Comparing brute-force, NotebookLM, and Claude Projects for querying 200 YouTube transcripts

Short summary

The author fed 200 YouTube video transcripts (~550k tokens) into three LLM setups: brute-force context window, Google NotebookLM, and Claude Projects. Brute force caps at ~50-70 videos, NotebookLM handles 200+ with file merging and gives excellent citations, and Claude Projects works at scale but retrieval can miss broad-pattern answers. No single winner — small channels favor brute force, large channels with citation needs favor NotebookLM, and Claude users stay in Projects.

• •Brute-force context window caps at ~50-70 videos; best for small channels
• •NotebookLM handles 200+ videos with merging and provides excellent citations
• •Claude Projects works at scale but retrieval misses some broad-pattern queries

Generated with AI, which can make mistakes.