I wonder if a distributed attempt to build a complementary index for YouTube could be an answer; wherein users install a browser plugin that captures the results and metadata of a given query or metadata when a video is watched.
This presumes you're doing the trillion dollar company a favor, but an independent index isn't likely to align with YT's engagement goals.
Free labour is how I was downloading YT videos to my unsupported Windows Phone device in 2011.
You could run classifiers against the video but $$$