What Is an AI Oral Exam?

What an AI oral exam is#

An oral exam is an assessment in which a student answers questions in conversation. The examiner can ask what the student meant, ask for an example or ask why a different answer would be wrong. Those follow-up questions are the reason oral exams reveal understanding that a written answer can hide.

An AI oral exam keeps that structure and uses an AI system as the examiner. The student speaks or types. The AI asks the first question, listens, and chooses the next question based on what the student actually said. The session produces a record, usually a transcript and sometimes a recording, and a draft score against criteria the instructor set. A person reviews the record and decides the grade.

An AI oral exam is a different thing from a multiple-choice quiz with a voice, from software that guesses whether text was written by AI, and from proctoring. It is a conversation designed to collect evidence of what the student understands.

Why oral exams are returning#

Oral examination is old. Joughin's analysis of the format identified dimensions along which oral assessments differ, including the content being assessed, the degree of interaction, authenticity and structure. Oral exams persisted in some fields, such as doctoral defenses and professional certification, and became rare in large undergraduate courses because they take examiner time.

Two changes brought them back into discussion. Generative AI made many written submissions weaker evidence of individual understanding. And AI made it possible to conduct a structured conversation with every student without scheduling a human for each. Akimov and Malin reported a case study of oral examination used as an online assessment tool. Sotiriadou and colleagues examined authentic assessment including interactive oral assessment as a way to uphold academic integrity and build employability skills.

How a typical AI oral exam works#

1. The instructor sets it up. They choose what is being assessed (a submitted paper, a presentation, a reading or the course content in general), write or approve the criteria, and set how long the session may last and how students may respond.
2. The system prepares questions. For a defense of submitted work, the AI reads the student's submission and drafts questions that refer to its claims, sources and choices. For a content exam, questions come from the material the instructor provides.
3. The student takes part. In a live format the student speaks with the AI in real time. In a recorded format the student answers each question in turn. In a text format the student types.
4. The AI follows up. When an answer is thin or vague, it asks for an example, a reason or a comparison. This is the part that distinguishes a conversation from a form.
5. The system keeps a record. A transcript, and where the assignment captures it, a recording.
6. The AI drafts scores. Each criterion gets a draft score with the evidence it rests on.
7. The instructor reviews. They read or listen to the evidence, adjust scores and decide what is released.

Common formats#

• Defense of submitted work. A student submits a paper, code or slide deck and answers questions about it.
• Oral examination on course content. A student answers questions about a unit or a reading without a submitted artifact.
• Cumulative or capstone examination. Questions draw on several sources across a course.
• Interactive scenario. The AI plays a counterpart, such as a client, and the student responds in role. This overlaps with AI role-play in education.
• Makeup assessment. An alternative route for a missed activity.

Strengths#

• Follow-up questions probe understanding in a way static questions cannot.
• A consistent examiner can talk with every student in a large course.
• The record gives you evidence for grading, review and appeals.
• Students can practice before the graded attempt.
• Text and spoken options can support accommodations.

Limits and risks#

An honest description of the format includes its weaknesses.

• Question quality varies. The AI may ask a question that is unclear or that misreads the submission. Instructors should preview questions where they can and review the transcript.
• Fluency can distort scores. A student who is anxious or who is still learning English may sound less sure than they are. Criteria should score reasoning, and a text option should be available.
• Technology can fail. Microphones, bandwidth and browsers vary. A fallback matters.
• Privacy applies. Recordings and transcripts are education records. Check where they are stored, how long they are kept and whether a vendor uses them to train models.
• The AI should not decide. A draft score is an input. A person should make the final judgment.
• It is not a substitute for every kind of assessment. Some outcomes need a written artifact or a supervised written exam.

Designing a fair AI oral exam#

• State the criteria and share them with students before the attempt.
• Keep sessions short enough to stay focused. Many tools document sessions between about six and twenty minutes.
• Offer a practice attempt that does not count.
• Provide accommodations: extended time, a text option and an alternative for students who cannot use the format.
• Preview or sample the questions, and read a sample of transcripts yourself.
• Decide in advance how you will handle a technical failure and an appeal.
• Keep a person in the loop for every grade.

Our oral assessment rubric guide gives criteria you can adapt, and the comparison of AI oral assessment platforms lists questions to put to vendors.

If you are planning a first oral exam, our blog playbook on how to run oral exams in college courses gives a practical sequence.

Questions to ask any vendor#

1. Who conducts the session: an AI, a person or both?
2. Where do the questions come from, and can I review them?
3. What evidence will I have after the session?
4. How are scores produced, and can I change them?
5. How do accommodations work?
6. How does the grade reach my LMS?
7. Where is data stored, for how long, and is it used to train models?

Where Rocketproof.ai fits#

Rocketproof.ai is one platform that runs AI oral exams. It builds questions from a student's own submission or from source material the instructor provides, supports spoken, live and typed responses, and keeps a transcript and rubric evidence for instructor review. The oral exams page describes the workflows, and the scalable oral exams guide covers running them in large classes. Several other products take different approaches, which the comparison guide describes.