Pick the ideas worth a question
Start from the syllabus or the learning outcomes, not from every line of the slides. For each lecture, list three to five ideas the instructor stressed: definitions that recur, formulas used in examples, distinctions the slides contrast.
If an idea would not appear on the exam, it does not need a question. Fewer, better questions beat a long list you never finish.
Write three types of question per idea
Recall: “What is a confidence interval?” Relationship: “What happens to the interval width if the sample size grows?” Application: “A study reports a 95% interval of 48 to 56. What can and can’t you conclude?”
Recall questions alone overestimate what you know. The relationship and application questions reveal whether you can use the idea, which is what most university exams test.
Keep the answer separate and answer later
Write the answer key on another page or the back of a card. Answer the questions a day or more later, without notes, then mark each one right, partly right or wrong.
Spacing the attempts helps memory (Cepeda et al., 2006). Answering right after writing the question mostly measures short-term recall.
Using AI to generate the questions
An AI tool can draft questions from your notes in seconds. Ask for the three types above and for the passage each answer comes from. Then review: delete questions about details your course does not cover, and fix any answer your notes contradict.
Watch for weak distractors in multiple choice (options nobody would pick) and for questions answerable by grammar alone. Evaluo, our product, generates questions by topic from your file and shows explanations you can check against it.
Try it yourself
Worked example: confidence intervals in introductory statistics
Your lecture notes say: a 95% confidence interval for a mean is roughly the sample mean ± 1.96 × (s / √n). The example: n = 100 students, mean study time 52 minutes, standard deviation 20 minutes.
Answer before you look
Think of your answer first, then open it to compare.
Recall. What does a 95% confidence interval estimate?Show answerHide
A range of plausible values for the population mean, built by a method that captures it in about 95% of repeated samples.
Application. Compute the interval for n = 100, mean 52, s = 20.Show answerHide
Standard error = 20 / √100 = 2; margin = 1.96 × 2 = 3.92; interval ≈ 48.08 to 55.92 minutes.
Relationship. What happens to the margin if n rises to 400?Show answerHide
Standard error = 20 / √400 = 1; margin = 1.96, so the interval halves in width.
Trap. Is there a 95% probability that the true mean is between 48.08 and 55.92?Show answerHide
Under the usual frequentist reading, no: the 95% describes the method across repeated samples, not this one interval.
The calculation, step by step
| Step | Working | Result |
|---|---|---|
| Standard error | 20 / √100 | 2 |
| Margin (95%) | 1.96 × 2 | 3.92 |
| Interval | 52 ± 3.92 | 48.08 to 55.92 |
| With n = 400 | 1.96 × 20 / √400 | margin 1.96 |
Three days later
- 1.You answer all four without notes, miss the trap question and add it to your mistake list.
How to check your work
If your margin came out as 39.2, you used s instead of s / √n. If your interval for n = 400 was a quarter of the width, you divided by n instead of √n.
Have your notes at hand?
Evaluo builds topic-by-topic questions from your PDF or Word file and shows what to review. Free plan: 1 study material per month.
Common mistakes
Turning sentences into fill-in-the-blanks
You end up testing word recognition, not understanding. Rephrase the question so the wording does not give the answer away.
Better: rephrase so the wording does not give the answer away.
Only writing definition questions
They miss whether you can apply the idea. Add at least one relationship or application question per idea.
Better: add a relationship or application question per idea.
Answering right after writing
The answer is still in short-term memory. Wait a day or more before the first attempt.
Better: wait a day before the first attempt.
Trusting AI questions without review
Generated questions can include wrong keys or topics outside your course. Check each one against your notes.
Better: check each answer against your notes.
Frequently asked questions
How many practice questions should I write per lecture?
Three to five key ideas with two or three questions each is usually enough for one lecture. Add more for topics you keep missing.
Are AI-generated practice questions reliable?
They are a useful first draft, not an answer key you can trust blindly. Check each answer against your notes and drop questions on content your course does not cover.
Should practice questions match the exam format?
Mostly yes. If the exam is multiple choice, practice with options; if it is problems, practice full solutions. Keep some open questions anyway, since they show gaps that options can hide.
Continue learning this topic
About this guide
Written by Gonzalo Olmos, founder of Evaluo. Published November 16, 2026.
Sources
References on learning strategies. The original activities in this guide are educational examples and should be checked against your course materials.