A swipe is a small vote
A recommendation system does not need you to name an interest. It can infer one from what you do: whether you watch to the end, move on quickly, replay a clip, or choose to like, share, or comment. Each action is an imperfect clue. Taken together, they help the system estimate which videos are more likely to hold your attention.
That makes ordinary scrolling useful feedback. Pausing on a clip about a very specific hobby may matter even if you never follow its creator. The system can then test more videos with related themes, sounds, or formats and see how you respond.
Why the feed can narrow so fast
Recommendations are a feedback loop: the app selects videos, your behavior supplies new evidence, and later selections reflect that evidence. A strong response to one niche topic can lead to more of it; skipping those videos can steer the feed elsewhere. The loop can run with every new swipe, so the feed may seem to “get” a new interest before you have put it into words.
It is not literally reading a mind, and a single viewing does not reveal a complete picture of someone. The system is making predictions from behavior, then adjusting when those predictions seem to fit—or do not.
The tradeoff of a well-tuned feed
A feed that quickly finds relevant clips saves the work of searching, but it can also make continuing effortless: each video is chosen as a plausible next one, and the next swipe provides another chance to refine the guess. That is why a few minutes can stretch into a much longer session—not because every recommendation is irresistible, but because the stream keeps responding to small signals as you watch.
