Anki cannot read a PDF. Everything between the document and a reviewable card is a translation step, and there are four ways to do it. Only one of them keeps a file out of your downloads folder.
Open the file and try to select a sentence with your cursor. If the text highlights, the PDF has a text layer and everything below applies. If the whole page selects as one block, or nothing selects at all, it is a scan — images of text. No conversion tool reads those without OCR, and any that claims to is either running OCR silently or guessing from layout.
For scans: run OCR first (your PDF reader or any OCR tool can add a text layer), then convert. Doing it the other way round produces empty or garbled cards, and you will not notice until you are reviewing them at midnight.
| Route | Steps | Failure mode |
|---|---|---|
| Type them yourself | Read the page, decide, type front and back in the Anki editor. | Slow — minutes per page. But the writing is itself a rehearsal, which is why some people deliberately keep this for high-value material. |
| Converter → APKG | Upload the PDF, get an APKG, double-click to import. | Structure-dependent: works well on bullet-heavy slides, poorly on dense prose. Re-running the same chapter imports duplicates. |
| AI → CSV → import | Paste text into a generator, get CSV, File → Import, map columns to fields. | Field mapping is where people give up. Multi-line answers and commas break naive CSVs, and encoding issues mangle accents. |
| AI → write directly | Open the PDF in the browser, generate, review in a panel, write to the deck via AnkiConnect. | Needs desktop Anki running with the free AnkiConnect add-on. No file is produced, so nothing to share with a classmate. |
A 300-page textbook converted in one pass produces a deck nobody can review. Not because the cards are wrong, but because a deck of 2,000 undifferentiated cards has no shape: you cannot tell what belongs to this week's lecture, you cannot suspend a section you have not covered, and the review queue becomes a wall.
Split by chapter or lecture. Generate one deck per unit, or one deck with tags per unit. Two thousand cards in ten tagged decks is a study plan; two thousand cards in one deck is a reason to quit.
The same rule applies to card count. Twenty to thirty cards per chapter is usually the honest number. If a tool offers to produce sixty from a chapter, it is padding — and padding is what makes people abandon decks.
lecture-07 means you can find and fix a bad batch later instead of hunting through the deck.AnkiMate is a browser extension that handles the fourth route. Open the PDF in your browser, select the passage or the page, pick one of nine subject-tuned templates (medicine, language, law, programming, GRE vocabulary, reversed, list breakdown, general, and Cloze on Pro), review the generated cards in the side panel, and write them into the deck you choose through AnkiConnect.
Three generations a day are free and writing into Anki is unlimited free. Pro is $19.9 once — Cloze, batch generation for long articles, and future templates.
Requirements, stated plainly: desktop Anki with the free AnkiConnect add-on (code 2055492159), and a text-layer PDF. Scanned PDFs need OCR first. On mobile only, use a converter and an APKG.
No. Anki imports text, CSV and APKG files. A PDF has to be turned into one of those first, either by a converter or by a tool that writes notes into your collection over AnkiConnect.
They need OCR to add a text layer before any tool can read them. Without that step you get empty cards or layout noise.
Usually twenty to thirty for a lecture, fewer for a dense textbook section. If a generator returns twice that, most of them are padding and should be deleted.
Cards are written on desktop through AnkiConnect, then sync to mobile through AnkiWeb as usual. The writing step itself requires the desktop app.
The full guide to getting your reading into Anki → · Converter versus direct write, side by side →