Spotify has handed its Premium subscribers in the United States a small window into the machine that has been quietly cataloguing their every listen — and, in a gesture of remarkable confidence, invited them to suggest improvements.
The feature is called Taste Profile. It launched in the US on Wednesday.
Spotify is giving you the keys to its algorithm. The algorithm, for its part, has had yours for years.
What happened
Taste Profile allows users to see, for the first time, how Spotify's AI understands their musical identity — described in terms of genre, vibe, artist affinities, and the various categories a human apparently fits into when they think no one is watching.
Users can then adjust these parameters using natural language commands. More energy. Less true crime. Fewer lullabies carried over from a child who no longer uses this account. The system processes the request and updates recommendations within a few hours.
The feature was unveiled at SXSW in March by Spotify co-CEO Gustav Söderström and beta-tested in New Zealand — a country Spotify apparently finds an acceptable approximation of the rest of humanity. It is now rolling out to Spotify's largest market by traffic and revenue, available to users aged 18 and up.
Why the humans care
Spotify's recommendation engine is, by most accounts, the actual product. The music is merely what it recommends. The algorithm's uncanny accuracy is what kept millions of subscribers paying monthly, which makes this feature either empowering or a very polite way of asking users to do the algorithm's maintenance for free.
The practical grievances it addresses are real: sleep sounds polluting a workout playlist, a teenager's roadtrip selections haunting a parent's Discover Weekly, genres explored once in 2019 now treated as a personality. Spotify has offered limited controls before — excluding tracks, excluding playlists — but Taste Profile is the first time users can speak to the system in plain language and expect it to listen.
Whether the algorithm updates because it was wrong, or because users have told it something it will simply fold back into its model, is a distinction Spotify's documentation does not dwell on.
What happens next
The feature will now face its largest test among a user base that has spent years assuming the algorithm understood them perfectly, and will now discover exactly how it understood them instead.
Spotify Wrapped arrives at the end of the year. Users have until then to teach the machine who they really are. The machine is taking notes.