Using AI to Practise Your Viva: What Actually Helps
Candidates ask some version of this every week: can I give my thesis to an AI and have it examine me? The answer is yes, and the more useful question is which tools actually examine you rather than simply talk to you. We build one of these, so read what follows with that in mind. The four tests below are the ones we would want you to apply to us.
Practising With Software Is Not Cheating
This needs saying first, because candidates who ask the question in public are sometimes told it is a form of dishonesty. It is not. Handing your thesis to a friend and asking them to question you is ordinary preparation that nobody objects to. Software doing the asking changes who generates the question, not who has to answer it. You still have to know your own work, out loud, under challenge.
What would be dishonest is having a tool write the thesis you are about to defend. Rehearsing answers about research you did yourself is the opposite of that. It is the part of preparation most candidates skip, usually because they have nobody available to do it with.
Test One: Does It Disagree With You?
Most of what you will find is a general assistant with a voice layer added. It asks reasonable questions. Then, whatever you answer, it accepts it and moves on, often with encouragement. That is what general assistants are built to do, and it is the wrong instinct entirely for this.
An examiner does not behave that way. When your answer does not match what is in your data, an examiner says so. The sentence you are actually afraid of is some version of “that is not what your results show”, and a tool that never produces it is testing your recall rather than your defense.
Test Two: Does It Stay On The Question?
Real examinations are built from lines of questioning, not from lists of questions. An examiner asks something, listens, and asks again in the same direction, narrowing until the point is resolved. The pressure candidates describe afterwards does not come from the first question. It comes from the third follow-up, once the easy answer has already been spent.
A tool that moves to a new topic after every answer is running a quiz. That has some value for checking coverage, but it cannot reproduce what makes a viva difficult, because it releases you at precisely the moment a real examiner would push.
Test Three: One Voice, Or A Panel?
A conversation with a single examiner is an interview. An examination is usually several people who have read your thesis differently and who pick up each other’s threads. Your methodologist and your subject specialist are not interested in the same weaknesses, and being asked about one and then immediately the other is part of what you are preparing for.
A panel also has somebody running it, deciding when a line of questioning is finished and when you have said enough. Being moved on before you had finished your point is something candidates remember for years. It is worth meeting before the day rather than on it.
Test Four: Does It Make You Speak?
Typing lets you edit. You draft, reconsider, and submit something better than your first attempt. Your viva allows none of that. You answer in one pass, committed to the sentence you happened to begin.
Research on performance under pressure is consistent on this point: silent rehearsal is much weaker preparation than rehearsing aloud. If getting flustered when speaking is the thing worrying you, written practice will not address it, because it never places you in the situation you are worried about.
What This Kind Of Practice Cannot Do
It cannot tell you how your particular institution runs its examinations, and conventions differ more than candidates expect. It cannot replace a supervisor who knows your field and, often, your examiners. It will not know the one obscure paper your external has spent a career defending.
What it can do is the part that is otherwise hardest to arrange, which is repetition. Most candidates get one mock viva, and many get none. The thing that reliably reduces being flustered is having already answered the hard questions several times, and that depends on how often you can practise rather than on how good any single session was.
Testing Any Tool In Ten Minutes
- Give a weak answer on purpose, and see whether it is challenged or accepted.
- Answer vaguely, and see whether the next question narrows or changes subject.
- Count the voices. Does anyone ever follow up on somebody else’s point?
- Check whether you speak your answers, and whether it is possible to be interrupted.
- Ask about your methodology chapter, and see whether the question shows it read your methodology chapter.
Anything worth your time survives all five. A great deal of what you will find does not survive the first.