YouTubers know nearly nothing about the recommendation algorithm, which is a giant ML model with thousands of features. Google has published several papers on how the algorithm works (search for “YouTube”
here), though they don’t mention the features used in the models for obvious reasons. Based on my experience it tends to weight watch history really heavily. So if people watch your videos a lot, lots of your new videos will be recommended to them, but if they don’t, fewer of your videos will be shown to them, even if they are subscribed to you. This implies that it would be better to make fewer higher quality videos than more lower quality videos. As an example, compare the average views to subscriber ratio of channels like StuffMadeHere and NileRed to LinusTechTips. The first two channels upload rarely and frequently get significantly more views than subscribers, while Linus struggles to get more than 15% of his subscribers to watch a given video.
If a YouTuber can’t show data explaining how a trick works, don’t believe them.