Case study: Creating a podcast using Gen AI to support student learning

Department of Chemistry
David Smith

David Smith describes how he used Gemini Notebook to create a podcast to support student learning of organic reaction mechanisms, uploading his own article on the topic as source material. He spent between 1 and 2 hours reworking and editing the podcast before releasing it to students as an optional additional resource ideal for consolidation week on a year 1 Undergraduate module.

Reflecting on his experiences from the teaching, David notes that the quality of the podcast exceeded all his initial expectations but that some issues remain, not least in the inability to change the voices to vary podcasts and increase diversity/inclusivity and the limits to the available options to edit the podcast contents after production.

From the student perspective, David reports on the results of a survey he distributed to obtain feedback which suggested that attitudes to the use of Gen AI in education in general had a bearing on attitudes towards the podcast with those more positive about the former also being more positive about the latter. Those who listened to the podcast tended to be positive about its contents and its educational value, while those who did not tended to report that they were not in favour of uses of AI in Higher Education.

Watch their presentation:

AI Podcasts to support student learning (Panopto viewer) (9 mins 55 secs, UoY log-in required)

Transcript

So it was about just over a year ago that I was sitting in a university session run by Google, and they were demonstrating the podcast capacities of what was then called Google Notebook, which is now Gemini Notebook. And this is the ability to create a two voice podcast from some source material that you create. And they demonstrated it with some history material. And I thought, wow, that's impressive. There is no way that will work for chemistry. There is not a chance that with highly technical chemical material, it will work. And at the time I was teaching a first year course and there was a really good test case for potentially creating a podcast that might support student learning in that course. So I thought I'd have a go.

So in particular, I teach an area of chemistry which I've developed a specific approach to teaching, which we only use here in York, to teach students how to draw reaction mechanisms. And this means that students are a little bit lacking in additional resources in this area. If they go and look in a textbook, it won't be taught quite the same way that I explain it in my lectures. And so I wanted to create some extra resources that they could engage with in different ways that would allow them to hear and learn about this approach to mechanisms. And I thought a simple podcast that they could listen to while they were cooking their tea or commuting to campus might be quite an interesting resource for them to engage with.

So the way you go about doing this is you feed your source material into the Google notebook. And because I'd written a paper explaining exactly this approach to teaching mechanisms, that was the ideal source material to feed to Google Notebook to prime things with. So I fed that paper into Gemini Notebook. And when you do that, you can do a variety of things, as I'm sure many of you know. And one of the things that you can do is you can create audio output, and in particular to create a two voice podcast, you want to create a deep dive between the two voices. You get some options in terms of how long you want it to be. I went for about 15 minutes for the deep dive that the voices were doing. There was a reasonable amount of material that they were going to be talking through and presenting, and I thought that was a reasonable length for students to engage with while travelling or doing another task. So you feed it in and within five minutes you have a podcast basically, with two voices. So you've done the first step of this process. You've generated that initial output. And what I did is I listened to it actually, after listening to a bit of it, I created a transcript and skimmed through the transcript. And it's a bit quicker than listening all the way through the 15 minutes. And I was surprised. It was pretty good. There were some choices I wouldn't have made in terms of how things were explained, and there were a few technical errors that the podcast made in pronunciations of technical words, things like that. A few little slips and it occasionally got distracted on what was slightly less important bits of my source material. So I actually went back in and edited my source material and threw out some bits. I think I would have written a much better paper on how to teach this if I'd listened to this AI podcast in the first place, because it showed me what was distracting. So I threw out the bits that were a bit irrelevant, used the modified source, and created a couple more inputs. And once I'd done those and looked at the transcripts, I was able to essentially combine the best bits of those three different outputs into a single, coherent podcast. That process took me about 1 to 2 hours, I would say, but less time by far than it would have taken me to find another academic, decide what we wanted to talk about, chat the thing through, get the kind of polish on it, edit it all up. Would have taken a lot Longer. Within a day, so I started this in the morning, and by the end of the afternoon - And I didn't work on it all day - I had a podcast that was subtitled and could be launched onto the VLE.

Um, so I created the podcast. I also put one image in there to help the students with a bit of the technical description. And I'll have a go at playing a bit of the audio here. And you'll hear two voices. You don't have any options at the moment to change the voices that you hear. They're sort of generic American white voices.

"Let's talk about organic chemistry or chem 101. If you've been through it you probably remember that feeling. Maybe dread.

Oh, definitely. It often feels less like learning and more like surviving a massive info dump.

Exactly like this huge mountain of reactions, weird named mechanisms, endless facts, all just stuff to cram in, endure. And then let's be honest, forget immediately after the exam.

That's such a common experience. I remember students just trying to make the arrows work to get the product, not really thinking about why the electrons move that way.

Yeah, like reverse engineering it without understanding the cause and effect.

Precisely. And, you know, the way it's traditionally taught kind of encourages that you race through so many different reaction types without really nailing down the fundamentals first.

I definitely remember that feeling like trying to memorise a cookbook with a thousand slightly different recipes. The sheer volume of stuff can push actual problem solving aside. Instead of a library of facts to memorise, What if organic chemistry could be more like an operating system? Something logical? That's the idea behind the source material we're looking at today. It proposes this really elegant alternative a simple set of priority rules.

Okay. Priority rules..."

So I'll leave that there. Um, it shows you the kind of thing that you can do. So you can see the two voices talk very naturally. They bounce things off one another. None of those things or lots of those things are not really written explicitly in my original script. That's what the AI has come up with, and it does a pretty good job of presenting the more technical material as well.

Um, so having done that, the question is really what did the students think? There was a 15 minute podcast. I released it to the students. They could use it in Consolidation week to help them think through the material before going into their tutorial that came after consolidation week, where they were going to solve some problems using these priority rules. And so I asked the students what they thought about it. I did declare that AI had been used to make the podcast because I thought that was only fair. So as is always the case, not every student responded to the survey, but I used a specific survey tool and asked the students, um, these were responses of students who had listened to the podcast and they felt it was scientifically accurate, they felt it reinforced their understanding, and they wanted more podcasts to be available to support their learning. They said things like "it was a really helpful tool to review the topic", "help them to feel more confident", "I wish there could be a podcast series for all the lectures", "I felt like I was listening to an actual podcast", "It was interesting how it was asking questions that students would have doubts on, and then explaining them really well". There was then a second group of students who didn't listen to the podcast, and I asked lots of reasons why they might not have engaged with it, and the only one that got any traction was they didn't listen to the podcast because they didn't agree with the use of AI, and the kind of responses that the students gave were things like this. "In an age of trying to be climate conscious, I do not think it has a place in academic spaces", "for a university which preaches sustainability, encouraging its use is irresponsible", "not keen on generative AI being used to assert the actual human knowledge of professors", "Unsure whether a machine that blatantly plagiarised human work while hallucinating incorrect information holds any place in an academic institution" and one student simply responded "I hate AI". Um. Interestingly, I also asked the students whether they used AI in their own studies, and we found that students were more likely to have engaged with the podcast if they themselves used AI, and the ones who didn't engage with the podcast were less likely to be using AI to support their own studies. And there is a clear divide between our students. I'm not saying that the numbers are reflective of the total numbers. I suspect there was a vocal minority who really hooked on to this survey and wanted to make their opinions felt, but they certainly did make their opinions felt and in the module feedback.

So my reflections, my concerns are the ethics of AI generated outputs, sustainability and student opinion, the limitations with the fixed non diverse voices, and the lack of editing capacity, you can't edit out individual words or individual sentences. You have to kind of go in and use an editing tool and, you know, you can't do it in a textual way. You can't delete something in the transcript, and it disappears from the podcast, and you can't incorporate images. And I'm sure those things will come into future versions of Gemini podcasts. And some different software's already solved some of those problems. There's lots of potential. Of course, we've already seen Caroline showing how it can be used creatively to get students to engage with these things. There's another paper been published by chemists elsewhere who've created podcasts as pre lecture material, and students can either read through the notes before going into the lab, or listen to the podcast before doing the little quiz before starting in the lab. And they found that it was a very quick way of producing audio material to support an entire lab course. And that paper has just come out in Journal of Chemical Education. So people are thinking about using podcasting in lots of different ways, either for content but also more creatively, getting students to perhaps create and error check their own podcasts. And that's it for me.

You can also read David's article reporting on the experience: Creating a Podcast Using Generative Artificial Intelligence to Support Student Learning of Organic Reaction Mechanisms

Reference:
Smith, D. K. (2026) Creating a Podcast Using Generative Artificial Intelligence to Support Student Learning of Organic Reaction Mechanisms, Journal of Chemical Education. https://doi.org/10.1021/acs.jchemed.5c01652