Oral Exams & AI: A Practical Alternative to Take-Home Writing

By /

Take-home essays used to be one of the most trusted assessment formats in higher education. Students read, they thought, they wrote. The paper reflected the process. Generative AI broke that chain. Now a student can produce a competent essay in minutes without having done any of the reading, and detection tools can’t reliably tell the difference. So what do you do if your course depends on writing as proof of learning?

Hartmann (2025) tried something different. In her paper published in College Teaching, she describes redesigning an upper-level humanities course around oral exams. And she doesn’t just swap the format. She rethinks the entire course using backward design, building every class session around the skills students will need when they sit across from her and talk through their ideas out loud.

I find this paper compelling not because oral exams are new, but because Hartmann takes the logistics seriously. Most faculty who consider oral assessment get stuck on the practical objections: too time-consuming, too stressful for students, too hard to grade fairly. Hartmann works through every one of those concerns with data, planning details, and student feedback. The result is a model that other instructors can actually use.

Why Written Assignments Are Vulnerable

Hartmann frames the problem clearly: “Generative AI threatens to render traditional take-home writing assignments obsolete while turning instructors into plagiarism detectives” (p. 2). That’s a sentence worth pondering about. The first half describes a pedagogical crisis. The second describes a professional one. Faculty didn’t sign up to police student work. And the tools designed to catch AI-generated text are unreliable enough that relying on them creates new problems, false accusations, damaged trust, wasted time.

I’ve written about this from several angles. Corbin, Bearman, Boud, and Dawson (2025) framed AI and assessment as a wicked problem, one with no clean solution, only trade-offs that require professional judgment. Perkins and Roe (2025) argued in their chapter on the future of assessment that retreating to proctored exams won’t protect validity in the long run. And the AI Assessment Scale from Perkins, Roe, and Furze (2024) offers a framework for deciding how much AI involvement is appropriate for different types of tasks.

Hartmann’s contribution fits alongside all of this, but she’s solving a very specific version of the problem. In a small upper-level humanities course built around close reading and argumentation, oral exams offer something that no written take-home assignment can guarantee anymore: evidence that the student did the thinking.

The Course Redesign

What I appreciate about this paper is that Hartmann doesn’t present oral exams as a quick fix. She redesigns the entire course around them. Students practice thinking on their feet through structured cold calling throughout the semester. They annotate readings collaboratively on a social platform, which builds familiarity with the texts before the exam. Laptops and phones are removed from the classroom to keep attention on discussion and close reading.

Every component feeds the final assessment. The oral exam doesn’t arrive as a surprise at the end. Students have been preparing for it all semester, even if they don’t always realize it.

She uses backward design here, and it shows. The exam questions go out in advance. Students can veto one question they don’t want to answer. They’re allowed a single page of handwritten notes. The sessions are short, individual, and graded immediately using a rubric. Hartmann tracked her own time and found something that will surprise a lot of skeptics:

The total time spent administering and grading oral exams for the Tibetan Buddhism Fall 2024 course was less (thirteen hours) than the total time spent grading twenty-five midterm papers and twenty-five final papers for my Tibetan Buddhism Fall 2021 course (fifteen hours, plus procrastination time).(p. 5)

Thirteen hours compared to fifteen, plus all the mental overhead of reading through papers wondering which ones were AI-assisted. That comparison alone should get more faculty thinking about this option.

Anxiety Is Real, But It’s Manageable

The biggest objection I hear from colleagues about oral exams is student anxiety. And Hartmann doesn’t dismiss it. She takes it seriously and builds her response around research on exposure and preparation. Students who practice speaking about ideas throughout the semester, through cold calling and discussion, develop the confidence they need by exam time. Predictable routines help. So do clear rubrics and the option to record sessions.

Hartmann uses a gym analogy that I think will resonate with a lot of educators: “if I want students to treat the classroom like a gym, I should be willing to be the ‘personal trainer’ who pushes them even when they do not want to be pushed” (p. 6). The discomfort is part of the learning. And the skill students build, thinking clearly under pressure and articulating ideas verbally, is exactly the kind of competency that employers say graduates lack.

Student reflections in the paper support this. Many reported that preparing for the oral exam pushed them to understand the material more deeply than they would have for a written paper. Several described relief once the conversation started, realizing they knew more than they thought. The format made their learning visible to them in a way that writing sometimes doesn’t.

Oral Exams & AI

Where Oral Exams Fit (and Where They Don’t)

Hartmann is realistic about the limits of her approach. Oral exams work well in smaller, upper-level courses where the instructor can schedule individual sessions. They complement writing instruction but don’t replace it. Students still need to learn how to write, and writing remains a core skill in the humanities. But when the goal is to assess whether students have actually done the intellectual work of reading, analyzing, and forming arguments, oral exams offer something that a take-home essay in 2025 simply cannot: certainty that the thinking belongs to the student.

And this connects to a point I’ve been making across multiple posts. The research on cognitive surrender (Shaw & Nave, 2026) shows that students tend to defer to AI outputs without critically evaluating them. Fan et al. (2025) found that AI improved essay quality but triggered metacognitive laziness, where students stop monitoring their own thinking. Oral exams short-circuit that pattern. You can’t outsource a live conversation. You either know the material or you don’t, and both you and your instructor will know which one it is within the first two minutes.

Hartmann herself frames this beautifully when she writes about the purpose of liberal arts education: “In liberal arts education, the primary objective isn’t simply to produce polished essays (we have plenty), but to build intellectual muscle—the fundamental reshaping of how students think, question, and understand the world.” (p. 2). Oral exams test that muscle directly. A polished essay might hide weak understanding but a conversation won’t.

I’m not suggesting every course should switch to oral exams. The logistics don’t scale easily to large lecture courses, and writing remains essential to academic development. But Hartmann offers a well-documented, thoughtfully designed model for courses where it does make sense. She’s done the work of tracking time, addressing common objections, and sharing student feedback. The barriers are lower than most people assume.

For faculty still stuck in the detection cycle, chasing AI-generated text with unreliable tools and losing both time and trust in the process, this paper offers a real alternative. Design your course around the thinking you want students to do. Then assess that thinking in a format AI will find hard to fake.

References

Leave a Comment

Your email address will not be published. Required fields are marked *

Scroll to Top