Make flashcards and practice exams from lecture notes with AI
Testing yourself is one of the most reliable ways to remember things. A chatbot with your notes can write the questions in seconds. Making sure they're good questions with correct answers is still on you.
Start from your notes, not the chatbot's memory
If you ask for "flashcards on cell membranes" with nothing attached, you get generic textbook cards. Some will use terms your course doesn't, some will cover things your lecturer skipped, and some will miss what your lecturer kept coming back to.
Give it the week's material instead. Convert the slides and readings with CourseFeed, paste or attach the result, and tell the chatbot to use only that. Every file in the document starts with a # Source: heading, and slides keep their slide numbers, so you can say exactly which file the cards should come from and trace each card back to where it came from.
Prompts for flashcards
This one works for most courses:
"Using only the Source section for Week 5 slides.pptx, make 20 flashcards. One fact per card. The front is a question, not just a term. The back is the answer in one or two sentences, using the wording on the slides. End each back with the slide number in brackets."
Every part of that prompt has a reason. One fact per card, because a card that asks for six things gets marked half right forever. If something really is a list, like the stages of a process, ask for one card per stage plus one for the order.
A question on the front makes you recall more. "What does the sodium-potassium pump move, and in which direction?" is harder than just "Sodium-potassium pump", and that's the point.
The slides' wording matters because your exam uses your course's terms, not the chatbot's. And the slide number lets you check a card later without hunting for it.
For definitions and key phrases, cloze cards (fill in the blank) are often quicker to review. Anki marks the hidden part like this: The {{c1::resting potential}} is about -70 mV. Ask for "10 cloze cards in Anki cloze format" and you'll get that syntax.
Only ask for cards in both directions when it makes sense, like vocabulary in a language course. For most subjects a reversed card ("-70 mV" → "?") isn't a useful question.
Getting the cards into Anki
Anki imports plain text files with one card per line and the fields separated by a tab, a semicolon or a comma. Roughly:
- Ask for the right format: "Put the cards in a code block, one card per line, with a tab between the front and the back. No header row and no numbering."
- Copy the code block into a plain text editor like Notepad, or TextEdit in plain text mode, and save it as a .txt file with UTF-8 encoding.
- In the Anki desktop app, go to File, then Import, and pick the file. Check the preview: the separator, the note type (Basic for question cards, Cloze for cloze cards), the deck, and which column goes to Front and which to Back. Then import.
Tabs sometimes turn into spaces when you copy out of a chat window. If the preview puts the whole line in one field, ask for semicolons instead, with each field wrapped in double quotes so a semicolon inside an answer doesn't split the card. Some chatbots can also give you a downloadable .txt or .csv file directly, which skips the copying problem completely.
Anki's menus move around between versions, so if something doesn't match, check the importing section of Anki's manual. Quizlet and other flashcard apps have their own import options, and most will take a pasted list with a separator between term and definition.
Practice exams
Flashcards drill single facts. A practice exam checks whether you can actually use them, and it's as close as you'll get to the real thing before the real thing.
Something to start with:
"Write a 30-minute practice test on the Week 5 slides and reading: 8 multiple-choice questions, 4 short-answer questions and 1 longer question that needs ideas from two different slides. Don't show the answers. When I reply, mark my answers against the notes and give the slide number for every correction."
To make it closer to your actual exam:
- Copy the format. Your syllabus usually says what the exam looks like, so tell the chatbot: "The midterm is 40 multiple-choice questions and two short essays."
- Use past papers as a model. If your course gives you past questions, paste one in and ask: "Write five new questions in this style on this week's material."
- Go one at a time. "Ask me one question at a time and wait for my answer" stops you from reading ahead, and you get feedback while the question's still fresh.
- Ask for better wrong answers: "Each wrong option should be a real misconception, not something obviously wrong. Don't use 'all of the above', and keep the options about the same length." Chatbot-written multiple choice often gives the answer away by making it the longest or most detailed option.
Check the answers before you learn them
AI-written questions go wrong in a few predictable ways. The answer key can be wrong. Two options can both be right. A question can test something that isn't in your notes. Or the answer comes from general knowledge and clashes with how your course does it, with a different notation, sign convention or definition from your lecturer's.
A wrong flashcard is worse than no card, because you'll practise the mistake every few days. So:
- Use the slide numbers. Check a handful of cards against the slides before you import, and check every card you get wrong while reviewing.
- Ask for a second pass in the same chat: "Check each card against the notes. List any whose answer isn't directly supported by the notes, and say why." It catches some mistakes, but it's the same chatbot checking its own work, so it doesn't replace you looking.
- Fix or delete bad cards as soon as you find them. Anki lets you edit a card during review, or suspend it to get it out of the way.
The other common problem is questions that are too easy. Chatbots like writing questions you can answer just by recognizing a phrase from the slides. If you're getting everything right, ask for harder ones: "Write questions that apply these ideas to new examples that aren't in the notes", or "Ask why and what-would-happen-if questions, not definitions."
Spaced repetition, briefly
Cards work best reviewed a little at a time over weeks, not crammed in one night. The idea behind spaced repetition is to review each card right around when you'd otherwise forget it. Each time you remember it, the next review gets pushed further out. Anki does the scheduling for you, as long as you're honest when you mark your answers.
A few habits that help:
- Make each week's cards in the week of the lecture, so they have time to space out before the exam.
- Only make cards for what you actually need to know. Reviews pile up, and a few hundred new cards a week turns into an hour a day pretty quickly.
- Keep one deck per course, so you can study one without the others mixed in.
- Do a practice exam at the end of each week, and mixed ones covering several weeks as the exam gets closer.
Check your course's rules
Making study material from your own course notes is usually fine. Some courses, though, don't allow uploading lecture material to AI tools, or sharing it, and a public Quizlet set built from your lecturer's slides can break both rules. If you're not sure, check the syllabus or ask. How to study a week of lectures with AI covers the rest of a week's routine, and summarizing lecture slides is a good step to do before this one.