Music streaming platforms use two main mechanisms for music curation: One mechanism is the human-curated playlist – a sequence of songs focused on a specific theme. The second mechanism relies on a so-called recommender system, which automatically suggests what a user should listen to next based on the user’s past behavior on the platform. While platforms invest considerable resources to develop their recommender systems to maximize consumer satisfaction and loyalty, there is limited research on the actual impact of these systems on consumer satisfaction. This paper addresses this gap and assesses the impact of algorithmic and human music recommendations on listener satisfaction. Our approach builds on an inherent feature of music listening experiences – that people tend to listen to songs they like multiple times. Our dataset is user-level data of 9,778 random premium subscribers of a leading European music streaming platform who discovered 4,136 unique new songs during three months, yielding a dataset of 1,952,195 song discoveries. Our findings show that on- platform curation is less effective than off-platform information in suggesting satisfying content to users and that human-curated playlists are, on average, more effective than algorithmic recommendations. Further, we find that on-platform curation is more effective than off-platform information in suggesting relevant, unfamiliar content, especially content from foreign low-popularity artists which the users are unfamiliar with. Finally, within the on-platform curation discovery modes, human-curated playlists are always more effective than algorithmic recommendations in suggesting relevant, albeit more familiar, content. Conversely, algorithmic recommendations do better than human- curated playlists when suggesting relevant unfamiliar content.