How to Run Oral Exams in College Courses

College oral exams can be run at scale by choosing the right response format, grounding questions in student work, and using instructor controlled rubrics. This playbook is for faculty and teaching center staff who want to pilot an oral exam in one course without adding a week of scheduling. It covers the three response formats Rocketproof supports, how to write prompts from course artifacts, how to grade fairly, and what to check before you launch.

Choosing among recorded audio, text chat, and live voice

Rocketproof calls the student-facing format the response engine, and you pick one per assignment. All three are conducted by the AI, so none of them needs you in the room. They differ in how the student experiences the exam and in what support they need.

• Recorded audio (shown to students as Spoken Response). The student reads each question as text, then answers by speaking. It needs a microphone, gives the student time to think before each answer, and is the default for a new assignment. Consider it when you want spoken explanation without real-time pressure.
• Text chat (Text Conversation). The student exchanges typed messages with the AI. It needs no microphone or camera. Consider it when students may be uncomfortable speaking, bandwidth is limited, or speaking is not a learning outcome.
• Live voice (Live Conversation). The student talks with the AI, which speaks back in real time. It needs a microphone and a steady connection. Consider it when thinking aloud, answering follow-ups on the spot, or professional conversation is the skill you are assessing.

A few practical points follow.

• Match the format to the outcome. If you are checking whether a student can explain a method, recorded audio or text chat is usually enough. Reserve live voice for outcomes where real-time conversation is the point.
• Large enrollment courses are usually easiest with recorded audio or text chat. Both let students work inside a window you set, and neither depends on a live connection holding up for the whole conversation. Live voice also runs without you, but it asks more of each student's device and network, so plan more support time for it.
• Offer text chat as a route, not a penalty. Rocketproof can offer a typed option as an accommodation. Where speaking is not a stated learning outcome, treat typed answers as equal evidence.
• Do not mix formats inside one rubric without thinking it through. A student who types gives you no delivery to observe, so score the content of the answer, not how it was said.

Which engines a given assignment type supports is shown in the assignment form. Check it there rather than assuming every combination is available.

Designing prompts tied to course artifacts

The strongest oral exams start from something the student made. An artifact can be a paper, a lab report, a slide deck, code, a case analysis, or a reading response. Questions drawn from it are hard to answer from memory or from generated text, because they ask about this student's choices.

For upload-based assignments, Rocketproof reads the submitted document and prepares a questioning plan tied to its specific arguments, methods and sources. You can review the plan and add your own questions before students begin. For a reading assessment, you supply the source or the topics.

A good prompt does one of four jobs:

1. Explain a decision. "Why did you choose this method over the alternative you mention in section 2?"
2. Test an assumption. "Which assumption in your estimate would change your recommendation most if it were wrong?"
3. Apply to a new constraint. "A stakeholder tells you the budget is cut by a third. What changes?"
4. Surface limits. "What is the weakest part of this argument, and how would you strengthen it?"

Illustrative examples by discipline follow. They show the shape of a prompt and are not a tested question bank. Adapt them to your course.

• Nursing: "Walk me through why you prioritized the second intervention in your care plan. What finding would make you reverse that order?"
• Computer science: "In your submitted code, what happens if the input list is empty? Show me where you handle it, or what you would change."
• Political science: "Your memo recommends option B. Respond to a colleague who says option A serves rural districts better."
• History: "You rely heavily on one primary source. What are its limits, and what would a second source add?"
• Business: "Your forecast assumes a 5 percent churn rate. Which input is the least certain, and why?"

Keep the anchor questions the same for every student so the exam is comparable, and let the follow-ups branch from what each student actually says. If your prompts only work when the student has memorized a definition, rewrite them so they need the student's own work.

For a worked example of grounding questions in a submission, see Alternatives to AI Detection for College Assessments. For the product side, see AI oral exams and oral defense.

Setting rubrics instructors control

The rubric is where faculty judgment stays central. You write it, students see it in advance, and you decide what is released. A starting set of criteria might be:

1. Accuracy and use of course concepts
2. Justification of decisions
3. Response to follow-up questions
4. Awareness of limits and trade-offs

Write level descriptors that fit your discipline. Keep delivery out of the score unless speaking is a stated outcome, because accent, pace and hesitation are not evidence of understanding.

On the control question, the rules are simple:

• You write the rubric and can add your own questions.
• AI scores are drafts. Rocketproof prepares rubric-aligned draft scores with evidence from the transcript. You review them and change any you disagree with.
• Nothing is final until you release it. Rocketproof does not issue final grades on its own.
• Integrity concerns follow your institution's process. A transcript is one piece of evidence, not a finding.

Rocketproof is an authentic assessment, simulation and course intelligence platform. It is not an AI detector, and it does not proctor or judge whether a student used AI.

Scheduling and workload planning

The question most faculty ask first is how much time an oral exam takes per student. There are two separate numbers.

• Student time. This is whatever you set. A focused ten-minute conversation often gives enough evidence for one learning outcome. Longer exams are possible, but each added minute is a minute you will also review.
• Your review time. This depends on your rubric, the length of the conversation and how closely you read. We do not publish a figure here because it varies by course. Time yourself on five or six sessions in your pilot, then multiply by your enrollment.

The arithmetic for live, instructor-run orals shows why format matters. Sixty students at ten minutes each is ten hours of conversation before scheduling gaps, late starts and make-up slots. With AI-facilitated formats, students complete the conversation inside a window you set, and your time moves from conducting conversations to reviewing them.

Some scheduling choices that keep the load manageable:

• Open a multi-day window and let students pick their own time.
• Use the exam on the assignment that matters most, not on every assignment.
• Set accommodation time in advance. Extended time can be set per student or for a whole course.
• Reserve a make-up path for students who miss the window or hit a technical problem.

Scheduling models for very large courses vary by institution, so check what your teaching center already uses for makeup exams and late work.

Preparing students for conversational assessment

Students do better, and support requests drop, when the format is no surprise. Before the graded window:

1. Say why. Explain that the conversation checks understanding of their own work and is not an accusation.
2. Publish the rubric and the topics or artifact the questions will draw on.
3. Run a practice round. Rocketproof can offer a practice run that stays out of grades, and an assignment can be marked practice so it is excluded from the gradebook and from grade passback.
4. Check the device. For recorded audio and live voice, ask students to test their microphone and connection ahead of time.
5. Explain the AI's role. The AI asks questions and follows up on what the student says. It does not decide the grade.
6. Tell students about the alternatives. If typed responses or extra time are available as accommodations, say how to request them.

On accessibility, Rocketproof's assignment settings include an audio-only option, a typed response option that needs no camera or microphone, extended time, a pause control during recording, and a caption-first review view. Whether these meet a particular institutional or legal accessibility standard is a question for your disability services office, so involve them before the pilot. This article makes no claim of conformance with any specific standard.

Reviewing evidence and finalizing grades

After students finish, the work moves to you.

1. Read the transcript and, if you wish, listen to the audio for moments that matter.
2. Check the draft rubric scores against the evidence cited from the transcript.
3. Change anything you disagree with and add your own feedback.
4. Release the grade. Where LTI grade passback is configured, a released grade can pass to the LMS gradebook.

If a conversation raises a concern, treat it as a reason to follow your institution's normal procedure, not as an automatic finding. A thin answer can reflect anxiety, a bad connection or a poor question as easily as a gap in understanding.

For Canvas, Rocketproof supports LTI 1.3 launch from the course and grade passback, and Canvas is the platform verified end to end. Blackboard, Moodle and D2L Brightspace are supported on the same LTI 1.3 standard, but a launch has not yet been observed on a vendor test system, so check with us before planning a term around them. Students can also use a direct assignment link without an LMS. See the LMS integration page for current status by platform.

Pilot checklist

Start with one course and one assignment.

1. Pick the assignment. Choose a project where explaining the work is part of the learning outcome.
2. Check your policy. Confirm what your institution allows, and loop in your integrity office and disability services.
3. Choose the format. Pick recorded audio, text chat or live voice for the outcome you are assessing.
4. Define what you will assess. Write two to four outcomes the conversation must show.
5. Draft and publish the rubric.
6. Review the question plan and add your own questions.
7. Run a practice round with students.
8. Set the accommodation path for typed responses and extended time.
9. Decide how grades reach students, with Canvas passback or another route.
10. Time your review, gather student feedback and adjust before expanding.

For more on scaling the format, see how to run oral exams for large classes. The conversational format also supports AI role-play simulations, and the broader approach is on the authentic assessment page.

Frequently Asked Questions

How much time does an oral exam take per student?

It depends on two things you control. Student time is the length you set, and ten focused minutes is a common starting point. Your review time depends on your rubric and how closely you read, so time five or six sessions in your pilot and multiply. AI-facilitated formats remove the scheduling and conducting time that makes live orals expensive.

Which format fits large enrollment courses?

Recorded audio and text chat are usually the easiest to run at scale, because students answer inside a window you set and neither needs a live exchange to hold up for the whole exam. Live voice also runs without you, but it asks more of each student's device and connection, so plan more support time for it.

How should prompts connect to assignments?

Draw them from the student's own submission, such as a paper, code, slides or a case analysis, and ask students to explain a decision, test an assumption, apply the work to a new constraint or name its limits. Rocketproof can prepare a questioning plan from an uploaded document, and you can review it and add your own questions.

How are oral exams graded fairly?

Publish the rubric in advance, score what you can observe in the answer, and keep delivery out of the score unless speaking is a stated outcome. Rocketproof drafts rubric-aligned scores with transcript evidence, and the instructor reviews, changes and releases them. Offering text responses and extended time as accommodations also helps keep the exam fair.

How do students prepare for an AI facilitated oral?

Tell them why the exam exists, share the rubric and the artifact the questions will draw on, and offer a practice round that does not count toward their grade. Ask them to test their microphone and connection beforehand, and explain how to request typed responses or extra time.

See how oral exams can work in your course format

Request a demo to see how oral exams can work in your course format, including recorded audio, text chat and live voice, along with the transcript and instructor rubric review. Request a demo, or explore the public demo on your own.