How it works, step by step
- Install the appand open it. There's no sign-up screen to get past.
- Choose to build a deck from a document and select your PDF — from phone storage, Drive, or wherever you saved it.
- Let the AI read it. It pulls out the facts worth testing and writes them as questions with short answers. This step needs a connection.
- Prune, then review.Delete anything you don't need, then start a session — the spaced repetition scheduler handles the timing from there.
Which PDFs produce good decks
The quality of a generated deck is mostly decided before you press the button. Three things matter:
It needs a text layer.If you can select and copy text in the PDF, the AI can read it. If the file is a scan — a photographed page saved as a PDF — there's nothing to read. Use photo import for those, which runs OCR first.
Size it like a study session, not a library.One chapter, one lecture, one past paper. A whole textbook forces the model to summarise at a level so high that the cards stop being testable. If you have a 600-page book, split it and import the chapter you're on.
Prose and definitions beat tables and equations.Terminology, mechanisms, dates, classifications and definitions convert cleanly. Dense mathematical derivations don't — flashcards are the wrong tool for those anyway; do problems instead.
One deck a day, free
The free tier includes 1AI generation per day. In practice that means choosing the chapter that matters most today, which is a better study habit than bulk-importing a semester you'll never review. If you're building decks for a whole syllabus in one sitting, Premium removes the daily cap. Reviewing existing decks is never limited.
What to do with the deck once it exists
A generated deck is a draft. The students who get value from this workflow all do the same three things afterwards:
- Cut ruthlessly.If a card asks something you'd never be examined on, delete it. A 40-card deck you finish beats a 200-card deck you abandon.
- Split compound cards. Any card whose answer is a list of five things is really five cards. Splitting them is what makes the recall reliable.
- Review the day you make it, then let the scheduler take over. The first review is the one that decides whether anything sticks.
Why this beats making cards by hand
Nobody disputes that hand-written cards are slightly better cards. The problem is the hour it takes to write forty of them, which is an hour not spent reviewing. The evidence for spaced repetition is about review, not about authorship — so the trade that actually pays is: generate the draft in a minute, spend five minutes pruning it, and put the other fifty-four minutes into review sessions.
This is the same argument we make on the Anki comparison page, where it's the central difference between the two apps.