What the packager adds to the transcript
Instead of pasting raw captions, you get the transcript wrapped with a prompt and context: the instruction you chose, the video title and channel, and clear TRANSCRIPT START / TRANSCRIPT END markers. That structure helps the model understand the task and keep the transcript separate from your instructions.
- Summary: the main points of the video.
- Takeaways: the key lessons or insights.
- Action Items: steps and recommendations mentioned.
- Study Quiz: questions to test understanding.
Long videos are split for you
Transcripts over about 3,000 words are divided into parts of roughly 3,000 words, each labeled “Part 1 of N” and so on, and each copyable on its own. The tool also shows an estimated token count (about 1.33 tokens per word) so you can judge whether a model will accept it in one message.
- Send the parts in order, and tell the model to wait until it has all parts before answering.
- Then ask your question once, referring to the full video.
- Ask it to cite timestamps or quote lines so you can check its answer against the source.
Keep the answer honest
Language models can fill gaps with plausible guesses. Tell the model to answer only from the supplied transcript and to say when something isn’t in it. Check any specific claim, figure, or quote against the video before relying on it.