PDF to Quiz AI: Turn Documents into Tests

The short version

Upload the PDF to an AI quiz generator, choose the question types and how many you want, then read through the output and delete the weak questions before you study. The generation takes a couple of minutes. The review pass takes five, and skipping it is the main reason people conclude AI quizzes are useless.

Scholarly's quiz generator does this from a PDF, and the practice test generator builds a longer, exam-shaped version from the same source. Both keep the questions editable, which matters more than it sounds.

Why quiz yourself instead of rereading

This is the part of the pitch that does not depend on any product.

Roediger and Karpicke's retrieval-practice experiments compared students who reread material against students who tested themselves on it. The rereading group felt more prepared. The testing group retained more when tested later. The gap between how prepared you feel and how prepared you are is exactly what rereading widens and what quizzing closes.

The 2013 review by Dunlosky, Rawson, Marsh, Nathan and Willingham in Psychological Science in the Public Interest assessed ten study techniques and rated practice testing as one of only two with high utility. Rereading and highlighting, the two things students actually do most, were rated low.

You should be sceptical of articles that put a precise number on this. Retention gains vary enormously with material, delay, and question quality, and the specific percentages that circulate in marketing copy are rarely traceable to a study. The direction of the finding is solid. The exact size is not a fixed constant.

Turning a PDF into a quiz, step by step

1. Pick the right slice of the document. A 200-page textbook generates shallow questions because the model has to spread its attention. One chapter or one lecture produces sharper ones. If the tool has a page-range option, use it.

2. Upload the PDF. Text-based PDFs work best. A scanned document is an image until something runs OCR over it, so if your file is photographs of pages, expect weaker results.

3. Choose question types deliberately. Multiple choice is fast to answer and auto-gradable but easy to answer by elimination. Short answer forces genuine retrieval, which is the point. A mix, weighted toward short answer for material you need to understand rather than recognise, is usually right.

4. Generate, then review. This is the step that decides quality. Delete questions that quiz trivia (a figure number, a date in a citation), fix any that give the answer away in the phrasing, and rewrite anything ambiguous. The review pass is itself a first study session, which is why it costs less than it looks.

5. Take it properly. Closed book, no scrolling back to the source. Then check. The questions you get wrong are the entire output of the exercise; the ones you get right mostly confirm what you already knew.

6. Come back to the failures. Retest the missed questions a few days later rather than immediately. Spacing the retest is what converts a good session into retention.

What AI quiz generators are good and bad at

They are good at: definitions, terminology, factual recall, comparisons that the source states explicitly, and generating enough volume that you actually have something to practise on.

They are unreliable at: anything requiring multi-step reasoning across distant parts of a document, questions about material the PDF only implies, plausible wrong answers for multiple choice (distractors are the hardest part to generate well), and anything involving diagrams or notation that did not survive text extraction.

They will occasionally be wrong. A generated question can contain a subtly incorrect premise, and studying a wrong question teaches you a wrong thing. This is the single strongest argument for reading the output before you use it, and for preferring tools that show which part of the source a question came from.

Tools worth knowing

Pricing for third-party tools changes often, so check each product's own pricing page rather than trusting a figure in a comparison article, including this one.

Scholarly generates quizzes and practice tests from PDFs, slides, recordings, videos, and notes, keeps every question editable, and links them to flashcards and notes built from the same source. Free to start with no credit card. The free plan is capped at 1 lifetime AI generation and 1 lifetime file upload, with an 8 MB file limit and the first 32 pages of a PDF read by the agent. Plus is $8 per month or $76 per year and Premium is $30 per month or $144 per year, as of August 2026. See pricing for the current figures.

Quizizz is built for live, gamified classroom quizzing with a large shared question library. Strong for engagement in a classroom, weaker as a way to generate questions from your own document.

Kahoot is the classroom competition format. Excellent for group review, but it assumes live participants and is not a solo revision tool.

Google Forms is free with any Google account and fine for a hand-written quiz you want to distribute and auto-grade. There is no document-to-quiz generation, so you are writing every question yourself.

Formative is aimed at teachers who want real-time visibility into student responses and standards alignment, rather than at students revising alone.

The honest split: if you want to deliver a quiz to a group, the classroom tools win. If you want to generate one from a document you already have, that is a different job.

Writing better questions than the default output

Even a good generator produces generic questions unless you push it. Two things help most.

Aim higher up Bloom's taxonomy. Most generated questions sit at the Remember level: define this, name that. Ask explicitly for questions at the Apply and Analyse levels, meaning questions that require using the concept in a new situation or explaining why something follows. Those are harder to answer and much closer to what an exam asks.

Match the format to the assessment. If your exam is essay-based, a deck of multiple choice questions is practising the wrong skill. Generate short-answer and explain-why questions instead, answer them in writing, and compare against the source.

Question type	Best for	Auto-gradable
Multiple choice	Broad coverage, fast checking	Yes
True/false	Quick misconception checks	Yes
Fill in the blank	Terminology and formulas	Yes
Short answer	Genuine retrieval, explanations	Partially
Essay	Argument and synthesis	No
Matching	Relationships between terms	Yes

For teachers

The time saving is real but it lands in a different place than people expect. Generation is fast; verification is not. A generated 20-question quiz still needs a read-through before it goes in front of a class, because a question with a wrong premise is worse than no quiz.

Where it genuinely helps: producing a first draft from your own course materials rather than from a generic bank, building low-stakes retrieval checks often enough that students actually get spaced practice, and generating several variants of the same assessment.

Where it does not: anything you would not personally sign off on. Treat generated questions the way you would treat a first draft from a teaching assistant.

Common questions

Can AI make a quiz from any PDF? From most text-based PDFs, yes. Scanned documents need OCR first, and heavily visual material such as diagram-driven slides produces weak questions because the meaning is in the images rather than the text.

Are AI-generated quiz questions accurate? Usually, but not always, and the errors are the subtle kind. Read the questions before you study them. Tools that ground each question in a specific passage of your source make that check much faster.

Is there a free way to do this? Yes. Scholarly's free plan includes an AI generation you can spend on a quiz, and there are other free tiers around. Free tiers across the category are generally capped in some way, so expect limits rather than unlimited use.

How many questions should a practice quiz have? Ten to twenty for a single lecture or chapter. Enough to cover the material, short enough that you will actually sit down and do it. Frequency beats length here.

Should I use multiple choice or short answer? Short answer, when you can be bothered to mark yourself honestly. It forces genuine retrieval rather than recognition. Multiple choice is a reasonable compromise when you need volume or speed.

Getting started

Take the most recent lecture PDF you have not revised, generate ten questions from it, delete the three weakest, and answer the rest closed book. That whole loop takes about fifteen minutes and will tell you more about where you stand than an hour of rereading.

Create a free Scholarly account if you want to try it on your own material. No credit card required.