Facebook Reels has revamped its video recommendation system by prioritizing user feedback over traditional engagement metrics like likes and watch time. This shift is driven by the new User True Interest Survey (UTIS) model, which helps surface niche content and enhance user satisfaction. The platform now combines user surveys with machine learning to understand user preferences more accurately, leading to improvements in recommendation quality. The UTIS model utilizes a simple question posed to users about how well a video matches their interests, collecting thousands of responses daily. This direct feedback revealed that earlier methods only achieved 48.3% precision in identifying true interests. By addressing biases and refining the dataset, the new model increased accuracy from 59.5% to 71.5%, precision from 48.3% to 63.2%, and recall from 45.4% to 66.1%. These metrics indicate a substantial enhancement in identifying what users genuinely want to see. Implementation of the UTIS model has shown promising results. In A/B testing with over 10 million users, the model led to a 5.4% increase in high survey ratings and a 5.2% boost in overall user engagement. It has also helped deliver more high-quality, niche content while reducing generic recommendations. Despite these advancements, challenges remain, including serving users with limited interaction histories and improving the diversity of recommendations. The team is exploring advanced techniques like large language models to further refine personalization.