I still have doubts. Google uses anchors heavily in search, so it's hard to prove if a result is due to matches in captions or in a website linking to the video. Would be good to find proof one way or another.

FWIW I tried to find the norm MacDonald clip where he sings about the "impossible bat" and YouTube was spot on.

I suspect what you see is the search "learning" from the users -- someone is looking for a specific video that they've seen before but need to refine the search terms a few times before they find it. Thereafter, Google remembers the entire sequence of searches before the user clicks on a video and associates them with the video even though none of those terms might actually exist in their indexed video metadata. Google.com Search does this very obviously.

I mean Google search does it obviously using page rank which is that if someone links to the page with that word it uses it. Me I was searching for arcane words that I remembered hearing in the middle of a podcast. I doubt anyone actually linked or searched for the same with relevance to that podcast. Also the solution you're saying needs far more intuitive and intricate development than just indexing captions lol.