Recommendation runs on a small set of behaviors you can observe in your own analytics. Clicks tell the system whether a thumbnail and title earn attention. Watch time tells it whether the video kept the promise those elements made.
Session behavior adds context. What a viewer watches before and after your video shapes how the system places you in their feed. A video that ends a session gets judged differently from one that starts a longer run.
A high click rate on a short watch is a warning, not a win. The system reads the mismatch and pulls back reach to protect the viewer experience. It would rather show something less flashy that people actually finish.
A low click rate on a long watch suggests a strong video with a weak wrapper. Reach may grow slowly until the title and thumbnail catch up with the quality inside.
Signals are not votes. They are evidence the system uses to test a hypothesis about what a specific viewer will accept next.
YouTube runs surveys that ask viewers to rate content. Those answers feed a satisfaction signal that no creator can see directly, and it can override a healthy-looking watch curve.
Likes, shares, saves, and not-interested taps also register. They are sparse, so the system treats them as hints rather than verdicts. Still, a cluster of not-interested taps on your video is a strong negative.
The same video can hold one viewer for eight minutes and lose another in twelve seconds. The system ranks per viewer, not per video, and that changes how you should read your average.
Your channel average hides the viewers who love you. The people who watch everything you publish are the ones the system learns from fastest, and they pull new similar viewers toward your work.
The recommendation algorithm does not judge your video in one pass. It runs a series of small tests and watches what real viewers do with each one.
The first hour after upload is a test window, not a verdict. The algorithm samples a small audience and decides from their behavior whether to widen the reach.
A video can get a second push from the algorithm hours or days after it seems dead. The system keeps re-testing content when a new audience matches it.
Analyze a channel, extract its structural DNA, and generate scripts engineered from patterns that actually work.
Analyze a Channel