Which tool to choose
- You want to answer out loud, like in a real oral: use an assistant with voice conversation mode (the ChatGPT or Gemini app offers it for free). You speak, the AI listens and responds. It's the only way to also train the voice that trembles.
- You want an examiner built on your material: NotebookLM generates questions from the documents you upload and points you back to the source for each correct answer. Great when the professor quizzes "on the book."
- You only want to train the content in writing: any assistant in the text box will do.
How to do it
- Upload or paste the syllabus: the book's table of contents, handouts, the course topics. The more the AI knows the perimeter, the more the questions resemble the real ones.
- Assign the role before starting. The operational syntax:
Act as the professor examining me on this syllabus.
Quiz me one question at a time, from the simplest to the hardest.
After each of my answers: tell me what I got, what I got wrong or forgot, and pose a follow-up question as a demanding examiner would. Don't give me the right answer until I ask for it.
- Answer as if you were in front of the examining board: without rereading the book, from memory. If you use voice, turn on conversation mode and speak.
- When you get it wrong, don't move on. Ask: "explain where I lost the thread and ask me a similar question again." Repetition on the weak point is what fixes the concept.
- At the end, ask for the assessment. The operational syntax:
Let's stop. List the three topics I was weakest on and, for each, the exact question you'll ask me again tomorrow to check whether I've recovered.
A concrete example
Luca has a contemporary history oral. He loads the textbook's table of contents into NotebookLM and activates the examiner role. First, an easy question: the causes of the First World War. He answers well. The AI presses: "was the system of alliances a cause or an accelerator?" Luca hesitates. The AI points out that he's confused cause and trigger, asks him the question again on a similar case (the Sarajevo crisis), and this time Luca makes the distinction. At the real oral, the professor asks almost the same follow-up question. Luca had already seen it.
When it does NOT work (and how to fix it)
If the AI asks you questions that are too easy or generic
It means it doesn't have enough material to calibrate the difficulty. Remedy: load the detailed syllabus and add "raise the level: top-marks questions, with the connections between topics that a demanding professor expects."
If the AI accepts a wrong answer as good
Assistants tend to be accommodating. Remedy: in the role write "be strict, don't praise me out of courtesy: if the answer is incomplete tell me clearly." And verify for yourself the points you have doubts about, by opening the sources.
If you can't find the voice mode
The buttons move around. Remedy: look for an icon shaped like a headset, a microphone or a sound wave in the mobile app. If there isn't one, you simulate the oral in writing: write your answers off the cuff, without deleting, to mimic the fact that out loud you can't go back.
A tip from someone who really uses it
Record yourself while answering out loud and listen back. The AI evaluates the content, but in an oral exam how you speak also counts: the pauses, the "ums," the confidence. The content is corrected by the AI, the performance is corrected by your own ear.
Frequently asked questions
Do the AI's questions really resemble the professor's?
They resemble them if you give it the right syllabus and ask for connecting questions, not just definitions. No simulation guesses the exact question, but training yourself to reason under pressure is what saves you when the question you hadn't anticipated arrives.
Can I use it for foreign-language orals too?
Yes, and it's one of the best uses: the AI quizzes you and replies in the language, correcting pronunciation (in voice mode) and grammar. Set the role directly in that language.
Doesn't having AI quiz you create a false sense of security?
Only if the AI flatters you. An AI examiner tuned to be strict has the opposite effect: it shows you the cracks before the board finds them. False security comes from reviewing by reading; tackling the questions cold dismantles it.