Flashcard makers: what a good one does to your cards (and the cards it should refuse to make)

A flashcard maker is only as good as the cards it produces: one idea per card, a question on the front that forces recall, a short answer you can check in two seconds, a source you can open, and nothing invented. Judge any maker by running six checks on ten of its cards before you study them. Keep the cards it got right, rewrite the rest, and export the deck so it stays yours whatever tool you use next. Here are the six checks, the cards a generator should refuse to make, and ten generated chemistry cards checked in front of you.

Generated cards against written cards: what the research says

Start with the honest split, because it decides how to use any flashcard maker.

Writing a card teaches a little by itself. Slamecka and Graf (1978) showed what they named the generation effect: words people produced themselves were remembered better than words they only read. Writing "What does the mitochondrion make, and from what?" on a card is a small act of producing, and it leaves a small trace. A generated card leaves no trace at all until you study it, because you did not produce anything. You read it, if that.

Most of the learning, though, is in the studying, not the writing. Roediger and Karpicke (2006) had students either reread a passage or recall it from memory; a week later the recallers remembered more, even though the rereaders had felt more confident at the time. The attempt before the flip is the learning. Kornell (2009) then took the same question to flashcards. Learners who spaced their cards out beat learners who massed them for about nine in ten participants, and most of them believed the opposite about which had worked. Massed study feels better and tests worse.

So the honest conclusion: a generator saves the typing, not the thinking. The thinking is in the attempt before the flip, and the dates between flips. Those are the subject of the active recall guide and the spacing guide, and chapter 2 of the free playbook has the schedule on one page. Everything below is about the cards themselves, which a generator can get right or wrong before any studying happens.

The six checks to run on any flashcard generator's cards

Take ten cards from the maker's output at random. Run each check on each card. The rules come from Wozniak (1999), whose twenty rules for writing knowledge are the oldest and clearest statement of what a card should be, and from what the research on recall needs a card to do.

1. One idea per card. A card with a paragraph on the back tests nothing, because you cannot tell which part you missed. Fails: front "Describe VSEPR theory", back a six-line paragraph. Passes: front "What does VSEPR theory say electron pairs around an atom do?", back "they spread out to be as far apart as possible".
2. A question on the front, not a keyword. A label is not a prompt. "Chain rule" on a front tells you nothing to retrieve. Fails: "Le Chatelier's principle". Passes: "If you add more reactant to a system at equilibrium, which way does it shift?"
3. Nothing on the front that gives the answer away. Fails: "Define a strong acid: an acid that dissociates completely in water". Passes: "What does a strong acid do in water that a weak acid does not?"
4. A back you can check in two seconds. A definition, a formula with its symbols named, a one-line step, a number with its unit. Fails: a back that needs a paragraph to compare against. Passes: "109.5°".
5. A source you can open. The page, the minute of the recording, the slide. A card that cannot be traced cannot be checked, and a card that cannot be checked will be learned wrong if it is wrong.
6. Nothing invented. The card's fact is in your material, not in the model's memory. A generator given your chapter should write cards about your chapter. A card about a topic the chapter never mentions is the model filling a gap.

Score the ten. If more than two fail, the maker is writing summaries in the shape of cards, and its output needs rewriting before it is a deck. How to write a card that teaches is the guide to the rewriting.

The cards a generator should refuse to make

A good maker refuses some cards. A student checking generated cards should delete the same ones.

• A card from a sentence that is not a fact. "The teacher said this is important" is a remark, not knowledge. Nothing to recall.
• A card with a multi-part list on the back. "List the five electron-domain geometries and their angles" is five cards pretending to be one. Split it.
• A procedure as a card. "How do you solve a related-rates problem?" has a back nobody can check in two seconds, and knowing the steps in words is not the same as doing them. A procedure is a problem to redo, cold, from the question alone, and the maths study guide shows how.
• A card from a source the maker was not given. If you uploaded chapter 9 and a card is about chapter 12, the model reached into its own memory. Delete it, whatever it says.
• A duplicate of a card you already have. Two cards for one fact means two reviews for one memory and a schedule that lies about how much you know.
• A card whose answer is the question in other words. "What is a polar molecule?" with "a molecule that is polar" on the back. It happens more than you would think.

Verify and cite: the two things that make generated cards safe

Two rules, and they are not negotiable.

The fact on the card must be in your source. The card must say where. That is checks 5 and 6 again, and here is why they matter more for cards than for any other kind of note.

A spaced schedule is a machine for strengthening whatever you put in it. You see a card, you attempt it, you check the back, and the gap before the next review grows. Over a month the fact on the back is rehearsed into memory on a schedule designed to make it stick. If the back is wrong, the same machine makes the wrong fact stick, with the same efficiency. This is our reasoning rather than a cited result, but it follows from what spacing does: the schedule does not know which cards are true. A wrong card is the worst place an error can live, because you will meet it again and again and feel more sure each time.

So verify before the first review, not after the tenth. The verify rule for generated notes is the same as for generated cards, and the guide to notes from a lecture or a PDF explains why it has to happen within a day: after that, the memory of the class that would have caught the error is gone.

Worked example: ten generated cards from a chemistry chapter, checked

A Grade 12 chemistry (SCH4U) student gives a flashcard generator the chapter on molecular shape and polarity, the ninth idea in the course. The maker returns a set. Here are ten of its cards, in our wording, with the six checks run on each. The facts on the passing cards are the standard ones from the chapter's shapes table, and we re-derived each before writing it down.

Card (front → back)	Result	Why
"VSEPR" → a paragraph on the theory	Delete	Fails check 2 (a keyword front) and check 1
"What is the bond angle in a tetrahedral molecule?" → "109.5°"	Keep	Passes all six; the chapter's shapes table is the source
"Why is CO₂ non-polar although each C=O bond is polar?" → "it is linear, so the two bond dipoles point opposite ways and cancel"	Keep	One idea, checkable, in the chapter
"List the five electron-domain geometries and their angles" → five lines	Split	Fails checks 1 and 4; becomes five cards
"Is NH₃ polar?" → "yes: trigonal pyramidal, the dipoles do not cancel"	Keep	Short back with its reason
"What shape is a water molecule, and roughly what is its bond angle?" → "bent, about 104.5°"	Keep	Two linked facts the chapter states together
"Name the shape of a molecule with six electron domains and two lone pairs" → "square planar"	Keep	Checkable against the table
"What did the lecturer emphasise?" → "shapes come up on every test"	Delete	Fails check 6 (not a fact) and check 5 (no source)
"How many lone pairs does the central atom in SF₄ have, and what shape results?" → "one; seesaw"	Keep	Re-derived below
"VSEPR stands for ___" → "valence shell electron pair repulsion"	Keep	A fill-in-the-blank card passes; the blank is the question

The tally: seven to keep, one to split into five, two to delete. Eight of ten cards survive in some form, which is a good maker. The two deletions are the ones that would have cost the most: a paragraph card that tests nothing and a card with no fact on it.

The re-derivation for the SF₄ card, because every number on a kept card gets one. Sulfur has six valence electrons. Four of them are used in the four S–F bonds, leaving two electrons, which is one lone pair. Four bonding domains plus one lone pair is five electron domains, which is the trigonal bipyramidal arrangement; with one lone pair in the equatorial position, the molecular shape is the seesaw. The card is right. The square planar card: six domains with two lone pairs on opposite sides of an octahedral arrangement leave four atoms in a plane. Right. The tetrahedral angle of 109.5° and water's angle of about 104.5° are in every chemistry data table, and the "about" on the water card matters, because the chapter gives it as an approximation and so should the card. Every shape, with its angle and the reasoning behind it, is in the molecular geometry and bond angles guide.

Before any of the seven are studied, one more pass: each card gets its source written on it. "Chapter 9, shapes table, p. 3 of the handout." Check 5, done by hand in two minutes.

Importing a set you already have: make flashcards online without retyping

Most tools import a CSV file or a common deck format, so a set you made last year or a set a classmate shared does not need retyping. Import it. Then run the six checks on a sample of it too.

The reason is the one from the verify section. A set written by someone else carries someone else's errors, and you cannot see them, because you did not write the cards and you have not studied the material yet. A classmate's deck with one wrong formula in it will teach you that wrong formula on a schedule. Sample ten, check six, fix what fails.

Two questions to ask any maker about importing: can I import without retyping, and can I edit a card after it is in? A maker that imports but will not let you edit has made the errors permanent.

Exporting so the cards stay yours

One thing is non-negotiable, and it is not about the cards. The maker must export your deck in an open format: CSV, Markdown, or the .apkg deck format that many study tools read. If it does, a change of tool next year costs you an afternoon. If it does not, a change of tool costs you every card and every review you ever did. Cards you cannot export are rented, not owned, and the rent is the months of reviews you would lose by leaving.

Test the export in your first week, with a small deck, before the deck is big enough to matter. Open the exported file and read it. If the questions and answers are there in plain text, you are safe.

Free flashcard makers: what "free" should include

A free online flashcard maker should let you do the three things that are the method: make cards, study them with an attempt before the flip, and have them come back on a spaced schedule. It should also let you export, for the reason above. If the free tier does those four things, it is a complete study tool, and the research says the learning happens there.

What is reasonable to pay for: generating cards from long recordings or large files, storage for many sets, a voice that reads the cards, a tutor behind the deck. These cost the maker something per use, and a fee for them is fair.

The one test for any free tier: can you leave with your cards? A free tier that spaces reviews but will not export is a trap with a schedule. The same test, with nine others, is in the ten-point study-app checklist.

The six checks on one page

1. One idea per card.
2. A question on the front, not a keyword.
3. Nothing on the front that gives the answer away.
4. A back you can check in two seconds.
5. A source you can open.
6. Nothing invented.

Sample ten cards; more than two fails means rewriting before studying. Then the two non-negotiables: verify every kept card against your source before its first review, and export the deck in an open format in your first week. A generator saves the typing. The attempt before the flip, on a spaced schedule, is still yours to do.