Quick answer: to create an exam from a class with AI, first get its transcription -by recording the session and transcribing it with a tool like VOCAP- and then paste that text into an AI model (Claude or ChatGPT), asking it to generate questions of the type you need (multiple choice, true/false, short answer or essay) with their answers and a brief justification. Tell it to stick strictly to the class so the questions cover only what was explained, and ask for a mix of types and difficulties to build a realistic mock exam.
Rereading your notes gives you a false sense that you know the material. Then the exam arrives, the first question appears, and you realize that recognizing a concept isn't the same as being able to explain it. The most effective way to study isn't rereading: it's testing yourself with questions. The problem has always been how long it takes to make those questions. In 2026, AI solves that in seconds from the class itself.
In this guide you'll see how to turn a recorded lecture into a review quiz: the question types you can generate, copy-paste prompts for multiple choice, flashcards and mock exams, and how to make the questions genuinely resemble the ones that come up.
Why study with questions, not by rereading
Decades of research on learning point to the same thing: retrieving information from memory cements concepts far better than exposing yourself to them again. It's the so-called testing effect. Generating quizzes from your classes with AI leverages exactly that:
- You spot what you don't know. A question you can't answer pinpoints exactly the gap you need to review.
- You study actively. The effort of recalling, not passive rereading, is what consolidates memory.
- You review in a spaced way. With dozens of questions ready, you can self-assess several times over the days.
- You simulate the exam. Answering against the clock and without notes gets you used to the real conditions.
- You cover the whole syllabus. AI generates questions from every part of the class, not just from what you remember.
Question types AI can generate
Don't limit yourself to multiple choice. From the same class you can request very different formats depending on what you want to train:
| Question type | What it trains | When to use it |
|---|---|---|
| Multiple choice | Recognition and discrimination between similar concepts | Quick review and multiple-choice exams |
| True / false | Precision on definitions and nuances | Spotting concepts you think you know but confuse |
| Short answer | Active recall without cues | Definitions, dates, formulas, key terms |
| Essay | Explaining, connecting and arguing | Theory exams and open-ended questions |
| Flashcards | Spaced memorization of concept-definition pairs | Vocabulary, formulas and facts you need to retain |
Key point: the quality of the questions depends, above all, on the quality of the source transcription. If the class contains formulas, proper names or technical vocabulary and the transcription captures them poorly, the questions will inherit those errors. An accurate transcription is the foundation of a good quiz.
Step by step: from class to quiz
This workflow works with any class, whether online or in person.
Step 1 — Get the transcription
The raw material is the text of the class. If it was recorded on a platform with quality captions, download them. If it was in person, in another language, or you want real accuracy, transcribe the recording: upload the audio or video to an engine like Whisper. Tools like VOCAP do it in one step and return clean, punctuated text. For the previous step, check out our guides on how to transcribe university lectures with AI and on transcribing online classes.
Step 2 — Review key terms and concepts
Before generating questions, check that formulas, proper names and technical terms are spelled correctly in the transcription. An error in one concept spreads to every question that depends on it. With a good transcription engine, this is a matter of seconds.
Step 3 — Choose the question type
Decide what you want to train: multiple choice for a quick review, short answer for active recall, essay for theory, or flashcards to memorize. You can combine them. In the next section you'll find ready-made prompts for each case.
Step 4 — Generate the quiz
Paste the transcription, add the prompt above it and run. Always ask for the answers separately (or at the end) so you can try to answer first. Check that the questions are correct and based on the class; if any is ambiguous, ask the AI to rephrase it.
Step 5 — Self-assess and review
Answer without looking at the solutions, grade yourself and note the topics you get wrong so you can review them. Repeat the quiz a few days later: spaced repetition is what cements the concepts. Combine the questions with some structured notes made with AI to review the theory on the points you miss.
Doesn't the class have decent captions?
Transcribe the recording accurately and generate your quizzes from a reliable text. Try VOCAP free: 30 minutes, no card required.
Try VOCAP FreeCopy-paste prompts
Paste the class transcription and add one of these prompts above it.
Multiple choice
From this class transcription, create 10 multiple-choice questions with
4 options each. Stick to what appears in the class, don't add outside
topics. Give the correct answer and a brief justification at the end,
in a separate section so I can answer first myself. Transcription:
[paste the transcription here]
Short answer and essay
Generate 6 short-answer questions and 3 essay questions about this class,
ordered from easiest to hardest. For each one, add a model answer
afterward based only on the transcription and point to the topic or
section it comes from. Transcription:
[paste the transcription here]
Flashcards to memorize
Turn the concepts, definitions and formulas from this class into 15
flashcards in "question | answer" format, one per line, so I can
import them into a spaced-repetition app. Be concise and faithful to
the class. Transcription:
[paste the transcription here]
Building a full mock exam
To really review, isolated questions aren't enough: you should reproduce exam conditions. With the transcription of all the classes in the syllabus you can ask for a whole exam:
- Gather the syllabus. Transcribe all the relevant classes and combine them into a single text, or summarize them first in blocks if they're very long.
- Define the structure. State how many questions you want per topic, the mix of types, the estimated time and a point scale.
- Ask for the answers separately. That way you can take the exam "cold" and grade afterward, just like in real assessment.
If the syllabus spans several long sessions, summarize them first: our guide on summarizing classes and lessons with AI explains the block-by-block technique so you don't lose detail. And if you're working with thesis or research audio, check the one on transcription for theses and dissertations.
Turn any class into review questions
Accurate transcription with Whisper (OpenAI) + automatic summaries with Claude (Anthropic). Upload the audio and get the transcription ready to generate your quizzes. From €1/hour.
Start Free with VOCAPHow to make the questions resemble the real exam
A generic quiz helps, but one that imitates the professor's style helps much more. Some tweaks that make the difference:
- Keep the AI tied to the class. Explicitly tell it not to introduce topics that weren't mentioned; that way you avoid studying what isn't on the test.
- Copy the exam format. If you know the exam is multiple choice, or two essay questions, ask for exactly that format.
- Adjust the difficulty. Ask for "application" or "comparison" questions instead of just definitions if the exam requires reasoning.
- Mark the source. Ask each question to point to the topic or minute it comes from, so you can review the relevant part of the class when you get it wrong.
Common mistakes that ruin a quiz
- Starting from a transcription with errors. The questions inherit and amplify the mistakes, especially in formulas and technical terms.
- Letting the AI make up content. If you don't tie it to the class, it will add topics from its general knowledge that may not be on the test.
- Looking at the answer before trying. The value is in the effort of recalling; always ask for the solutions separately.
- Generating the questions only once. Spaced review -repeating days later- is what cements the concepts.
- Studying only with multiple choice. Combine formats: multiple choice trains recognition, short answer trains recalling without cues.
Frequently asked questions
How do you create an exam from a class with AI?
In three steps: get the class transcription (by recording it and transcribing it with a tool like VOCAP), paste that text into an AI model like Claude or ChatGPT, and ask it to generate questions of the type you need with their answers and a brief justification. For a full exam, ask it to mix types and difficulties and to rely only on the transcription.
What types of questions can AI generate from a class?
Almost every format: multiple choice, true/false, short answer, essay questions, fill-in-the-blank, matching concepts and flashcards. The key is to request the format explicitly and state how many questions you want of each type. You can also ask each question to point to the topic or minute it comes from.
Do the questions AI generates match what's on the exam?
They match if the AI works from the real transcription of the class and you tell it to stick strictly to it, without adding outside topics. If you know your professor's criteria -how much weight theory carries versus problems- state that so the questions resemble the real ones.
How do I build a full mock exam from several classes?
Transcribe all the classes in the syllabus and combine them into one text, or summarize them first in blocks. Then ask for an exam with a specific structure: number of questions per topic, a mix of types, an estimated time and a point scale, with the answers separate so you can self-assess first and grade afterward.
Are AI-generated quizzes actually useful for studying?
Yes, and they're among the most effective ways to study. Answering questions triggers memory retrieval (the testing effect), which cements concepts far better than rereading. AI makes generating dozens of questions take seconds, so you can self-assess repeatedly and in a spaced way. The important thing is to try to answer before looking at the solution.
Do I need the transcription or can I give the audio directly to the AI?
You need the transcription as an intermediate step: text models work with text. First you transcribe the audio and then generate the questions from that text. VOCAP does the transcription with Whisper and delivers clean, punctuated text in a single step, ready to paste into the AI.