AI quiz generator

Quizzes that are actually marked.

Nine question types, from multiple choice to essays, code and numeric answers. Six of them are checked by code and cannot be marked wrongly; the other three are graded against a rubric written before you saw the question.

Generate a quiz Try the code runner

Checked, not just shown

Practice that checks your work.

A chat tells you “great answer!” every time. Here a written answer is marked against a rubric written first, code runs against real tests, and a speaking drill hears what actually came out of your mouth.

9question types, all graded
4live code runtimes

01You write · rubric grading

A rubric marks it, line by line.

7/8Check yourselfMark complete

Check yourself

2Short answerMarked by AI1 attempt

Your shot tastes sharp and sour. What went wrong, and what would you change?

Sour means the water rushed through before the sweetness dissolved. More contact fixes it: grind finer. Bitter is the opposite — go coarser.

24 words

Submit

Passed

80 / 100 · pass mark 70

You tied the sourness to the water leaving too soon and named the right lever. You didn’t say why a finer grind slows it down.

Rubric

Explains the cause· weight 0.590%

Proposes a specific fix· weight 0.570%

Try a different answer

02You code · it runs it

It runs your code. The tests decide.

5/8ExerciseMark complete

Exercise3 checksPYTHON

total() must refuse a negative price. Make all three cases pass.

Write the functionpython · sandbox · 5sready

1def total(prices):2    """Sum a cart."""3    if any(p < 0 for p in prices):4        raise ValueError("negative")5    return sum(prices)

returns the right totalinput [3, 1, 5] → 9
handles an empty cartinput [] → 0
rejects a negative priceinput [3, -1, 5] → ValueError

Run again3/3 passing · 38ms

03You talk · it talks back

It hears you, and coaches the slip.

4/7Say itMark complete

Say it

At the ticket desk in Madrid. Say each one out loud.

¿Dónde está la estación de tren?

Where is the train station?

/ˈdon.de esˈta la es.taˈθjon de tren/

Speak

100%heard: “dónde está la estación de tren”

¿Dóndeestálaestacióndetren?

That’s the one: es-TÁ, “is”. Next phrase.

The nine types

Multiple choice is one of nine.

Recognition is the easiest thing to fake your way through, which is why a generator that only makes multiple choice can tell you almost nothing.

Multiple choice, and multi-select
The two everyone generates. Worth having, and worth not over-using: they test recognition, which is the easiest thing to pass without knowing anything. Multi-select is markedly harder because you cannot eliminate.
Cloze, ordering and numeric
Fill the gap in a sentence. Put five steps in the right sequence. Give a number, checked with a tolerance so 9.81 and 9.8 are both right. All three are objectively markable and none of them can be guessed from the options.
Short answer and essay
You write it in your own words and it is marked against a rubric that came with the question, point by point, with written feedback saying which part you missed. This is the type most quiz generators cannot do, because the marking is the hard half.
Code, run against tests
JavaScript, Python, SQL and HTML run in your own browser and pass or fail against real tests. Nothing is asked to judge whether your code looks right — it either produces the expected output or it does not.
Projects, against a rubric
A piece of work rather than an answer, assessed on stated criteria. What you use when the thing being learned only shows up at the scale of a whole task.

How marking works

The hard half is the marking.

Generating a question is easy. Deciding honestly whether an answer was right, and saying why it was not, is the part that makes practice worth doing.

Six of the nine cannot be marked wrongly
Multiple choice, multi-select, cloze, ordering, numeric and code are all checked by code, not by a model — against the recorded answer, the sequence, the tolerance, or the test suite. There is no judgement involved, so there is nothing to hallucinate.
The other three ship a rubric
A short answer, essay or project is generated together with the criteria it will be marked against, before you ever see the question. Marking then happens against that rubric rather than against a fresh opinion, which is what stops the same answer scoring differently twice.
You get told what was wrong with it
"Incorrect" teaches nothing. Every marked answer comes back with which part was missing or mistaken, so the attempt is worth more than the score. This is the whole reason to answer a question rather than read the answer.
What you got wrong comes back
A question you missed is not just logged, it is scheduled — returned for review before the material decays, and more often than the things you got right. A quiz you take once and never see again is a measurement, not practice.

Questions

Questions about questions.

The six we hear most. The intro film answers most of them in a minute and a half.

Is the AI quiz generator free?

Yes, within the same limits as the rest of the product: 12 fully-written lessons a month on the free plan, and the questions inside those lessons come with them at no extra cost. Answering, grading and review are never metered — you can retake a quiz as many times as you like, and every attempt is marked in full. Pro raises the writing allowance to 150 lessons a month.

How is a written answer actually graded?

Against a rubric that was generated with the question, not with the answer. The criteria exist before you attempt it, so grading is a comparison rather than an opinion formed on the spot — which is what keeps the same answer from scoring differently on two runs. You see the rubric after marking, along with which specific criteria you met and missed.

Can I generate a quiz from my own material?

Yes. Upload a PDF, a slide deck, your notes or a transcript, and the questions are built from what those documents say rather than from general knowledge about the subject — which is the difference between a quiz that prepares you for your exam and one that prepares you for a similar exam somewhere else.

Can it make an exam rather than a quiz?

Yes — longer, mixed-type assessments over a whole section are one of the lesson formats, distinct from the short checks inside a single lesson. A section exam draws on everything up to that point rather than only the last thing you read, which is a much better predictor of whether you actually know the material.

Can I export the questions?

You can publish a course and share it as a link that anyone can read without an account, which covers handing a quiz to a class or a colleague. There is no question-bank export to CSV or QTI today — if you need one for an LMS, say so; it is a reasonable thing to want and knowing who wants it is what decides whether it gets built.

How many questions does it generate per lesson?

As many as the lesson needs, which is usually three to eight, and the types are chosen by what the lesson teaches rather than by a quota. A lesson about a procedure gets ordering and a checklist; a lesson about two things people confuse gets discrimination questions; a lesson about a calculation gets numeric answers with tolerances. A fixed "five multiple choice per lesson" is the pattern this deliberately avoids.