Quiz Generator
Turn your own PDFs and Word documents — scanned pages and photos included, read with OCR — or just a topic into a printable quiz. Choose how many multiple choice, true/false, fill-in-the-blank, short answer, definition and matching questions you want, then download the paper and its answer key as PDF or Word. Every question is built from sentences your material actually contains, and the whole thing runs in your browser — your documents are never uploaded.
How to use Quiz Generator: Open Quiz Generator — no sign-up, and nothing to install. Add your study material: one or more PDF, Word (.docx) or image files — scanned pages and photos are read with OCR. Several files are treated as a single body of material, so a quiz can span a whole set of notes. No document? Choose "From a topic", type the subject and tick the Wikipedia articles to build from. Narrow it down if you want to: give each file a page range (all, 1,5,7,9 or 2-7) — PDF pages are exact, Word pages come from the file where it records them and are estimated where it does not — and click any detected topic to leave it out. The work itself runs inside your browser; the only thing that leaves your device is the small request described below.
Add PDF, Word or image files
PDF, .docx, PNG or JPG. Scanned pages and photos of text are read with OCR. All in your browser — never uploaded.
Turn your own PDFs and Word documents — scanned pages and photos included, read with OCR — or just a topic into a printable quiz. Choose how many multiple choice, true/false, fill-in-the-blank, short answer, definition and matching questions you want, then download the paper and its answer key as PDF or Word. Every question is built from sentences your material actually contains, and the whole thing runs in your browser — your documents are never uploaded. Quiz Generator is a free, browser-based tool in the Filevane pdf tools collection — there is nothing to install, no account to create and no watermark on your results. It works on desktop, tablet and mobile, and because the work happens on your own device it stays fast even with large inputs.
How to use Quiz Generator
- Open Quiz Generator — no sign-up, and nothing to install.
- Add your study material: one or more PDF, Word (.docx) or image files — scanned pages and photos are read with OCR. Several files are treated as a single body of material, so a quiz can span a whole set of notes. No document? Choose "From a topic", type the subject and tick the Wikipedia articles to build from.
- Narrow it down if you want to: give each file a page range (all, 1,5,7,9 or 2-7) — PDF pages are exact, Word pages come from the file where it records them and are estimated where it does not — and click any detected topic to leave it out.
- Set how many questions you want of each type — multiple choice, choose-all-that-apply, true/false, fill in the blank, short answer, definitions and matching each have their own number. Each one shows roughly how many your material can support.
- Pick a difficulty if you want one, and how many options each multiple-choice question should have.
- Click Generate. The questions are built in your browser from sentences your documents actually contain, then shown for you to check.
- Download the paper and its answer key as PDF, Word or plain text. Your documents are never uploaded; a topic search sends only the topic to Wikipedia.
Why use Quiz Generator?
- 100% free with no limits, no sign-up and no watermarks.
- Private by design — your files are processed in your browser and never uploaded to a server.
- Fast and lightweight, with instant results even on a phone or a slow connection.
- Works everywhere — any modern browser on Windows, macOS, Linux, Android or iOS.
- Part of a full pdf tools suite, so related quiz tasks are one click away.
What people use Quiz Generator for
- Setting conceptual questions — how a mechanism works, how two ideas differ — rather than pure recall, from material you already teach from.
- Turning a lecture handout or textbook chapter into a revision quiz before an exam.
- Building a quick comprehension check from the training manual a new starter has just been given.
- Making a printable worksheet with an answer key from your own teaching notes, without retyping any of it.
- Producing several different papers from the same material, so a class does not all sit an identical test.
- Quizzing one chapter of a long PDF at a time by giving the page range, rather than splitting the file first.
- Self-testing on a set of notes: generate questions, answer them, then check against the key.
- Quizzing a scanned worksheet or a photographed textbook page — the text is read with OCR first.
- Making a quick revision quiz on a topic with no notes to hand, built from the Wikipedia article on it.
Quiz Generator: limits and things worth knowing
- No AI model is involved, and nothing is invented. Every question is assembled from sentences your documents already contain — a multiple-choice question's correct answer is the definition your material gives for a term, and the wrong options are the definitions it gives for other terms. Because nothing is written from scratch, no question can state a fact your source does not.
- The material you feed it matters more than any setting. Explanatory prose with clear definitions — "X is a ...", "X refers to ...", "X means ..." — is what most question types are built from. Slide decks of bullet fragments and pages of tables produce far fewer questions.
- Short answer and the Conceptual multiple-choice style ask about understanding rather than recall: how something works, how two things of the same category differ, or how two terms connect. They are built only where the material states the link itself — a "... by ..." clause, two definitions sharing a category, two defined terms in one sentence, or a question your material already answers — so the question probes comprehension without the tool inferring anything your source does not say. Straight "What is X?" recall stays in the Definitions type, so the two never duplicate each other.
- Multiple-choice options come in three shapes. Long statements asks which description fits a term, so the options are full descriptions and there can only be as many questions as your material has definitions. Short terms blanks the key term out of a sentence and offers a list of terms instead — that works on any sentence mentioning a defined term, so it yields considerably more questions from the same documents. The Conceptual style asks how or why instead. A mix rotates between all three.
- Every file has its own page range, so a 300-page book can be quizzed one chapter at a time: type all, 1,5,7,9 or 2-7. PDF page numbers are exact. A .docx stores no pagination of its own, so its pages come either from the boundaries Word left behind when it last laid the file out, or — for a file nothing has ever paginated, such as a Google Docs export — from an estimate based on how much text fits on a page. Estimated pages say so, and a page break you inserted yourself is always obeyed exactly.
- The topics your material covers are listed as soon as it is read — from the document's own headings where it has them, and from the most distinctive words in each passage where it does not. Leave any of them out with a click, and see afterwards how many the paper actually covered.
- Difficulty is what a question makes you do, not how long it is: recall of one stated fact is easy, choosing among plausible alternatives is medium, and reasoning across the material is hard. Because every question is assembled rather than written, the operation it demands is known exactly — so each question carries its true difficulty, and the mix you got is reported rather than promised.
- Two questions on one paper are never allowed to answer each other. A matching exercise pairs five terms with their descriptions, so nothing else on that paper may define those terms; a sentence printed whole is never also printed with a gap in it. This is why asking for several types at once can yield fewer than the per-type figures suggest — those figures are what each type could give you on its own.
- With "Ask the same material more than one way" ticked, a term already used can be asked about again in a different form once every unused one is spent — defined in one question, gapped in another. It never asks the same question twice.
- Each type shows how many questions your material can support before you generate. That figure is produced by actually building them and counting what survives, not by estimating, so the number you see is the number you get.
- If a type produces fewer questions than you asked for, the tool says so and says why. Multiple choice needs several definitions to build plausible wrong answers and matching needs five per question; when there are not enough it builds fewer rather than padding the paper.
- Add as many PDFs and Word files as you like — they are treated as one body of material, and each question records which file it came from.
- Scanned PDFs work. Any page with no text layer is read with OCR in your browser, page by page, so a PDF that mixes typed pages with scanned ones keeps the exact wording of the typed pages. Words inside pictures on typed pages — screenshots, labelled diagrams — are read on request, and only what the text layer lacks is added. Images (PNG, JPG, WebP, BMP) are read the same way. OCR is English-only here, and a faint or skewed scan gives rougher text and fewer questions.
- With no document at all, type a topic: the tool searches Wikipedia or Simple English Wikipedia from your browser, you tick up to five articles, and their text becomes the material. Nothing is written for you — the questions come from the articles' own sentences, each is traced to its article, and the paper and key end with the attribution Wikipedia's CC BY-SA licence asks for. The search sends only your topic and the titles you pick.
- Half the true/false statements are false. Where your material lists misconceptions in a table headed like "What students think | What is actually true", those come first — false because you said so, and labelled "Listed as a misconception in". The rest are altered so the answer is genuinely false, and labelled "Altered from" rather than "Source", because your document does not say them. A misconception is never offered as a true statement or a model answer.
- Word files with structure give far more. A two-column table headed "Term | Meaning" (or Definition, Description) is read as a glossary, so every row becomes a definition for multiple choice, matching, definitions and short answers. Questions you already wrote with their answers — "Q1. … ?" followed by "Answer: …" — become short answers word for word, and their answers are never printed elsewhere on the same paper.
Is it private and secure?
Quiz Generator does its work on your own device. It is one of only five tools on Filevane that contacts the network at all, and it is published as a named exception on our privacy proof page rather than quietly excluded from the claim — the specifics are immediately below.
Network activity: this tool sends one thing. Files you add are never uploaded — they are read, OCR'd where they are scans, and turned into questions inside your browser. Two things do touch the network, and only when you use them. Reading a scanned page or an image downloads the text-recognition engine and its English model from this site itself the first time (data comes in; none goes out, and no third party is involved). And the "From a topic" mode sends the topic you type, then the titles of the articles you pick, to Wikipedia (en.wikipedia.org or simple.wikipedia.org) to fetch their text.
You do not have to take that on trust — our privacy proof page publishes the Content-Security-Policy that enforces it, lists every third party it permits, and shows how to verify the whole claim yourself.
We also measured the competition. Our PDF upload audit recorded how many bytes each of 20 online PDF tools actually transmits: 12 of the 13 measurable services uploaded the whole test file, and Filevane transmitted 0.00 MB.
Learn more
PDF to Word, Excel and PowerPoint: What Actually Converts — What PDF to Word, Excel and PowerPoint conversion keeps and loses, why scanned PDFs need OCR first, and how to convert Office files to PDF — in your browser.
Frequently asked questions
How do I make a quiz from a PDF or Word document?
Add one or more PDF, .docx or image files — scanned pages are read with OCR — or type a topic and pick Wikipedia articles instead. Then choose how many questions you want of each type — multiple choice, true/false, fill in the blank, short answer, definitions or matching — then click Generate. You get a preview you can check, and downloads as PDF, Word or plain text with a separate answer key.
Can I use several documents at once?
Yes. Add as many PDFs and Word files as you like and they are treated as one body of material, so a quiz can span a whole set of lecture notes rather than a single handout. Each question records which file it came from, and you can print that on the paper if you want to.
Can I quiz only part of a document?
Yes. Every file you add has a Pages box: leave it as all, or type the pages you want — 1,5,7,9 for single pages, 2-7 for a run, or a mixture of both. Only those pages are read, so a 300-page textbook can be quizzed one chapter at a time. PDFs have real pages, so those numbers are exact. A Word file stores no pagination of its own — where its pages fall depends on the fonts and paper size of whatever opens it — so page numbers there come from one of two places: the boundaries Word left behind the last time it laid the file out, which are exact, or an estimate worked out from the text when the file has never been laid out at all. Estimated pages are labelled as such, any page break you inserted yourself is obeyed exactly, and the rest may sit a page either side of what Word shows you.
Can I pick which topics the quiz covers?
Yes. The topics your material covers are listed as soon as the files are read — taken from the document's own headings where it has them, and from the most distinctive words in each passage where it does not. Click a topic to leave it out, and no question will be built from it. After generating, the tool reports how many of the topics in play the paper actually covered.
Can I set the difficulty?
Yes, and difficulty here means what a question makes you do rather than how long it is. Easy is recall of a single stated fact; medium is choosing the right answer among plausible alternatives from the same material; hard is reasoning across it — mechanisms, contrasts and links between ideas. Choosing a difficulty makes the generator prefer questions of that kind where the material supports them. Some types can only ask one way — a definition is always recall — so the mix you actually got is shown after generating rather than promised in advance, and every question carries its own label.
Can questions have more than one correct answer?
Yes — the "Choose all that apply" type builds questions where several options are right. The correct options are statements your material makes about the term, and the wrong ones are real sentences about a different term, so nothing is altered or invented. You can also set 3, 4 or 5 options per question for both this and ordinary multiple choice.
Can it ask about the same material in more than one way?
Yes, and this is how a short document can still fill a long paper. With "Ask the same material more than one way" ticked, a term already used once can be asked about again in a different form — defined in one question, used in a gapped sentence in another — but only once every unused one has been spent. It never asks the same question twice, and it never puts two questions on one paper that answer each other: a matching exercise that pairs a term with its description means no other question may define that term, and a sentence printed whole is never also printed with a gap in it. That is why asking for lots of types at once can produce fewer questions than the per-type figures suggest.
Can I choose how many questions of each type?
Yes — every type has its own number box, so you can ask for ten multiple choice, five true/false and three matching in one paper. Each type shows roughly how many your material can support before you generate, so you are not guessing.
Does this use AI, and can it invent things?
It does not use a language model, and it cannot invent anything. Every question is assembled from sentences your documents — or the Wikipedia articles you picked — already contain — the correct answer to a multiple-choice question is the definition your material gives, and the wrong options are definitions it gives for other terms. Because nothing is written from scratch, no question can state a fact your source does not.
Can I choose long or short multiple-choice options?
Yes, and there is a third option. Long statements asks which full description fits a term, so the options are complete sentences. Short terms turns it round: a sentence has its key term blanked out and you pick the term from a short list — that also gives you far more questions, because it works on any sentence rather than only the definitions. Conceptual asks how something works or how two ideas connect. Or choose a mix of all three.
Can it ask conceptual questions, not just definitions?
Yes. Short answer questions are conceptual by design — they ask how something works ("How does regularisation reduce overfitting?"), how two things of the same kind differ ("What is the difference between classification and regression?"), or how two ideas connect ("How are backpropagation and gradient descent related?"). Multiple choice has a Conceptual style that does the same. Straight "What is X?" recall lives in the separate Definitions type, so the two never overlap. Conceptual questions are only built where your material states the link itself — a causal clause, two terms of the same category, or two terms in one sentence — so nothing is inferred that your source does not say.
Why did it make fewer questions than I asked for?
One of two reasons, and the notice tells you which. Either your material ran out of sentences that can support that kind of question fairly — multiple choice needs several definitions to build plausible wrong answers, and matching needs five — or the sentences it would have used are already asked about elsewhere on the paper, and asking twice would give the answer away. The per-type figures shown before you generate are what each type could give you on its own, so they do not add up when you ask for several at once.
What kind of document works best?
Explanatory prose with clear definitions — lecture notes, textbook chapters, revision summaries, training manuals. Sentences of the form "X is a …", "X refers to …" or "X means …" are what most question types are built from. Slide decks of bullet fragments and tables of figures produce far fewer questions.
Does it work with scanned PDFs?
Yes. Any page that has no text layer — a scan, a photographed handout, slides exported as pictures — is read with OCR (optical character recognition) automatically, right in your browser, and the file shows how many of its pages were read that way. A PDF that mixes real text with scanned pages is handled page by page, so the text pages keep their exact wording. If a page has real text AND pictures with words in them, such as a screenshot or a labelled diagram, click "Also read text inside pictures" under the file: those pages are OCR'd too, and only the words the text layer did not already contain are added. OCR reads English, and clean scans at around 300 dpi work best; a faint or skewed scan gives fewer and rougher questions.
Can I make a quiz from a photo or screenshot?
Yes. Add PNG, JPG, WebP or BMP images alongside your PDFs and Word files and the text in them is read with OCR, then treated like any other document. A photo of a textbook page taken square-on in good light works well.
Can I make a quiz from just a topic, without a document?
Yes. Choose "From a topic", type what the quiz should be about, and the tool searches Wikipedia — or Simple English Wikipedia, whose shorter sentences suit younger learners — and lists the matching articles. Tick up to five, and their text becomes the material, exactly as if you had uploaded it: you still choose the question types, how many of each, and the difficulty. Every question is still built from sentences the articles actually contain, each one records which article it came from, and the downloaded paper credits the articles, as Wikipedia's licence asks.
Are my documents uploaded anywhere?
No. The files are read, scanned pages OCR'd, the questions built and the PDF and Word files written entirely inside your browser. Nothing you add is sent to a server. Two features do use the network, and only when you use them: the first scanned page downloads the English text-recognition model (a download to you — your file is not sent), and "From a topic" sends the topic you type and the articles you pick to Wikipedia to fetch their text.
Related tools
See all free pdf tools or browse the full tool catalogue.