How to Make Anki Cards From Lecture Slides with AI (Without a Bloated Deck)

Why make cards from your own lecture slides

Large pre-made decks are built for standardised exams. Most medical school exams are written by your own faculty, and their emphasis, classifications and examples come from their slides. Cards built from those slides test what your exam is likely to test, in the same words. If your school uses faculty-written block exams, the companion guide on how to study for in-house exams covers the rest of the lecture workflow.

The evidence for flashcards is strongest at the level of principles. Retrieval beats rereading (Karpicke and Roediger, 2008). In a randomised study of 80 medical students, Schmidmaier and colleagues (2011) found that re-testing with electronic flashcards beat re-studying at one week, but the advantage had gone at six months: testing needs spacing to last. Spaced questions sent after a clerkship improved end-of-year retention in a randomised trial by Kerfoot and colleagues (2007). Evidence on Anki itself is observational: Deng, Gluckstein and Larsen (2015) found that unique Anki cards seen were associated with USMLE Step 1 scores in 72 students, and Wothe and colleagues (2023) reported an association between daily Anki use and Step 1 scores in 165 students.

Tool options: what kind of tool to use

Most pages ranking for this topic are written by the companies selling the tools, and some quote how many hours the manual method takes without saying where the number comes from. Here is a neutral way to compare the four main routes. The feature that matters most is not speed; it is how quickly you can check each card against its source slide.

Manual or AI, which is faster overall? AI removes the typing, but the pruning and checking stay with you, and for image-heavy lectures the manual route loses less than you would think. The honest test is to time yourself on one real lecture both ways and compare the kept cards, not the generated ones. If you are still choosing a general tool, the AI study assistant guide has a one-chapter acceptance test you can reuse.

The lecture-to-Anki workflow

1. Same day, mark the slides: highlight what the lecturer emphasised or repeated. This is the one input AI cannot infer.
2. Generate a draft from the slides (and the lecture transcript if you have one), asking for short cards with the slide number in an Extra field.
3. Prune: delete trivia, split long answers, merge duplicates, check every number.
4. Add image occlusion for anatomy, histology and pathway diagrams.
5. Tag and import: hierarchical tags such as cardio::pharm::lecture05, then import into Anki.
6. After a week, repair leeches: rewrite or split the cards you keep failing.

If you already have "term - definition" lines, the free slides to Anki converter turns them into an importable file in your browser, without an account. Starting from PDF handouts instead? See PDF to flashcards.

Worked example: one slide, before and after pruning

Example A self-made text slide from a renal pharmacology lecture reads: "Loop diuretics (e.g. furosemide): inhibit the Na-K-2Cl cotransporter (NKCC2) in the thick ascending limb of the loop of Henle. Increase urinary calcium excretion. Adverse effects: hypokalaemia; ototoxicity at high doses." The lecturer stressed the calcium point, because thiazides do the opposite. Here is a typical AI draft and what happens to each card.

Two minutes of pruning turned six cards into four sharper ones, and one of the deletions caught an invented fact. That is the realistic value of AI here: a fast first draft that you are still responsible for.

Card rules, with medical examples

Piotr Wozniak's "minimum information principle" still sums it up: each card should ask one simple thing.

Use cloze for definitions, values and names, basic for "why" and "how", and image occlusion for anything labelled. A deck of only clozes trains you to recognise sentences, not to reason.

Image occlusion for slide diagrams

Since version 23.10, Anki includes image occlusion as a built-in note type, so you no longer need an add-on. AI tools are still unreliable on labelled images, which makes this the step worth doing by hand.

1. Screenshot the diagram from the slide, cropped to the part you need.
2. In Anki, click Add and choose the Image Occlusion note type, then select the screenshot.
3. Draw a shape over each label you want to recall. Each shape becomes its own card.
4. Choose how the other labels behave while you answer (hidden or visible). Hiding them all is harder and better for structures that are easy to guess by position.
5. Add the lecture tag and slide number in the extra field, as for any other card.

CSV or .apkg: how to import

For AI drafts, text files are the practical choice. Recent versions of Anki also read optional headers at the top of the file, such as #separator, #html, #tags and #columns, so a file can go straight into the right deck with the right fields.

• Save as UTF-8 so accents and symbols survive.
• Use one separator (comma, semicolon or tab) consistently.
• Map Front, Back, Extra and Tags in the import window.
• Enable HTML only if your cards contain formatting.
• Import into a lecture sub-deck or tag, so you can suspend or review it as a block.

How many Anki cards per lecture?

A fixed number does not exist, but "it depends" is not an answer either. Use the lecture type as a starting range, then let one week of real review time correct it.

Two controls keep it sustainable. First, cap new cards per day in the deck options: every new card creates reviews for weeks. The Anki manual gives a useful rule of thumb: if you consistently learn 20 new cards a day, expect roughly 200 reviews a day. Second, watch your review time for a week before raising the cap. With FSRS, the scheduler built into Anki since 23.10, the default desired retention is 90%; the manual warns that above 90% the workload increases very quickly, and above 97% it can be overwhelming. To fit reviews around a fixed exam date, see the spaced repetition schedule for exams.

Worked example: a week's card budget

Example A second-year week with five lectures, using the starting ranges above:

Scaling the manual's rule of thumb (20 new cards a day, roughly 200 reviews a day) down to 16 gives on the order of 160 reviews a day once the deck is running. That is a rough proportion, not a prediction: time a few days of your own reviews and multiply. If your average is 10 seconds a card, 160 reviews take about 27 minutes before any new cards. If that does not fit next to lectures, lower the new-card limit or keep fewer cards per lecture; do not start skipping reviews. To place whole topics, not single cards, around the exam date, plan topic reviews around the exam date.

Combining your lecture cards with AnKing

Many US students already use AnKing, a large community deck mapped to board exams. You do not have to choose. A common hybrid is:

1. Keep AnKing suspended by default, then unsuspend only the cards whose tags match the lecture you just had.
2. Make lecture cards only for the gaps: details, classifications or examples your lecturer stressed that the unsuspended cards do not cover. Often that is a handful per lecture, not a full deck.
3. Tag your own cards differently (for example #lecture::renal::07) so you can review them as a block before the in-house exam.
4. When the two disagree, your lecturer's version wins for the in-house exam; note the difference in the Extra field.

Ten-minute quality check before studying a new batch

• Every number matches the slide.
• No card needs the original slide to make sense.
• No answer is longer than a line.
• Near-duplicate cards are merged or deleted.
• Each card has a lecture tag and the slide number.

Leech repair table

A leech is a card you keep forgetting. By default, when a card has lapsed 8 times Anki tags the note as a leech and suspends the card, and the manual's advice is to change how the information is presented, delete it if it is not worth the time, or wait while you learn a confusable card first. In practice, the symptom tells you which fix to use:

How to do this with Memoniq

Memoniq generates flashcards from the slides, notes, recordings or YouTube lectures you upload to a notebook, alongside a summary, written-exam quizzes and a daily spaced-repetition review in the app. On the Student plan you can export the cards for Anki as text or CSV and keep studying them with Anki's scheduler, FSRS included. The first 14 days after sign-up include Student features, with no card required.

Treat the output as a draft: apply the pruning rules above, delete what you do not need and check facts against your slides, because AI can make mistakes.

Limits

Flashcards train recall of facts; they do not train clinical reasoning or explaining a topic out loud, so pair them with questions and oral practice. AI extraction is weaker on image-heavy slides and on anything the lecturer said but did not write. The card ranges above are a starting heuristic, not data. The Anki-specific studies cited are observational and from US schools.

Next read: how to prepare for a medical viva, for the reasoning side of the same lectures.

Sources

• Anki - Changes in 23.10 (built-in image occlusion and FSRS)
• Anki Manual - Importing text files
• Anki Manual - Packaged decks (.apkg)
• Anki Manual - Editing, note types, cloze and image occlusion
• Anki Manual - Deck options and FSRS
• AnkiHub Community - Suspend/unsuspend cards
• Wozniak P. - Twenty rules of formulating knowledge
• Schmidmaier R et al. (2011), Using electronic flashcards to promote learning in medical students: retesting versus restudying, Medical Education
• Deng F, Gluckstein JA, Larsen DP (2015), Student-directed retrieval practice is a predictor of medical licensing examination performance
• Wothe JK et al. (2023), Academic and wellness outcomes associated with use of Anki spaced repetition software in medical school
• Kerfoot BP et al. (2007), Spaced education improves the retention of clinical knowledge by medical students
• Karpicke & Roediger (2008), The critical importance of retrieval for learning, Science