YouTube Lecture Notes with AI: From Playlist to Study Guide

Midterm season, 14 recorded lectures unwatched, and rewatching at 2x speed still means seven hours. Making YouTube lecture notes with AI got my review down to about 20 minutes per lecture: fetch the transcript, run one prompt that produces Cornell-style notes plus Anki-ready flashcards, done. The full setup is below, and a five-lecture module costs exactly nothing.

Getting the lecture transcript

University channels — MIT OCW, Stanford Online, freeCodeCamp — generally have decent captions, which is what the youtube2text.org API reads:

curl -s "https://youtube2text.org/api/transcribe?url=https://www.youtube.com/watch?v=LECTURE_ID" \
  -H "Authorization: Bearer yt_your_key"

The transcript comes back in result.content. A 50-minute lecture is roughly 55-70k characters, which fits in one Claude or GPT context alongside the prompt with plenty of headroom — the default maxChars of 150000 covers even a three-hour recorded seminar in a single call. No captions means a TRANSCRIPT_UNAVAILABLE 404, which mostly happens on freshly uploaded recordings.

The prompt: Cornell notes plus flashcards in one pass

I run both outputs in one prompt because the model has already "read" the lecture — asking twice wastes tokens and produces inconsistent terminology between notes and cards.

You are making study notes from a lecture transcript.

Course: {COURSE NAME}
Lecture: {NUMBER AND TITLE}

Produce two sections:

1. CORNELL NOTES
   - Cue column: one key question or term per concept
   - Notes: compressed explanations. Keep every worked
     example and every number the lecturer uses.
   - Summary: five sentences maximum, plain words.

2. FLASHCARDS
   - 12-15 question-answer pairs, one per line, formatted
     question<TAB>answer for Anki import
   - Test understanding, not trivia: "Why does X hold?"
     beats "In what year..."

Add nothing that is not in the transcript. If the lecturer
hints something is exam-relevant, flag it with [EXAM].

Transcript:
{PASTE TRANSCRIPT}

Save the flashcard block as a .txt file and Anki imports it directly via File → Import. The [EXAM] flag has paid for itself more than once — lecturers telegraph their exams constantly, and the model catches phrasing like "you'll want to remember this" that I glaze over at 2x.

Quota math for a real course

The shared demo key allows 5 videos per month per IP. Your own free key (Google sign-in, no card, no phone number) is also 5 per month — that's a five-lecture module for free. A 24-lecture semester course fits the $5.99 basic tier's 50 videos with room for a second course. Since one lecture is one API call regardless of length, your quota spend equals your lecture count — batch the whole playlist in an evening and study off the notes for the rest of the term.

What the captions get wrong

The honest limitation: auto-generated captions handle spoken explanation well and mathematical notation badly. "Eigenvalue" survives; "x sub n plus one over two" comes out as word soup; Greek letters become phonetic guesses. For concept-heavy courses — biology, history, algorithms lectures where the professor explains in sentences — the notes are genuinely good. For derivation-heavy math, treat the notes as a map of what was covered and verify every formula against the slides. These notes complement the lecture; they don't replace watching the hard parts.

Where to take it from here

If you want the notes somewhere more permanent than a text file, pipe them into Notion. To batch an entire playlist with a loop and retry logic, the Python guide has working code. And if you'd rather skip scripting entirely, you can connect the API to Claude and just paste lecture URLs into chat.

Grab a free key at youtube2text.org/app/keys before your next lecture drops — future-you, the one with the exam on Thursday, says thanks.