You have course PDFs, lecture slides, and notes. The exam is in ten days. The default approach is to start reading from page one and hope it sticks. It works poorly, because reading produces a feeling of knowing rather than knowing itself. This article covers how to turn a PDF into flashcards that force you to retrieve, and what determines whether the result is worth using.

Summary

Output quality depends on input quality. A machine-readable PDF with clear structure produces far better cards than a blurry scan.
Don't make cards from everything. Select the chapters you don't know, not the whole document.
A good card has one question and one answer. Cards with a paragraph on the back don't work.
Generation is the start, not the end. Review the output and delete what misses.

Key takeaways

Active retrieval builds a more durable memory trace than rereading the same text.
Scanned documents without a text layer need OCR, or there's nothing for the tool to read.
FSRS scheduling means you mostly see the material you're weak on.
It's worth generating a quiz from the same PDF, since it tests understanding differently than cards do.

Why flashcards beat rereading

When you read a chapter for the third time, the text feels familiar, and your brain interprets that familiarity as "I know this." It's an illusion of fluency. In the exam you don't have the text in front of you, you have to retrieve the information from memory, which is a completely different operation.

A flashcard forces that operation. You see a question and must answer before revealing the solution. That effort is what strengthens the memory trace. This is why an hour of flashcards is usually worth more than an hour of reading, even though it feels less pleasant.

The second advantage is diagnostic. After an hour of reading you don't know what you know. After an hour of flashcards you know precisely.

Prepare the PDF so something can be extracted from it

This step determines the quality of everything downstream, and most people skip it.

Machine-readable text. Open the PDF and try selecting text with your cursor. If you can, you're fine. If you can't, you have an image scan and need OCR to convert it. Without that, there's nothing to generate from.

Structure helps. Documents with headings, bullet points, and clear definitions produce better results than unbroken prose. AI orients by the same cues a human reader does.

Select a range. Don't upload 300 pages at once. Take the chapters you don't know. Cards covering material you've already mastered waste time at every future review.

Check what the document contains. Charts, unlabeled tables, and equations transfer worse than prose. If your material is primarily visual, expect to fill gaps manually.

Step by step

Here's what the workflow looks like in practice. I'm using diStudero as the example, but the principle is general.

Create a folder for the subject. Keep material for one exam together.
Upload the document. Supported formats are PDF, DOCX, PPTX and TXT, with a 25,000-character limit per file. Instead of a file you can paste a web URL, which pulls in up to 100,000 characters, roughly 50 pages. Photos of handwritten notes go through OCR.
Generate flashcards. The free tier allows 10 per day; the paid tier removes the cap.
Review and edit the output. Don't skip this. Delete questions targeting trivia and fix ones that are ambiguously worded.
Review on schedule. Cards are scheduled by FSRS according to how well you know them, not in sequence.

How to recognize a good card

Generated content is raw material. The difference between a deck that works and one you abandon after two days comes down to these:

One question, one answer. A card with five bullet points on the back isn't a card, it's a summary. Split it.
The question must be answerable without context. "What is the third phase?" fails, because you don't know the third phase of what.
Avoid yes/no questions. A fifty percent guess rate trains nothing.
Test understanding, not just definitions. Alongside factual questions, add some "why" and "what happens if" ones.

Going through a generated deck and discarding a quarter of the questions is normal, and it's ten well-spent minutes.

Where flashcards stop working

Flashcards excel at facts, terms, definitions, and paired associations. They have a clear boundary.

They won't prepare you for an oral exam. There you need to speak coherently for several minutes and handle follow-up questions, which is a different skill from recognizing a term. An oral exam simulation, where you answer open-ended questions and get graded, fits better.

They fragment relationships. For understanding how concepts connect, a mind map works better, because cards break connected ideas into isolated items.

They don't work while commuting. Audio does. diStudero can generate an AI podcast from the same material, so you can listen on the way to campus.

A sensible approach is flashcards for facts, a quiz for self-testing, and one of the other formats for what cards can't cover.

What to do today

Take one document for the subject that worries you most, and make flashcards from a single chapter. Not the whole course pack, one chapter. Twenty minutes later you'll know two things: whether the format suits you, and far more importantly, how much of that chapter you actually don't know. That second finding usually justifies the twenty minutes on its own.