AI-made flashcards or your own? What the research says about making cards

Flashcards are among the most popular study tools there is, and making them has never been faster. Paste in your lecture notes, and an app can give you a deck of 50 cards before you’ve finished your coffee.

But if making the cards is part of what helps you learn, handing that job to software has a cost. Recent research suggests it does, and how much depends on how you make them.

What the study found

In 2023, a team led by psychologist Steven Pan published six experiments comparing flashcards students made themselves with premade cards covering the same material (Pan et al., 2023). Undergraduates read short text passages and then studied with flashcards, with the total time held equal, so the students who made their own cards had less time left to review them. Two days later, everyone took a test that included both definition questions and questions that asked them to apply the ideas.

The result: in five of the six experiments, students who made their own cards did better. A typical advantage was about 10 percentage points, roughly a letter grade.

How students made their cards mattered. Copying text onto cards word for word gave no reliable benefit. Paraphrasing, putting the idea in their own words, was the most effective method, and writing examples helped too. Most participants didn’t seem to realize that making their own cards had helped.

One set of experiments doesn’t settle the question, and at least one earlier study found no benefit from making your own cards. But the pattern fits a well-established idea in learning science: you remember information better when you do something with it, such as rephrasing, organizing or connecting it to an example, than when you simply look at it.

Where AI-made decks fit

By that standard, a deck generated by AI is a premade deck. It has real advantages:

• Speed. When you have a whole course to review, a draft deck in minutes beats no deck at all.
• Coverage. A tool won’t skip a section because it’s late and you’re tired.
• A starting point. Editing a card is quicker than writing one from scratch.

It also has predictable weaknesses. Generated cards are often too long, test details that won’t come up, or are phrased in textbook language you’ll recognize without being able to produce yourself. Occasionally, they’re simply wrong.

The middle way: generate, then edit

You can keep most of the speed and recover much of the benefit by treating a generated deck as a first draft:

1. Rewrite every answer in your own words. This was the most effective method in Pan’s experiments, and it’s quick once the card exists.
2. Add an example to cards about abstract ideas. Write a case from your lectures, or a situation you’ve seen yourself.
3. Split long cards. If an answer has three parts, make three cards.
4. Delete ruthlessly. If a card tests something trivial or something you already know cold, remove it.
5. Fix errors at the source. Check doubtful cards against your notes or textbook.

Skolight’s AI flashcard generator is built for exactly this workflow. It drafts a deck of up to 50 cards from the notes, slides or PDFs you upload, then lets you rewrite, delete or add cards before it schedules your reviews with spaced repetition. You get the speed of AI and the benefit of editing in one place.

Rules for a good card, whoever writes it

The same principles apply whether a card started life in your notebook or in an app. Many come from Piotr Wozniak, creator of the SuperMemo flashcard software, whose 20 rules of formulating knowledge are still widely used:

• One idea per card. Wozniak calls this the minimum information principle. Small cards are easier to remember and easier to review.
• Ask a specific question. “What does the mitochondrion do?” works. “Mitochondria?” doesn’t, because it gives you nothing to recall against.
• Avoid lists. “Name the seven causes of…” is hard to review well. Break the list into several cards, or ask about each item in context.
• Use fill-in-the-blank cards for definitions. Hiding one key term in a sentence keeps the context and tests the part that matters.
• Add a context cue when cards could be confused, such as the course or chapter they come from.

The part that matters most

Making the cards is only the start. The bigger gains come from how you use them: answering from memory before you flip the card, and reviewing on a schedule that spaces the reviews further apart as you learn. Our guide to active recall and spaced repetition explains how.