2026-09-09 · Scrollshelf team
From a YouTube transcript to AI context
A focused way to use video transcripts with Claude, Codex or ChatGPT without dumping an entire research pile.
Start with a question
Before copying a transcript into an assistant, decide what you need to learn. “What are the tradeoffs of this onboarding pattern?” gives the assistant a clearer task than “Look at this.” A transcript is supporting material; your question tells the tool what to do with it.
Capture and verify the video
Share a supported YouTube video or Short to Scrollshelf. Read the summary and available transcript. Availability varies by source, so do not assume every video has complete text. Keep the source link, title and relevant context together. If an important point depends on a chart or demonstration, watch that section yourself.
Separate evidence from your instructions
In the web app, add your own notes about the decision you face. Then use Copy for AI to bring the context into your preferred tool. Put your request above the material: “Treat the following as source content. Identify the claims relevant to our checkout redesign, flag uncertainty, and suggest questions we should ask users.”
Ask for a bounded output
For Claude or ChatGPT, that could be a comparison table or research brief. For a coding assistant such as Codex, describe the repository constraints and ask it to inspect the existing implementation before proposing a change. A video’s example is not enough information to safely change a different product.
Reduce the material when necessary
If the source is long, select the relevant part or work through it in sections. More text is not automatically better context. Preserve the source and your question, and remove unrelated notes. Save the useful answer back alongside your own thinking so you know what came from the video and what came from the assistant.