed video summary, the vast majority of students did not choose one over the other. Instead, roughly 84% of participating students engaged with both formats.
This deep dive examines Kahlow’s classroom experiment, exploring the methodology behind the "Choose Your Own Adventure" module, the Canvas analytics that tracked student behavior, the critical nuances of completion rates, and what these findings mean for the broader evolution of instructional design in the age of generative AI.
Detailed Chronology: The "Choose Your Own Adventure" Experiment
To move past speculation and observe actual student behavior, Kahlow designed an intentional, transparent classroom activity for the opening module of a graduate-level course on multimedia design tools.
Setting the Stage: Complete Transparency
In modern pedagogical research, trust and transparency are paramount, particularly when introducing experimental or emerging technologies. Kahlow wanted to ensure that her students were under no illusions about what they were accessing. Before diving into the official course content and materials page for the first module, students encountered a prominent prompt labeled: "Choose your own adventure!"
The prompt outlined two distinct ways to review the core concepts of the module overview:
- Option One: A 13-minute, 31-second audio podcast featuring Kahlow’s real voice explaining the module’s foundational concepts.
- Option Two: A 3-minute, 51-second AI-generated explainer video created using Google’s NotebookLM, which ingested the transcript of Kahlow’s original podcast and condensed it into a shorter visual presentation.
Students were given absolute agency. They could choose to engage with the instructor’s podcast, the AI-generated video, both resources, or neither resource at all. Crucially, there were no grade penalties or extra-credit rewards tied to their choice. The explicit goal was not to test whether AI could outperform human teaching, but rather to observe how students naturally navigated an environment where an AI-generated resource was offered transparently alongside a traditional instructor-created artifact.
Tracking the Digital Footprint
With 32 graduate students enrolled in the course, the digital traces left behind in the learning management system provided an objective window into student behavior. Kahlow pulled the Canvas Studio view analytics for both resources, allowing her to track not just which links were clicked, but the depth of engagement and completion rates for each format.
The results challenged conventional assumptions about student laziness or a preference for the path of least resistance.
Out of 32 enrolled students, 29 accessed at least one of the resources. When analyzing the patterns of access, the data revealed an unexpected trend: the single most common behavior was not choosing the human-made podcast or the AI video in isolation. It was opening both.
Supporting Context & Metrics: Deconstructing the Data
A closer examination of the Canvas analytics reveals fascinating details about how graduate students consume supplementary course materials.
Engagement Breakdown
- Total Enrollment: 32 students
- Total Unique Access: 29 students accessed at least one resource.
- Podcast Engagement: 25 unique students opened the instructor-created podcast.
- AI Video Engagement: 25 unique students opened the AI-generated explainer video.
- Single-Source Users (Podcast Only): 4 students.
- Single-Source Users (AI Video Only): 4 students.
- Dual-Source Users (Both): 21 students.
Among the students who accessed either resource, an impressive 84% accessed both. Furthermore, among the minority of students who chose only one resource, there was a dead split: four students opened solely the video, and four opened solely the podcast. The data demonstrated that the AI-generated resource did not render the instructor-created resource obsolete, nor did the presence of the professor’s podcast make the AI tool irrelevant.
Completion Rates: The Time Commitment Factor
While opening a resource indicates initial interest, completion rates reveal true follow-through. Here, noticeable differences emerged between the two formats, largely driven by the stark contrast in length.
-
The AI-Generated Video (3 minutes, 51 seconds):
- Average completion rate: 89.7%
- Median completion rate: 100% (indicating at least half of the students who opened the video watched the entirety of it).
- 17 out of 25 students (68%) reached 100% completion.
- 20 out of 25 students (80%) completed at least 90% of the video.
-
The Instructor-Created Podcast (13 minutes, 31 seconds):
- Average completion rate: 76.1%
- Median completion rate: 100% (indicating at least half of the students listened to the entire recording).
- 15 out of 25 students (60%) reached 100% completion.
- 16 out of 25 students (64%) completed at least 90% of the podcast.
While both resources enjoyed high overall completion rates from those who opened them, the shorter AI-generated video was finished more consistently from start to finish. This outcome aligns with established multimedia learning principles: shorter chunks of multimedia content are inherently easier to digest in a single sitting. However, it adds an important layer of nuance to the "students chose both" finding. Students were not merely clicking both links; when opting for a faster review mechanism, they were significantly more likely to cross the finish line with the concise AI-generated summary.
Official Perspectives and Pedagogical Implications
Jess Kahlow, reflecting on the outcomes of her classroom experiment, noted that her initial expectations were defied.
"Going into this, I really did not know what to expect," Kahlow stated. "However, I did hope that at least some students would listen to my podcast even though it was longer. That said, I was happy to see that most students chose to spend additional time engaging with both resources, rather than treating the shorter option as a replacement for the longer one."
Rather than asking the binary question, "Which version should I choose?" students appeared to be adopting a more sophisticated cognitive strategy: "How can these different versions help me learn?"
AI as an Additional Pathway
Educational technology experts have long argued that Universal Design for Learning (UDL) principles require multiple means of representation, expression, and engagement. By providing an AI-generated explainer video derived directly from a faculty member’s core lecture or podcast, educators can cater to diverse learning styles—auditory, visual, and summary-oriented—without sacrificing the authentic voice and guidance of the instructor.
In this context, artificial intelligence functions less like a disruptive competitor to human educators and more like an instructional amplifier. It creates secondary and tertiary pathways for students to reinforce complex concepts, review material before examinations, or quickly catch up on foundational terminology.
The Power of Transparency
A critical pillar of Kahlow’s methodology was radical transparency. At no point were students tricked into interacting with synthetic media. They knew precisely that NotebookLM had processed their professor’s original transcript to generate the short video.
This transparency shifts the classroom dynamic from one of suspicion—where students wonder if professors are outsourcing their labor to machines—to one of collaborative exploration. Students are invited to witness firsthand how generative tools can synthesize and condense information, modeling professional practices for the ethical and practical use of AI in modern workplaces.
Future Outlook: Redefining the Role of AI in Higher Education
As universities look toward the horizon of generative AI integration, case studies like Kahlow’s provide a much-needed antidote to generalized panic. The assumption that AI will disintermediate professors assumes a zero-sum game where machine-generated content automatically displaces human-generated instruction.
The empirical evidence from this graduate multimedia design course points toward a more collaborative future:
- Evolution of Instructional Design: Faculty members may increasingly use generative tools to automatically spin off derivative study aids—glossaries, flashcards, summary videos, and podcasts—from their primary lectures, tailoring content density to different student needs.
- Shift in Student Metacognition: When given agency and transparency, students do not look for shortcuts to avoid learning; rather, they stack learning resources. They use quick AI summaries to prime their brains before diving into deep-dive faculty lectures, or they use human podcasts for foundational context and AI videos for rapid revision.
- Addressing the Caveats: Researchers note important limitations to keep in mind. This study involved graduate students in a multimedia design program—an audience inherently more tech-savvy, curious, and motivated to explore AI tools than a general-education undergraduate population. Future research must test these frameworks across diverse disciplines, introductory courses, and larger cohorts.
Ultimately, the future of artificial intelligence in higher education will not be defined by whether machines can replace professors. It will be defined by how creatively and transparently educators can harness generative technologies to expand the breadth, accessibility, and depth of the learning experience. As Kahlow challenges her peers: If you gave your students the choice between an AI-generated explanation and one you created yourself, would they choose one, or would they choose both?
References
- Cast. (2024). CAST Universal Design for Learning Guidelines version 3.0. Retrieved from https://udlguidelines.cast.org/more/about-guidelines-3-0/
- Damm, C., & Eaton, L. (2026). From prompt to practice: A framework for transparent GenAI use in higher education. EDUCAUSE Review. Retrieved from https://er.educause.edu/articles/2026/3/from-prompt-to-practice-a-framework-for-transparent-genai-use-in-higher-education
- Google. NotebookLM. Retrieved from https://notebooklm.google.com
- Litton, E. (2021). Videos in Online Courses: Viewing Patterns and Student Performance. Journal of Effective Teaching in Higher Education, 4(3), Article 3. https://doi.org/10.36021/jethe.v4i3.247
- Mayer, R. E. (2021). Multimedia Learning (3rd ed.). Cambridge University Press.
