We use GPT 3.5 and 4 for creating live quizzes at Faabul Live Quizzes. You really need to use a good prompt, ideally provide example(s) how the quiz should look like. Include the number of questions that you want, what type of quiz you need, how many answers it should offer.
With GPT 3.5, the results can be mixed, even though the prompt is pretty good. With GPT 4, the results are awesome, more predictable and you will get what you ask for in the prompt.