Satisfaction survey

YouTube shows these prompts to a small fraction of viewers on a small fraction of videos, asking them to rate what they just watched from one to five stars. Four and five star ratings feed what YouTube calls valued watch time. Ratings of one or two are excluded from it. Because so few people answer, models trained on the responses predict the likely rating for everyone else.

Studio never reports your scores back. Most individual videos never gather enough answers to say anything reliable about them. The practical consequence is that padding a runtime to bank minutes can pull satisfaction down and cost you recommendations later, while the raw watch time still looks perfectly healthy.