Exam prep · Updated 16 September 2026

Oral Exams Are Back: How to Prepare With AI (Safely)

Your course replaced a take-home essay with a twenty-minute conversation, and nobody explained how to study for that. The research has a surprisingly encouraging answer — and it is the opposite of what most students do with a chatbot.

Prepare for an oral exam by talking, not by reading. The strongest finding in the research is that students underperform in orals mainly because the format is unfamiliar, not because they know less — and most students actually score higher speaking than writing. So the job is to make the experience ordinary before exam day: get the format details from your instructor, then rehearse out loud against something that asks unscripted follow-up questions. A chatbot is very good at being that something. It is very bad at the thing students usually ask it for, which is a script.

Why this is suddenly happening to you

Written work stopped being evidence. In March 2026 reporting for the Associated Press, Jocelyn Gecker described the pattern that pushed faculty over the edge: assignments coming back flawless from students who could not discuss them. Bruce Lenthall, who directs Penn's Center for Teaching and Learning, reported "a massive shift toward in-person assessments." Cornell's Chris Schaffer added twenty-minute Socratic defences after written problem sets. One religious studies course swapped its final exam for a thirty-minute "final conversation"; an engineering professor now runs four-minute interviews with every student in a class of 180.

Institutions are writing it into policy rather than improvising. The University of Sydney's two-lane assessment framework splits everything into secured and open work, and its list of secured tasks names "in person interactive oral assessments; viva voces; contemporaneous in-class assessments and skill development; tests and exams." Sydney is explicit about why it did not simply ban AI instead: "it is already not possible to restrict AI use in assessments which are not secured," and "humans cannot accurately detect good AI use." The oral is what you get when a university concludes that detection does not work.

The research says you will probably do better than you fear

This is the part no one tells students, and it comes from the best available evidence. Nallaya, Gentili, Weeks and Baldock published a PRISMA systematic review of oral assessment in Issues in Educational Research in 2024, screening 2,199 records down to 368 full-text reviews and finally 17 studies that met the inclusion criteria.

Their headline for anxious readers: "the analysis suggests that most students performed better in oral assessments than they did in written assessments." Speaking is not the harder format. It is the unfamiliar one.

And the review is specific about where the fear comes from. Students "were anxious about oral assessments as they were not familiar with this type of assessment," having "previously experienced only paper-based tests." Elsewhere: "since students in some educational settings had not experienced or trained for the oral assessment, they did not know what to expect, thus contributing to heightened anxiety."

The script is the failure mode

Here is where most students use AI in exactly the wrong direction. Asking a chatbot to predict the questions and draft good answers feels like efficient preparation. It produces the precise failure the professors are describing: polished sentences with nothing behind them, which survive until the first follow-up.

The format is designed to defeat this. The University of Auckland's teaching guidance defines interactive orals as "brief, unscripted one-to-one conversations simulating real-world scenarios." Writing in Times Higher Education in August 2025, Popi Sotiriadou of Griffith University and Dani Logan-Fleming of Torrens University are blunter: "Not all oral assessments qualify as IOs. Presentations, viva voce examinations or rehearsed monologues do not deliver the same personalised, unscripted experience." They advise colleagues to tell students plainly that orals "demand flexible thinking, not memorisation of pre-prepared answers," because "success comes from true engagement and critical thought, making reliance on external scripts or AI-generated content futile."

A memorised answer is also the most brittle thing you can walk in with. If you rehearse one path through a topic and the examiner enters from a different angle, you have nothing — whereas a topic you can explain three ways survives any opening question. That is why every prompt below is a questioner, and none of them writes an answer.

Ask for the four things you are supposed to have

The single highest-value thing you can do this week takes five minutes and is not studying. The Nallaya review found the reliability of orals depends on design choices that are meant to be shared with you: "the measures taken to enhance reliability of the oral assessment included predetermining the allocated time, asking each student the same question, and establishing assessment criteria prior to the assessment." On rubrics: they "should also be presented to the students prior to the assessment so that they know what aspects will be assessed during the assessment and can prepare appropriately." Assessors, it adds, "need to ensure that the topics being tested in the assessment have been covered in class." That these are stated as recommendations tells you how often they are missed: one included study found students "may not be clear about the assessed criteria or have the necessary experience or skills to adequately prepare."

So email your instructor and ask for four things: how long it runs, roughly how many questions, what you may bring into the room, and the rubric or criteria. You are not asking for a favour or a hint. You are asking for the conditions the literature says a fair oral is built on, and instructors usually have all four written down already.

The fairness problem, stated honestly

Orals are not a neutral upgrade, and you deserve the unflattering half. Writing in EducationHQ in August 2026, Nicole Brownlie of the University of Southern Queensland warns that when an oral is used mainly as an "AI lie detector", "it can turn assessment into a test of confidence, rather than learning or effort." She cites German research finding students reported more anxiety in orals than in written exams, and Norwegian work on consistency in which presentations ran anywhere from 5 to 16 minutes, follow-up discussions from 7 to 23, and some students faced fewer than ten questions while others faced nearly fifty. That last study was in secondary schools, so treat it as an illustration of how much informal variation is possible rather than a measurement of your course.

The Nallaya review reaches the same place from the data: "confident and extroverted students will perform better than shy and introverted students," and this assessment type "can disadvantage students from EAL backgrounds." None of that means the format is illegitimate — it means rehearsal is levelling rather than vanity, and that if you have a disability, a speech difference, or significant anxiety, you should raise it early. The University of Guelph's guidance to its own staff tells them to "decide what alternative assessment options will be available for students who may be disadvantaged by, or less comfortable with, oral assessment (e.g., students with hearing or speech impairments, anxiety, non-native speakers)." Departments are expected to have thought about this. Ask what they decided.

Split the job

The workflow, step by step

Four prompts. Paste them into ChatGPT, Claude, Gemini or whatever you already use. Speak your answers out loud even if you then type them — the point is to practise producing sentences, not to produce text. Dictation makes this easier and closer to the real thing.

1. The unscripted examiner. The core drill. Run it until being questioned feels boring.

You are examining me in a [length]-minute oral exam on [topic].

Rules:
- Ask ONE question, then stop and wait for my answer.
- After every answer I give, ask a follow-up about something I
  actually said - not the next question on your list.
- Never give me a model answer, a correction, or the right answer.
- If I am vague, say "that was vague" and ask me to be specific.
- If I use a term without explaining it, ask me what it means.
- Do not tell me whether I am doing well.

Start with an easy opening question. Get harder as we go.
Stop after [length] minutes and only then tell me which two
answers were weakest and why.

2. The follow-up generator. Students prepare for "what is X". Examiners ask the second question.

Here are the topics on my oral exam:

[list them]

Generate 20 FOLLOW-UP questions only - the kind asked after a
student has already given a textbook definition. Use forms like:
- Why does that work?
- What would change if [condition] were different?
- How do you know that, rather than the opposite?
- Where does that explanation break down?
- Can you give an example that is not the one in the textbook?

Do NOT answer any of them. Do NOT include definition questions.

3. The borrowed-vocabulary audit. This is the prompt that finds the cracks before an examiner does.

Below is a transcript of me explaining [topic] out loud from
memory, including the messy bits:

[paste or dictate it]

Do NOT rewrite it, improve it, or explain the topic to me.

1. List every technical term I used but never explained.
2. List every claim I asserted without saying why it is true.
3. List anything I said that sounded confident but was actually
   circular or empty.

Then ask me about item 1 on each list, one at a time, and wait.

Terms you cannot unpack are exactly what a follow-up question targets. Better to find them at your desk.

4. The dress rehearsal. Only useful once you have the real details.

My oral exam is [length] minutes, roughly [N] questions, on
[topics]. I [may / may not] bring notes. The rubric says:

[paste the criteria you were given]

Run it exactly to those constraints, in real time. Do not invent
details I did not give you - if something is missing, tell me what
to go and ask my instructor instead of guessing.

Afterwards, score me against that rubric only, quoting the
criterion for each judgement. Then tell me the one topic I should
spend tomorrow on.

Where the line is

Rehearsing with a chatbot before a secured oral is ordinary studying, in the same category as being quizzed by a friend — the assessment is secured at the moment you speak, so there is no question about who is answering. The essay-era worry does not transfer. Two caveats apply. Sydney's framework notes that AI use inside a secured assessment "may or may not be allowed by examiner", so the rules for the room itself are your instructor's to set. And memorised AI-written answers are the one preparation style that is both risky and ineffective — Brownlie names it as a live concern about orals, and the unscripted format defeats it anyway. If your syllabus wording is unclear, our class AI policy checklist and guide to homework help without cheating cover how to read it and how to ask in writing.

Related reading

• The rest of your exam prep. The practice-test and flashcard system and the seven-day exam plan, both of which end in retrieval practice rather than re-reading.
• Speaking under pressure, elsewhere. Presentation coaching and behavioural interview practice, which train the same muscle for a different room.
• Why assessment changed. Agentic browsers and coursework, and what to do if you are falsely accused.
• Subjects where narration is the whole skill. Multivariable calculus and law school case briefs.
• Before you paste anything. The privacy checklist for class materials.

FAQ

Are oral exams just a way to catch students using AI?

That is how they are often reported, but it is not how they are supposed to be used. The University of Guelph tells its own staff that oral assessment should not be used "solely for preventing academic dishonesty or to proctor students." Reading it as an interrogation also makes you worse at it, because you prepare to defend rather than to explain. Prepare as though the examiner wants to hear you think, which is usually true.

Should I memorise answers before an oral exam?

No, and this is the most common mistake. Interactive orals are defined as "unscripted" conversations, and Sotiriadou and Logan-Fleming advise instructors to tell students the format demands "flexible thinking, not memorisation of pre-prepared answers." A memorised answer also breaks the moment an examiner enters the topic from an unexpected angle. Aim to explain each topic three different ways instead of one way perfectly.

Is it cheating to practise for an oral exam with a chatbot?

Practising is ordinary studying — the assessment is secured at the point you speak, so nothing about your preparation changes who gives the answers. Being quizzed, having a definition challenged, or asking for harder follow-up questions is no different from a study partner doing it. What does carry risk is memorising AI-written answers to recite. Check your course policy for rules about the room itself, since the University of Sydney notes AI use in a secured assessment "may or may not be allowed by examiner."

I get badly anxious speaking, or English is not my first language. Is this unfair?

The concern is real and documented rather than something you have imagined. The 2024 systematic review found that "confident and extroverted students will perform better than shy and introverted students" and that orals "can disadvantage students from EAL backgrounds." Two responses help. Rehearsal directly attacks the unfamiliarity the same review identifies as the main source of anxiety. And your department is expected to have considered alternatives for students disadvantaged by the format, so contact your instructor or disability service early rather than on the day.

How long is an oral exam and what can I bring?

There is no standard, which is exactly why you should ask. Reported formats range from four-minute interviews to thirty-minute conversations, and research on consistency has found large variation in both length and number of questions even within one setting. Rules about materials vary too — some orals allow a page of notes or a diagram, some allow nothing. Ask for the length, the approximate number of questions, what materials are permitted, and the rubric.

Bottom line

An oral exam is not a written exam read aloud, and it is not a lie detector. It is a conversation in which you do not control the order, cannot revise, and will be asked a second question about whatever you just said. The evidence says that is a format most students turn out to be good at once it stops being strange. So spend this week making it unstrange: get the four format details from your instructor, then talk through your material against something that keeps asking why. Use the chatbot as the examiner, never as the scriptwriter.