How to Use AI as an Oral Examiner Without Being Fed the Answer
Make the tool wait long enough for you to find out what you can explain
The question appears. Before you’ve formed an answer, the AI has supplied the explanation, the key points and a reassuring summary. You nod along. Your oral exam practice has quietly become another reading session.
The first instruction for an AI oral examiner should be to wait. Ask one question, stop and let you attempt an answer. That pause is the part where you discover what you can produce without seeing the expected response.
Choose a small objective and supply the relevant teaching material. If your faculty provides an answer framework or marking rubric, include it. Otherwise, ask for feedback on the elements supported by the source and avoid treating an invented score as an official assessment.
You can begin with this instruction:
Ask me one oral question at a time about this objective, using the material provided. Wait for my answer. Do not include the answer in the question or offer hints unless I request one. After my attempt, identify supported points, omissions and errors with source locations. Ask one follow-up about my weakest point, then wait again.
Set a rough answer length appropriate to the task. A definition might need a few sentences. Explaining a sequence will take longer. If every question becomes an uninterrupted five-minute speech, it gets harder to identify which part of the answer was actually uncertain.
Get Mohamed Jarray’s stories in your inbox
Join Medium for free to get updates from this writer.
You can type your responses or speak them and keep a brief record. The useful feature is committing to an explanation before receiving the correction, not the particular interface you use.
When you get stuck, ask for a limited hint. For example, request the category of the missing step or a reminder of the question’s scope. Keep track of whether you needed that help. An answer produced after several prompts isn’t the same performance as an answer produced independently.
Sometimes the best response is, “I can’t explain that yet.” Let the gap stay visible. Ask for the relevant source passage, work through it and return to the question later. Filling the silence with loosely related facts can hide the specific point you need to learn.
Feedback needs checking too. If the AI says you omitted something, locate it in the supplied material or rubric. If it penalises a different but accurate wording, ask which necessary concept is missing. Fluency and similarity to the model answer should not become substitutes for correctness.
You can make follow-up questions more useful by connecting them to your response. If you named the steps but couldn’t explain a connection, ask about that connection. If you stated a rule too broadly, ask when it applies. Avoid a stream of unrelated questions chosen merely to keep the chat moving.
After a few questions, close the conversation and record the result in ordinary language: “Could define the terms. Needed a hint for the mechanism. Didn’t recall the exception.” This is more actionable than a cheerful overall rating.
Then check one of those weak points in a later session, ideally with different wording. Otherwise, you may remember the sequence of the conversation more readily than the concept itself.
For actual viva preparation, include practice with another person when possible. AI questioning can help organise rehearsal, but it doesn’t reproduce every examiner’s expectations or the experience of responding under formal assessment conditions.
Before your next session, inspect the first prompt you send. If it asks for questions and complete answers together, decide whether that is really what you want. An oral examiner becomes more useful when it gives you room to speak.
