Skip to content
AI Playbook

Automation: Doc → Flashcards

📖 2 min read resourcesautomations
Turn any document or article into Anki-ready Q&A flashcards with an LLM, then write them straight to a CSV or Notion.

Feed in a doc, article, or your own notes; get back clean question/answer flashcards you can import into Anki (or drop into Notion) for spaced repetition. Great for turning the deep-dives on this site into study decks.

How it works

flowchart LR
    A[New doc<br/>upload / Drive / paste] --> B[LLM generates<br/>Q&A pairs]
    B --> C[Parse JSON<br/>to rows]
    C --> D[Write CSV<br/>Anki / Notion import]

You’ll need

  • A source document (PDF, Markdown, Google Doc, or pasted text)
  • An LLM key — Sonnet 4.6 or GPT-5.4 (better questions than the cheapest tier)
  • Anki (CSV import) or Notion

Build it in n8n

  1. Trigger → manual, a Webhook, or Drive “new file”.
  2. (If a file) Extract from File → plain text. For long docs, add a Split step to chunk by heading.
  3. HTTP Request → the AI transform node, input = {{ $json.text }}, system prompt below, max_tokens: 1500.
  4. Code node → return JSON.parse($json.content[0].text).cards to get an array.
  5. Split Out → one item per card.
  6. Convert to File (CSV) node → columns front, back → download/save. (Anki: File → Import → map columns. Or use a Notion node to create one DB row per card.)

The AI prompt

System prompt:

You create study flashcards from a document. Make atomic, recall-style cards.


Return ONLY valid JSON, no prose:
{"cards": [{"front": "<question>", "back": "<concise answer>"}]}


Rules:
- One idea per card. Prefer "why/how" questions over trivia.
- The answer must be fully supported by the document — never invent facts.
- 8-15 cards per ~1000 words. Skip filler and headings.

Zapier alternative

Trigger (Drive/Form) → Formatter → Extract text → Anthropic → Send Prompt (system prompt above) → Formatter → Parse JSON → Looping by Zapier → Google Sheets → Create Row (front, back). Export the sheet as CSV for Anki, or keep it as your deck.

Cost

A 1,500-word doc ≈ 2.5K input + ~1K output tokens. On Sonnet 4.6 (3/3/15 per 1M) that’s about $0.02 per document.

Variations

  • Point it at a playbook deep-dive URL (add an HTTP fetch step) to auto-build a deck for How LLMs Work or RAG.
  • Generate cloze deletions ({{c1::...}}) instead of Q&A for Anki cloze cards.
  • Add a difficulty tag and filter to “hard” cards only for exam crunch.