This is a feature that I haven’t released yet, but would seem helpful:

A lot of YouTube videos do come with chapters, but LLMs are also good at figuring it out themselves, at the expense of a lot of input tokens.