Most of the studying you do starts as a PDF. A textbook chapter, a slide deck, a clinical guideline, a research paper the professor assigned, a handout from class. The part that rarely gets done is the conversion — from "I read it once" to "I actually remember it three months from now." Flashcards close that gap, and AI now does the extraction work for you.
This is a hands-on 5-step guide to going from a PDF to a reviewable flashcard deck using AI, plus the five mistakes that produce junk cards.
What you need before you start
Two things: a PDF and an AI flashcard tool.
The PDF can be almost anything — a textbook chapter, exported lecture slides, a journal article, a government handout, a drug monograph, a statute excerpt. Text-based PDFs work best. Scanned PDFs (images of pages rather than real text) need OCR to work well; most modern AI tools handle this, but the output is only as clean as the scan.
The workflow below uses MintDeck's free AI flashcard generator because it accepts PDF uploads directly in the browser, requires no account to start, and produces cards that are immediately reviewable with spaced repetition. The same general steps work with any AI flashcard tool — the principles port over.
Step 1: Trim the PDF to what you actually need to study
This is the highest-leverage step and the one most people skip.
A 400-page textbook PDF dumped into an AI generator produces 400 pages of mediocre cards. A 15-page chapter produces a focused, reviewable deck. Before you upload anything, ask: which pages am I studying this week?
Three quick ways to trim a PDF:
- On macOS: Open in Preview → select the page thumbnails you want → File → Export as PDF.
- On iOS: Open in the Files app, long-press the PDF, tap Quick Actions → Create PDF, then delete the pages you don't need.
- In any browser: Use a free tool like PDF24 or Smallpdf's "split PDF" feature.
For slide decks specifically: trim anything that's a title slide, table of contents, agenda, or thank-you slide before uploading. Those pages produce low-signal cards that clutter your review queue.
Rule of thumb: 5–20 pages of focused content per generation. That's a reviewable deck. Beyond that, quality degrades.
Step 2: Upload the PDF and generate your deck
Open the AI flashcard generator and drag your trimmed PDF into the input area. On iOS, use the MintDeck app — it accepts PDF uploads directly from the Files app, iCloud Drive, and Google Drive.
The AI extracts the text, identifies the core concepts and relationships, and drafts a deck in about 30 seconds for a typical 10-page input. You'll get a mix of definition cards, "what is" prompts, and cause/effect or mechanism cards depending on the source material.
For text-based PDFs this is nearly lossless. For scanned PDFs, the tool runs OCR first — if your scan is blurry or skewed, the output will reflect that. A quick test: if you can select text with your cursor in a PDF reader, it's text-based and will generate cleanly. If you can't, it's a scan and OCR quality will vary.
Step 3: Review and edit the cards before you start studying
Every AI-generated deck needs a five-minute pass before it becomes your study deck. This isn't optional — it's what separates a useful deck from a deck that teaches you mistakes.
Three things to check on every card:
- Is the fact correct? LLMs occasionally paraphrase a source in a way that subtly changes the meaning. Cross-check any card that looks surprising against the original PDF.
- Is the card atomic? One idea per card. If a card front asks "What is X and Y?", split it.
- Is the card phrased like a question you'd actually ask yourself? Not "Mitochondria is..." but "What is the primary function of mitochondria?"
Delete duplicates and anything trivial. A tight 40-card deck beats a bloated 120-card deck every time. MintDeck's inline editor lets you edit, split, or delete cards directly from the deck screen — no separate editor to open. There's a longer deep-dive on editing in the edit-cards guide if you want it.
Step 4: Study with spaced repetition (FSRS)
This is the step that actually moves the material from "I read this" to "I remember this."
The scheduling algorithm matters. MintDeck uses FSRS by default — the Free Spaced Repetition Scheduler — which is trained on millions of real reviews and predicts your personal forgetting curve with measurably more accuracy than the SM-2 algorithm most flashcard apps still ship with. The practical effect: cards you find easy get pushed further out, cards you struggle with come back sooner, and your daily review queue stays short without losing retention. Why FSRS outperforms SM-2 has the full breakdown if you want the theory.
Target 10–20 minutes of review per day, every day. Consistency beats volume — fifteen minutes every morning produces more durable memory than a 90-minute cram on Sunday.
If the PDF included a diagram, a formula, or a key table, hold on to a photo or screenshot of it separately. Flashcards are for the recall layer; they're not a replacement for the underlying source.
Step 5: Expand the deck as you study more PDFs
A single-chapter deck is a starting point, not a finished product. Most effective study workflows look like this:
- Week 1: Generate chapter 1 cards. Review daily.
- Week 2: Add chapter 2 cards to the same deck. Keep reviewing.
- Week 4: You have a 150-card deck covering everything you've read so far, and FSRS is quietly scheduling older cards at longer intervals while keeping newer cards in frequent rotation.
That cumulative effect is what makes flashcards worth the setup. Every PDF you process this semester compounds into a single growing deck you can trust for an exam two months away.
For the general theory behind this workflow, how to make flashcards with AI and the science of spaced repetition are worth a read.
Common mistakes that produce junk cards
The five mistakes that turn a good PDF into a bad deck:
- Uploading the entire PDF instead of trimming first. A 200-page textbook dump produces cards that are too shallow to be useful and too numerous to review. Trim to 5–20 focused pages.
- Skipping the review-and-edit step. Every deck needs five minutes of editing before first review. AI is ~90% accurate; the remaining 10% is what you'd be memorizing incorrectly.
- Uploading scanned PDFs without checking OCR quality. If you can't select text in the PDF, the AI is working from OCR output. Bad scans produce bad cards. Rescan or retype if the extraction is noisy.
- Asking for too many cards per chapter. Card count ≠ understanding. A 40-card chapter deck outperforms a 120-card one. Let the AI produce a focused deck; delete anything trivial in review.
- Studying in long blocks instead of daily. Fifteen minutes a day beats a two-hour session once a week. The FSRS algorithm is built for daily cadence — using it otherwise wastes most of its benefit.
The bottom line
AI doesn't replace the learning — it removes the "I know I should make cards for this PDF but don't have time" problem that stops most people from using flashcards at all. The workflow is: trim the PDF, upload it, edit the draft, review daily with FSRS, expand as you go.
If you've got a textbook chapter or a slide deck open right now, the free AI flashcard generator takes about 90 seconds end-to-end on a 10-page PDF. The harder part is showing up for review tomorrow.
