Artificial intelligence has quickly become part of higher education. Faculty are using AI to brainstorm activities, summarize readings, create study materials, and develop instructional resources. Along with the excitement comes a common question: When students have a choice, will they choose instructor-created materials or the AI-generated version? Rather than speculate, I decided to look at my own students’ behavior. I pulled the Canvas Studio view analytics for both resources, so I could see exactly what students opened and how much of it they actually got through. Students did not appear to view AI-generated content and instructor-created content as competing options. Instead, the majority of students used the AI-generated resource as a supplement to the instructor-created resource.
What I did
For the first module in the course, I created two optional ways for students to review the same material. This was the module overview, before students had gotten into the actual course content, for a graduate-level course on multimedia design tools. Importantly, students knew exactly what they were choosing between because I feel this transparency is important. Before accessing the content, students saw the following prompt: Choose your own adventure! Before diving into the content and materials page, review one of the following materials. The first option is my real self talking at you, and the second option is a video that NotebookLM made based on the recording of the first option. The first option was a 13-minute, 31-second podcast that I recorded myself explaining the first module. The second option was a 3-minute, 51-second AI-generated explainer video created with NotebookLM using the transcript from my podcast. NotebookLM transformed my original podcast transcript into a shorter visual explanation covering the same core concepts. Students could choose to engage with the instructor podcast, the AI-generated video, both resources, or neither. There was no penalty or reward based on their choice. My goal was not to determine whether AI could replace instructor-created content. Instead, I wanted to understand how students would use an AI-generated resource when it was offered transparently alongside a resource created by their instructor.
What I found
There were 32 students enrolled in the course. To see how students actually engaged with each resource, I pulled the Canvas Studio analytics for both the podcast and the video. 25 unique students opened the podcast 25 unique students opened the video 4 students opened only the podcast 4 students opened only the video 21 students opened both the podcast and the video 29 students opened at least one resource (out of 32 enrolled) The most common pattern was not opening the instructor-created resource or the AI-generated resource on its own. It was opening both. Of everyone who accessed either resource, roughly 84% accessed both. Among the students who accessed only one resource, there was no strong preference either way: four opened only the video, and four opened only the podcast. The AI-generated resource did not replace the instructor-created resource. The instructor-created resource did not make the AI-generated resource irrelevant. Students engaged with both. Canvas Studio also tracks completion rate, or roughly, how much of each resource a student actually got through, not just whether they opened it. This is where a real difference showed up between the two formats. Completion patterns differed between the two resources. The NotebookLM-generated video had a higher average completion rate, with students completing an average of 89.7% of the video. The median completion rate was 100%, meaning that at least half of the students who accessed the video watched the entire resource. Seventeen of the 25 students (68%) who accessed the video reached 100% completion, and 20 of 25 students (80%) completed at least 90% of the video. The instructor-created podcast had a lower average completion rate, with students completing an average of 76.1% of the recording. The median completion rate was also 100%, indicating that at least half of the students who accessed the podcast listened to the entire recording. Fifteen of the 25 students (60%) reached 100% completion, while 16 of 25 students (64%) completed at least 90% of the podcast. Although the NotebookLM video showed higher completion rates, the two resources were not identical comparisons. The AI-generated video was significantly shorter than the instructor podcast (3:51 compared to 13:31), meaning students were choosing between not only two different sources of explanation but also two distinct levels of time commitment. Both resources were, on the whole, completed by most students who opened them. But the shorter AI-generated video was finished more consistently than the longer instructor podcast. That is not a huge surprise on its own since shorter content is generally easier to finish in one sitting, but it does add nuance to the “students chose both” finding. Students were not just opening both resources; among those who chose only one, the one they were more likely to actually finish was the AI-generated video. What Surprised Me Going into this, I really did not know what to expect, which is partly why I did it. However, I did hope that at least some students would listen to my podcast even though it was longer. That said, I was happy (relieved?) 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. What surprised me was not simply that students chose the AI-generated resource. It was that students rarely treated it as a replacement for the instructor-created resource. Instead, they appeared to use it as another opportunity to process the same concepts. This suggests students may not have been asking, “Which version should I choose?” They may have been asking, “How can these different versions help me learn?” At the same time, the completion-rate gap is a reminder that “opened” is not the same as “finished.” Some students who opened the podcast page did not get all the way through it, while the shorter video was more often completed start to finish.
What This Might Mean
This was a small classroom activity, not a controlled research study, so there are many possible explanations for these choices. Perhaps students appreciated having multiple ways to encounter the same concepts. Maybe some students preferred hearing the explanation directly from their instructor, while others preferred the concise format of the AI-generated explainer video and preferred to actually finish that shorter version even when they also sampled the longer one. It is also possible that students who were especially motivated chose to reinforce their understanding by engaging with both resources. Whatever the explanation, one thing became clear, and the analytics only strengthen it: Providing an AI-generated option did not reduce engagement with instructor-created content. Access to the podcast matched access to the video almost exactly. In this case, AI appeared to function less like a replacement and more like an additional pathway for learning.
Transparency Matters
One important part of this activity was that students were knowingly interacting with AI-generated content. They knew exactly how the AI resource was created. They knew that NotebookLM had used my original recording as the source material. They knew they were comparing my explanation with an AI-generated version of that same explanation. This transparency gave students the ability to make an informed choice. Rather than asking students to accept AI-generated content without context, this activity allowed them to explore how AI could fit into their own learning process.
Important Caveats
This comparison was not a perfect test of whether students preferred AI-generated or instructor-created content. The two resources differed in several ways: The instructor podcast was longer (13:31 compared to 3:51). The formats were different (audio podcast versus video explainer). The AI-generated resource was based on the instructor-created resource rather than independently created. Completion rate as tracked by Canvas measures how far a student progressed through a page, which is a proxy for engagement, not a guarantee of attention or learning. This was a graduate-level course on multimedia design tools, not a general student population. Students in this course may be more likely than average to explore an AI tool out of professional curiosity about how it works, and more likely to complete optional materials out of habit or motivation than students in an introductory or general-education course. This activity took place in Module 1, as part of the module overview, before students had gotten into the actual course content. Their engagement may reflect early-semester motivation or curiosity about the course itself, rather than how they would behave with a similar choice later in the term.
Takeaway
This experience challenged one of the assumptions I often hear about AI in education—that AI-generated content and instructor-created content are competing with each other. My students did not behave as though they were choosing one over the other, and the Canvas analytics back that up: roughly 84% of students who opened either resource opened both. Most chose both. Most also finished the video; a bit fewer finished the whole podcast, but nearly as many opened it in the first place. To me, this shows that when AI use is transparent and students are given agency, the AI-generated resource becomes another way to engage with the material rather than a replacement for instructor-created content. For me, that is an encouraging finding. The future of AI in education may not be about replacing instructor-created materials with AI-generated alternatives. Instead, AI may expand the number of ways students can access, review, and engage with course concepts. If you gave your students the choice between an AI-generated explanation and one you created yourself, what do you think they would choose? More importantly, would they choose one, or would they choose both?
References
Mayer, R. E. (2021). Multimedia Learning (3rd ed.). Cambridge University Press. Google.
NotebookLM. 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
Damm, C., & Eaton, L. (2026). From prompt to practice: A framework for transparent GenAI use in higher education.
EDUCAUSE Review. https://er.educause.edu/articles/2026/3/from-prompt-to-practice-a-framework-for-transparent-genai-use-in-higher-education
CAST (2024). CAST Universal Design for Learning Guidelines version 3.0. https://udlguidelines.cast.org/more/about-guidelines-3-0/
Jess Kahlow is an Adjunct Assistant Professor in the College of Education at the University of Texas at Arlington, where she teaches in the Department of Educational Leadership and Policy Studies and the Higher Education Adult Learning and Organizational Studies. Her work focuses on instructional design, educational technology, and the thoughtful integration of emerging technologies, including artificial intelligence, into higher education teaching and learning.
