Re-reading feels productive and isn’t. The part that actually builds memory is active recall: being forced to retrieve the answer before you see it. In the classic Roediger and Karpicke (2006) study, students who self-tested retained about 61% of a prose passage a week later, versus roughly 40% for the group that only re-read — testing produced substantially better long-term retention than restudying. The trouble is most AI-generated quizzes default to easy fact-recall, the questions you’d already pass. These 12 prompts push the model to target your real weak spots: misconceptions, application transfer, and the gap your last quiz exposed. See also flashcard prompts for spaced-repetition card decks.
TL;DR
- Paste your notes, transcript, or chapter and ask for mixed question types — never let the model give you all fact-recall.
- Always demand an answer key with a one-sentence explanation per question, or you learn nothing from a miss.
- For exam prep, the highest-leverage prompts here are #3 (misconception distractors), #6 (application transfer), and #8 (mistake-debrief).
- Best free option as of June 2026: upload your PDF to Gemini 3.1 Pro (10 files, up to 100MB each in one chat) or Claude (Sonnet 4.6, 1M-token context). Export the questions into Anki for spaced repetition.
Which AI to use for quizzes (June 2026)
All three major chat assistants accept document uploads and write solid draft questions for free. None gives you a shareable, graded quiz on its own — treat the output as a draft you review, then import the cards into Anki or a quiz tool.
| Tool | Free tier | Context for one long upload | Best for |
|---|---|---|---|
| Gemini 3.1 Pro | Yes | 1M tokens; 10 files, up to 100MB each per chat | Big PDFs, whole textbook chapters, slide decks |
| Claude (Sonnet 4.6) | Yes (limited) | 1M tokens standard | Long transcripts, careful rubrics, nuanced distractors |
| ChatGPT (GPT-5.5) | Yes (ads on US Free since Feb 2026) | ~320 pages in-app on Plus ($20); full 1M only on $200 Pro | Quick quizzes from short notes |
For repeated study, pair any of them with Anki, which is free and open-source and has the FSRS (Free Spaced Repetition Scheduler) algorithm built in since version 23.10 (no add-on required) — use prompt #4 below to generate import-ready cards.
Best for
- Exam prep (USMLE, bar, certifications)
- Reading retention
- Team training and onboarding
- Tutoring kids or younger students
- Self-quizzing on a podcast or lecture
1. Quiz from notes
Below are my notes on [topic]. Generate a 10-question quiz mixing question types:
- 4 fact-recall
- 4 application (apply concept to a NEW scenario, not one in the notes)
- 2 explain-in-your-own-words
Provide an answer key with a 1-sentence explanation per question. Mark which note passage each question comes from.
Notes: [paste]2. Quiz at 3 difficulty levels
From the notes below, generate 3 quiz tiers (5 questions each):
- Basic recall (definitions, named entities)
- Application (use the concept on a new example)
- Edge case / synthesis (combine two concepts, or find the boundary condition)
Label which tier matches which exam type (e.g., basic = passing the course, edge = research seminar).
Notes: [paste]3. Misconception-targeting quiz
Topic: [topic]. Common student misconceptions: [list 3-5].
Build 6 multiple-choice questions where each distractor mirrors a real misconception (not a random wrong answer).
For each question give: question stem, 4 options labeled A-D, correct answer letter, and a 1-sentence explanation of which misconception each wrong option represents.4. Spaced-repetition card set
From the notes below, generate 20 cards in Q-A format suitable for Anki or RemNote.
For each card: front (15 words max question), back (30 words max answer), difficulty tag (easy / medium / hard), and a suggested first-review interval (1d / 3d / 7d).
Order the deck by suggested first pass — easiest first to build momentum.
Output as a tab-separated table so I can paste it straight into Anki's import.
Notes: [paste]5. Visual / diagram quiz
Topic: [topic]. Generate 8 questions that require interpreting a visual element (diagram, chart, formula, anatomical image, code snippet).
If you can't render the image, give a clear alt-text description and ask the question against that.
For each: question | the visual element described | correct answer | explanation.6. Application quiz from a textbook chapter
Chapter content below. Generate 6 application questions where the student must transfer a concept from the chapter to a NEW scenario the chapter doesn't mention.
For each: scenario | question | model answer | rubric for partial credit (0 / 1 / 2 / 3 points with criteria).
Chapter: [paste]7. Self-explain quiz
Generate 5 prompts where the student must explain a concept in their own words (no multiple choice).
After each main prompt, add 1 follow-up probe that asks "but why?" or "what would change if [condition]?".
At the end, give a 4-criterion grading rubric so the student can self-assess (clarity, accuracy, depth, transfer).
Topic: [topic]8. Mistake-debrief quiz
I got these questions wrong on my last practice set: [paste list with my wrong answers].
Diagnose the underlying skill gap behind each mistake (not just "you got it wrong").
Then generate 6 new questions targeting the same underlying gap, at increasing difficulty.
Output: diagnosis paragraph | 6 new questions | answer key.9. Quiz from a video transcript
Transcript of a lecture / podcast / YouTube video below. Generate an 8-question quiz that covers the actual content density:
- Skip filler, intros, and tangents
- Weight questions by minutes spent on each subtopic
- Mix fact-recall (3) and application (5)
Include timestamps from the transcript so the student can re-watch the relevant section if they miss a question.
Transcript: [paste]10. Cumulative end-of-week quiz
Below are notes from 5 separate study sessions this week. Generate a 15-question cumulative quiz that:
- Forces interleaving (questions from session 1 mixed with session 4)
- Includes 3 "synthesis" questions that require combining two sessions' concepts
- Flags which session each question draws from in the answer key, not the question itself
Notes from all 5 sessions: [paste]11. Quiz with confidence calibration
Generate 10 questions on [topic]. For each, after the student answers, they must also rate their confidence 1-5.
Then in the answer key, give a calibration debrief: "You got 8/10 right, but rated 4-5 confidence on 9 questions — overconfidence on the 2 you missed."
This trains metacognition, not just content.12. Quiz audit and difficulty rebalance
Below is a quiz I already wrote. Audit it for:
- Difficulty distribution (too easy? too hard? all the same level?)
- Question-type balance (all fact-recall? no application?)
- Misconception coverage (are common wrong intuitions tested?)
- Clarity (any ambiguous stems or implausible distractors?)
Then rewrite the 3 weakest questions and explain what you changed.
Quiz: [paste]How to use these well
- Replace the
[topic]and[paste]placeholders with your own material before sending. Pasting the source text (not just a topic name) is what stops the model from inventing facts. - Answer the quiz before you scroll to the key. The retrieval attempt, not the reading, is what builds memory.
- Feed your wrong answers back in with prompt #8 the next day. Repeated targeting of the same gap is the whole game.
- Pour the keepers into Anki via prompt #4 so the FSRS scheduler resurfaces them right before you’d forget.
Common mistakes
- Only fact-recall questions — application is where real understanding lives.
- No answer key with explanation, so the student can’t learn from the miss.
- One difficulty for everyone, so strong students coast and weak ones drown.
- Distractors that are obviously wrong, turning a 4-choice MCQ into a 2-choice one.
- No interleaving across topics, so the brain learns to retrieve only when primed.
FAQ
Will the AI make up facts in the quiz? It can, especially on niche topics. Always paste the source notes or chapter into the prompt and ask it to mark which passage each question comes from (prompt #1). Then spot-check the answer key against your source before you trust it.
Which model is best for quizzes as of June 2026? For large documents, Gemini 3.1 Pro handles the biggest uploads (10 files, up to 100MB each, 1M-token context). Claude with Sonnet 4.6 writes the most careful rubrics and misconception distractors. For short notes, free ChatGPT (GPT-5.5) is fastest. All three are free for personal study.
How do I turn quiz output into Anki cards? Use prompt #4 and ask for a tab-separated table. In Anki, choose File then Import, set the field separator to Tab, and map column 1 to Front and column 2 to Back. Anki’s built-in FSRS scheduler then handles the review timing.
How many questions should one session have? Short and frequent beats long and rare. Ten well-targeted questions you actually retrieve from memory outperform a 50-question dump you skim. Run a small quiz daily and a cumulative one (prompt #10) once a week.
Multiple choice or open-ended? Open-ended (free recall) produces stronger memory because it forces full retrieval, but it is slower to self-grade. Use multiple choice (prompts #3, #5) for speed and broad coverage, and self-explain prompts (#7) for the concepts you most need to own.
Related
- Flashcard prompts
- Study plan prompts
- AI Quiz Generator: Turn Notes Into Self-Tests That Actually Build Mastery
Tags: #Prompt #Study #Quiz generation