How to Generate a Quiz with AI: Turn a Lesson or PDF into Questions

To generate a quiz with AI, choose a specific learning objective, provide an approved topic or clean lesson extract, and treat the output as a draft. Verify each key, alternative and explanation against the source before testing the learner view. This guide compares input choices, repairs a biology example and shows a complete three-question review workflow.

Three ways to feed the AI

When you generate a quiz with AI in Skolina, you start from one of three inputs, and the choice shapes the result more than any other setting.

• A topic. Type a subject line such as “the water cycle for Year 7” or “the causes of the First World War.” This is the fastest route, but also the loosest: the AI draws on general knowledge, so it may not match the vocabulary or emphasis of your own course.
• A pasted lesson. Copy the text of the lesson you actually taught and paste it in. The AI now works from your material, so questions tend to mirror what your Students saw in class, including the examples and terms you used.
• An uploaded PDF. Drop in a worksheet, a chapter, or a set of notes. This is handy when your content already lives in a document, and it keeps the generated questions anchored to that source.

A specific source makes alignment easier to check; it does not guarantee that the model will use it correctly. Inspect extracted text from a PDF, especially columns and symbols, and remove material the class has not studied. The PDF-to-quiz guide covers that extraction stage in more detail.

Tasks to request in an AI quiz draft

Request outputs you can check against the lesson rather than relying on a general promise of quality:

• First drafts at speed. It is very good at producing a batch of questions you can then prune and refine, which beats staring at a blank page.
• Variety of formats. Rather than defaulting to single-choice every time, you can ask for a mix across the eight question types: single and multiple choice, true/false, matching, fill-in-the-blanks, numeric, ordering, and open response. That variety alone makes a Quiz feel less monotonous.
• Recall and comprehension questions. Ask for a definition, a distinction or a sequence already present in the source. Check whether the question tests that content rather than a detail the model supplied itself.
• Rephrasing and distractors. It can suggest plausible wrong answers (distractors) for a multiple-choice item, which is often the hardest part of writing one well.

Where the AI slips

Knowing the failure modes is what separates a useful draft from a liability.

• Confident errors. The AI can state something false with total assurance, especially on niche or very recent topics. It does not “know” it is wrong.
• Shallow distractors. Wrong answers can be too obviously wrong, or accidentally also correct. Either way, the question stops measuring anything.
• Misaligned demand. A prompt labelled “analyse” may still ask for a definition. Inspect what the learner must do with the passage or data, rather than accepting the generated Bloom label.
• Ambiguity. A question that reads fine to the machine can be genuinely ambiguous to a fourteen-year-old. The AI rarely catches this.

If you want to sharpen the questions the AI hands back, our guide on how to write good multiple-choice questions covers distractors, stem clarity, and the common traps in detail.

Why human review is non-negotiable

This is the single most important habit to build: the AI produces a draft, never a finished Quiz. Always read every question before it reaches a Student. Check three things on each one — is the answer key actually correct, is the wording unambiguous, and does the question target what you meant to assess? Some repairs are small; others require replacing the item. Include that verification time when deciding whether generation helps your preparation.

A before-and-after example

To make that concrete, here is a fictional draft constructed to illustrate an option-quality problem in a biology lesson:

Which gas do plants absorb during photosynthesis? Oxygen — Carbon dioxide (marked correct) — Water — Sunlight

The answer key is right, but the alternatives are inconsistent. Sunlight is not a gas, and “water” leaves its state unspecified rather than naming a gas in the same way as oxygen or carbon dioxide. A learner can eliminate weak alternatives without resolving the intended photosynthesis distinction. Rewrite the options as named gases:

Which gas do plants absorb from the air during photosynthesis? Oxygen — Carbon dioxide (correct) — Nitrogen — Hydrogen

Now every option is a gas, so eliminating non-gases no longer supplies the answer. The question is specifically about photosynthesis: plants also carry out cellular respiration, which uses oxygen. If a learner chooses oxygen, ask which process they were thinking of before diagnosing a misconception. This revision improves the question’s focus; it does not establish a measured gain in learning.

Try that review on one lesson you already know well: create a free teacher account, generate a five-question draft, then mark each answer key and distractor as keep, rewrite or discard. The useful result is a trustworthy Quiz, not the number of questions produced.

This is also the honest case for and against the technology. If you want the broader picture of where automation helps and where it falls short in the classroom, we explore that in the opportunities and limits of AI in education. The short version for quiz-making: treat the AI as a fast but unreliable assistant, not as the author of record.

Review a three-question draft against a short source

Use this fictional lesson extract to make the review reproducible: “In this lesson, distinguish photosynthesis, in which plants use carbon dioxide from the air, from aerobic cellular respiration, which uses oxygen.” The extract supplies a narrow target. It does not justify questions about every plant process or conditions the class has not studied.

Ask for these three tasks explicitly instead of requesting “a difficult biology quiz”. Then solve the returned items without viewing the generated key. For every option, try to identify a condition under which it could also be correct. If the question omits that condition, repair the stem; do not make a reasonable interpretation wrong merely because it differs from the generated key.

Keep a small review record beside the draft: source sentence, intended evidence, checked answer and reason for a change. That record is external editorial work, not a promise of a source-tracking feature in Skolina. If the model invents a quotation, an experiment or a new fact, verify it independently or remove it rather than using the model’s own assurance as a source.

Before release, try one correct response, one plausible wrong response and one alternative wording of the open answer. Check feedback as well as scoring. After the lesson, use a different example to ask which process is being described; repeating the original option order checks memory of that quiz as well as subject knowledge. The Department for Education’s guidance places responsibility for the final material with the professional using it; its England-specific guidance is not an endorsement of a generator.

What it costs, and where your data goes

Two practical points before you click generate.

First, the cost. AI generation is available on every plan while the token balance is sufficient. Text generation costs one token per generated question, so a fifteen-question draft costs fifteen tokens. File extraction can add tokens according to the amount of text processed. New teachers also start with 100 tokens and 14 days of Pro access.

Second, where your data is processed. This matters and deserves a straight answer. Skolina hosts your account data in the EU. The AI generation itself, however, runs on OpenAI’s models, which means the text you submit for generation is processed outside the EU. Do not include sensitive or personal information in that content. It is a deliberate trade-off, not a hidden one.

A sensible workflow

Putting it together, a reliable routine looks like this:

1. Choose a clean source you can verify. A short, legible lesson extract can be easier to audit than a broad topic or an unreadable PDF.
2. Specify the intended evidence. Request formats and difficulty that suit the learning goal, then check the actual demand of every question.
3. Generate, then review every item. Fix the answer key, kill weak distractors, disambiguate wording.
4. Test it on yourself. Sit the Quiz as a Student would; the awkward questions reveal themselves instantly.

If AI generation is brand new to you, it pairs naturally with the fundamentals of creating an online quiz for your classroom, which covers the mechanics of building and sharing one from scratch.

Research and guidance worth checking

UNESCO’s guidance for generative AI in education and research recommends a human-centred, age-appropriate approach and explicit attention to privacy. For the narrow problem of option quality, a recent systematic review of automatic distractor generation shows both how active the field is and why generated distractors still need evaluation. Neither source certifies a particular generated Quiz; your subject knowledge and local policy remain decisive.

Skolina’s Token allowances and plan limits can change as the service evolves, so use the current pricing page as the source of truth rather than relying on an old screenshot or review.

Getting started

Try generation on a short lesson you can check, and compare the drafting plus review work with your normal preparation. New teacher accounts come with 100 tokens and 14 days of Pro access, so you can create a free account and generate your first AI quiz without committing to anything. Bring one real lesson, generate from it, and review what comes back: that single pass will tell you more than any feature list. Keep only questions whose sources, keys and feedback you can defend. The same approach works well beyond the classroom: here is how to create a quiz for corporate training and employee onboarding.