It is possible to get that by scraping the website instead of using the API, but you would need to simulate many different users with different random starting points. There are problems with that approach too. Ultimately to really answer the YouTube would need to give anonymised data of click’s on the recommendation links.