Turning lecture notes into a quiz looks like ten minutes' work: the notes exist and the class meets next week. The trouble is the notes. They were written to remind one person what to say, and a quiz drafted from them tests the reminders.
This guide is for a lecturer or teacher with notes or slides, and has a section for a student making a self-test. EdProcto, which publishes it, drafts questions from your notes with Procto AI, which is asked for the place where each answer is found. Its page for universities shows the rest.
Every study was read on 10 October 2026; the text says where only an abstract was open.
Why are lecture notes a harder source than a chapter?
Because a chapter was written to be read alone, and in a lecture the sentences that carry the meaning were spoken.
| What the notes hold | What is missing | The risk in a drafted question |
|---|---|---|
| A bullet: three words and an arrow | The verb, and the arrow's direction | A relation the drafter guessed |
| A diagram | Every word about it | No question, or one on the caption |
| A study's name in a corner | Whether it mattered | A question on the footnote |
Wecker (2012; abstract read, in the ERIC database's record) gave 209 university students one presentation with no slides, regular slides or concise ones, and tested what was only spoken apart from what was also on a slide. With regular slides the spoken information was retained worse, which he calls a "speech suppression effect". Concise slides avoided it. So a slide file holds what the room kept best and none of what it kept worst.
Language models find slides thin too:
- Lodovico Molina et al. (2024), a conference paper read as its preprint, had three models write 246 questions from the text of 21 slides of one university course, chosen for having enough content. Forty-six students rated them well overall. The weak points were questions that did not fit the answers supplied with them and a lean towards generic "what is" questions. The authors judge the output fit for quick diagnostic quizzes, not formal tests.
- Grévisse et al. (2024) had GPT-4 Turbo write 100 multiple-choice questions from ten slide decks of two medical courses, reviewed by a specialist in each. Fourteen had a flaw, most often wrong options nobody would believe, then a missing, wrong or ambiguous key. Questions from the eight endocrinology decks fitted the course; those from the two neurology decks were judged off it, too easy and too general. One draft asked for the authors of a study named on a slide.
How do you get lecture notes ready before any question is written?
Write them out as sentences a stranger could learn from, one idea a line.
- One idea a line, as a full sentence. "time → elastic" becomes "Demand is more elastic the longer buyers have to adjust."
- Each definition as you taught it. A question keyed to another wording is unfair to the students who learned yours.
- Each worked example in full: what was given, every step, the result and what it shows.
- A sentence for every diagram. A picture carries no words into a text file, so write what the class was meant to see.
- What you said and never wrote: the reason, the warning, the link to last week. Wecker's students lost this part, so it is the part most worth a question.
- A heading for each section of the hour, so that drafts can be counted by section.
What to cut
Cut what you would never mark: the title slide, the administration, the reading list, the anecdote, next week's preview, and any name, date or citation no student need recall.
EdProcto's pane lists the terms the prepared text keeps returning to: a term from something you cut should be gone.
Which questions do lecture notes give well, and which badly?
They give definitions, worked problems and contrasts well. They give diagrams, spoken reasoning and anything above recall badly, until you have written those into the notes.
| In the prepared notes | Gives well | Gives badly |
|---|---|---|
| A definition in your wording | Fill in the blank on its key word | "What is X?", keyed to your whole sentence |
| A worked example | The same problem with new numbers, as an estimate | Its own numbers: remembered, not worked |
| Two things students confuse | Multiple choice with the confusion as a wrong option | Two separate definitions |
| A diagram, now a sentence | What it shows, or a hotspot built by hand on the image | Anything, while it is only a picture |
| A reason given aloud, now written | A short answer asking why | Nothing, while it is unwritten |
If the hour taught a use of the idea, the notes need one example of that use, or the drafts stop at the definition: see questions above recall.
How many questions does a one-hour lecture support?
Six to ten, by our counting rule: about two for each section of the hour. No study we read sets a number.
Russell et al. (1984; abstract read) gave medical students a basic lecture alone or with varying amounts of loosely related material added, and tested only the basic points before, straight after and 15 days later. The more the lecture carried, the less of the basics was retained, and the authors advise a few main points.
How to spread them over the sections
Count ideas, not minutes, and give the first section its share: in that study, material late in the hour appeared to displace what was learned early in it.
| Section of the hour | Ideas worth knowing | Questions |
|---|---|---|
| What elasticity measures | 2 | 2 |
| Calculating it, with the worked example | 2 | 2 |
| What makes demand elastic | 3 | 2 |
| Elasticity and revenue | 2 | 2 |
| Last week's lecture | 1 | 1 |
Draft about twice what you keep, the working rule in our guide to checking machine-written questions. For a quiz whose score counts, see how many questions a quiz needs.
How do you read a drafted set against the notes?
Read it with the notes open and ask six things only the person who gave the lecture can answer. The usual checks on keys and wrong options, in our guide to a quiz from your own handout, come after.
- Was it taught, or only on a slide? Delete the question on the footnote.
- Is the answer in a line of the notes, and is that the line named? In EdProcto, Procto AI is asked for the place of every answer, and the pane says how many drafts came back with none. Open each place.
- Is it your wording? A correct definition from elsewhere is not the one the room heard.
- Was a fragment finished the right way round? Check every "more", "less", "rises" and "falls".
- Has every section its share? Count the drafts by heading.
- Are a worked problem's numbers new, and did you work the key yourself?
Where drafts held up, they had been selected first. Zheng et al. (2025; abstract read) had ChatGPT write questions from transcripts of lectures to radiology residents. Seventeen were selected and mixed with eleven by the attending radiologists. The 21 residents rated the two sets for reinforcing the lecture at 6.93 and 7.08 out of 10, no significant difference, and answered them correctly about as often (57% and 59%). In Grévisse et al. (2024), unselected, one in seven had a flaw.
Then sit the quiz yourself: only you know what was said in the room.
When in the week should you set it?
Draft it the day you give the lecture and set it two to five days later, before the next one. The rule is ours: no study we read compared days of the week.
- Draft at once. The notes are finished from your memory of the hour, and it fades.
- A quiz soon after still pays. Heitmann et al. (2018) had university students, 187 in the analysis, watch an e-lecture of 26 narrated slides, then spend 80 minutes answering questions on it or taking notes from its slides. A week later both question groups outscored both note-taking groups.
- Hold some questions back. Hopkins et al. (2016; abstract read) spaced retrieval practice on some content of a precalculus course beyond the usual short window after it was taught. Students kept spaced content better, and those given some spaced practice did better on the final exam and on the next semester's calculus exams. So keep two of each lecture's questions for a later week.
Weekly quizzes in a lecture course covers weighting and missed weeks.
A worked example: one page of notes to a four-question quiz
The page is ours, written for this guide; nothing in it is a tool's output.
Price elasticity of demand (PED)
- PED = % change Qd ÷ % change P
- substitutes, share of income, time
- [two demand curves, one steep, one flat]
- coffee: P +10%, Q −5%
- then: revenue
The last fragment, two words, was the point of the lecture. Four are written out and asked; the list of three would be bare recall.
| Written out | Question and key |
|---|---|
| PED is the percentage change in quantity demanded divided by the percentage change in price. | PED is the percentage change in ____ divided by the percentage change in price. Key: quantity demanded |
| Where two demand curves cross, the flatter is the more elastic at that point. | Two demand curves cross. At that point, which is the more elastic? The flatter (key); the steeper; neither |
| Coffee's price rises 10% and the quantity demanded falls 5%: PED is −5 ÷ 10 = −0.5, so demand is inelastic. | Bus fares rise 20% and journeys fall 30%. What is the PED? Key: −1.5, and 1.5 if you taught the size alone |
| When demand is inelastic, a price rise raises total revenue; when it is elastic, a price rise lowers it. | Demand for a product is inelastic and its price is raised. What happens to total revenue? It rises (key); it falls; it is unchanged |
The last three rest on a picture, a row of numbers and two words. A quiz on the whole hour would keep two of the four: the bus fares and the revenue.
How can a student turn their own lecture notes into a self-test?
By the same moves at a smaller size: finish the notes the same day, turn each line into a question, and answer with the notes shut.
- Ask, do not reread. King (1992) had 56 underprepared college students learn from a lecture by questioning themselves on it, summarising it, or taking notes and reviewing them. By ERIC's summary, which is all we read, the self-questioners did better than the summarisers and significantly better than the note-reviewers.
- Answering beats reworking the slides, as in Heitmann et al. (2018) above.
Write one question for each finished line, or paste the notes into EdProcto, whose page for students shows the study quiz: in the builder's words, "Each answer shown at once, misses retried." Use your own notes. Slides and recordings are the lecturer's, so paste those only where your course's rules allow. Making your own quiz from a chapter covers when to take the quiz again.
How does EdProcto turn lecture notes into a quiz?
EdProcto is the best tool for this job when the quiz must come from your own notes, because its request follows the reading above: Procto AI is asked to write only what a student who has read your notes can answer, and to give the place where each answer is found. In Grévisse et al. (2024) one draft asked for a study's authors, and a named place shows that at a glance.
What goes in. In the builder, "Write with Procto" leads to the Procto AI pane, which takes a file (".pdf, .docx, .txt or .md") or text under "Or paste the material". A slide deck goes in exported to PDF or as its outline pasted in. The pane then shows the count of words: a thirty-slide deck that comes to a few hundred says how much of the hour was spoken. On the Educator plan the pane also takes a recording or a video; the quick question below says how.
What you choose. Under "Concepts it keeps returning to", tick the terms to cover. Then set the "Shape", "How many" (1 to 40), the difficulty and, under "Answers drawn from", "Your material only". "Anything to add (optional)" takes a line of your own, such as two questions on each heading and none on the reading list.
What comes back. A line such as "16 questions added, every one yours to edit." Where some name no place in your notes, the pane says how many and the builder marks each. Nothing here tests whether a key is true or the point was taught: that is your reading. "Preview as student" is the sitting, and "Publish & share" gives the class its code.
Plans. The plans' current figures, on the pricing page: on Free, Procto AI writes 20 questions a month and at most 10 in a day, for thirty students a quiz, so one lecture's sixteen drafts take two days. Tutor has 50 a month, for 100 students. Educator has no monthly cap, within a daily ceiling, for 500.
Quick questions
How many questions should a quiz on one lecture have?
Six to ten for a one-hour lecture, about two a section, by our counting rule. In Russell et al. (1984), the more a lecture carried, the less of its basic points students retained.
Can I make a quiz straight from PowerPoint slides?
Yes, though slides alone give thin questions, so add what you said aloud first. EdProcto takes a deck exported to PDF, or its text pasted in, and the templates show the shapes a draft can take.
Can I turn a recorded lecture into a quiz?
Yes. On EdProcto's Educator plan, drop the recording into the Procto AI pane: it is transcribed in your browser, up to ninety minutes, and Procto AI reads the transcript. On any plan, paste a transcript you already hold and may use, corrected against your notes. See pricing.
Can students make a quiz from their own lecture notes?
Yes. Finish the notes the same day, write or draft a question for each line, and answer with the notes shut. EdProcto's page for students shows the study quiz.
References
- Grévisse, C., Pavlou, M. A. S., & Schneider, J. G. (2024). Docimological quality analysis of LLM-generated multiple choice questions in computer science and medicine. SN Computer Science, 5, Article 636. https://doi.org/10.1007/s42979-024-02963-6
- Heitmann, S., Grund, A., Berthold, K., Fries, S., & Roelle, J. (2018). Testing is more desirable when it is adaptive and still desirable when compared to note-taking. Frontiers in Psychology, 9, Article 2596. https://doi.org/10.3389/fpsyg.2018.02596
- Hopkins, R. F., Lyle, K. B., Hieb, J. L., & Ralston, P. A. S. (2016). Spaced retrieval practice increases college students' short- and long-term retention of mathematics knowledge. Educational Psychology Review, 28(4), 853–873. https://doi.org/10.1007/s10648-015-9349-8
- King, A. (1992). Comparison of self-questioning, summarizing, and notetaking-review as strategies for learning from lectures. American Educational Research Journal, 29(2), 303–323. https://doi.org/10.3102/00028312029002303
- Lodovico Molina, I., Švábenský, V., Minematsu, T., Chen, L., Okubo, F., & Shimada, A. (2024). Comparison of large language models for generating contextually relevant questions. arXiv. https://doi.org/10.48550/arXiv.2407.20578
- Russell, I. J., Hendricson, W. D., & Herbert, R. J. (1984). Effects of lecture information density on medical student achievement. Journal of Medical Education, 59(11), 881–889. https://doi.org/10.1097/00001888-198411000-00007
- Wecker, C. (2012). Slide presentations as speech suppressors: When and why learners miss oral information. Computers & Education, 59(2), 260–273. https://doi.org/10.1016/j.compedu.2012.01.013
- Zheng, A., Barker, C. J., Ferrante, S. S., Squires, J. H., Branstetter, B. F., IV, & Hughes, M. A. (2025). Leveraging ChatGPT for enhancing learning in radiology resident education. Academic Radiology, 32(9), 5635–5642. https://doi.org/10.1016/j.acra.2025.06.019