A quiz generator is a genuinely useful AI build — teachers, course creators, and trainers all need to turn material into assessments, and doing it by hand is tedious. Done naively it produces lopsided, ambiguous questions; done well it produces balanced, validated quizzes. This tutorial covers the right approach and includes a builder that generates the exact prompt and JSON schema to use.
Step 1 — Extract concepts before writing questions
The biggest mistake is asking the model “write ten questions about this text” in one shot — it clusters questions around whatever stood out and skips half the material. The fix is two passes. First ask the model to list the testable concepts and facts in the source text. Then ask it to write one question per concept. This makes coverage a deliberate, inspectable step rather than a lucky side effect, and it lets you control quiz length precisely.
Step 2 — Generate structured questions
Every question must come back as structured JSON, not prose. Define a schema —
a question stem, an options array, the index of the correct answer, and an
explanation — and instruct the model to return only conforming JSON. The
explanation field does double duty: it helps learners and it gives you something
to validate the answer key against later.
The hard craft is distractor quality. A good wrong option is plausible to someone who half-knows the material — a common misconception or a true-but- irrelevant fact — and clearly wrong to someone who understands it. Instruct the model explicitly to avoid joke options, “all of the above,” and absurd choices that let test-takers eliminate by guessing.
Step 3 — Validate, then export
Never trust the answer key blindly. Run a validation pass in code: confirm each question has exactly one correct answer, no duplicate options, and a non-empty stem; regenerate any that fail. For higher stakes, add a second, independent model call that re-derives the answer from the stem and options alone — disagreement between the generator and the checker flags a likely error for human review. Once validated, the JSON exports cleanly into any quiz UI.
Use the builder below to assemble the concept-extraction prompt, the question schema, and the validation rules for your own generator, then read how to build a chatbot with the OpenAI API for the structured-output call pattern and how LLMs work to understand why a separate verification pass catches the model’s confident mistakes.