Spotify Taste Profile: See What AI Thinks You Like
Spotify rolled Taste Profile to US Premium on Sep 23 after a NZ beta. How to read your AI summary, edit it in plain words, and what it teaches builders.
On this page
Spotify just showed everyone its homework. On September 23 Taste Profile reached US Premium users after a New Zealand beta: an AI written summary of your taste plus plain word controls to argue back. Here is how to use it, where it stops, and the pattern every builder should steal.
Takeaways
What shipped: visible AI summary of your taste across music, podcasts, and audiobooks, editable with natural language. How to reach it: profile picture, then Taste Profile, then write the change under Tell us more; Home updates within hours. Killer uses: silence the kids music on your account, trim sleep sounds, explore a new genre without starting over. Builder pattern: let users read the model's model of them and correct it, instead of hiding the profile behind a black box.
How do you actually use it?
Open Spotify on your phone, tap your profile picture, scroll to Taste Profile. You will see how Spotify currently understands you: artists and genres you love, habits like marathon training upbeat tracks or weekday commute news podcasts. Under Tell us more, write the adjustment in your own words and send. Spotify says Home recommendations reflect the change within a few hours, and support docs say a few minutes for the profile itself. The edit shapes Discover Weekly, recommendations, and Wrapped downstream, so small corrections compound.
open: profile picture -> Taste Profile -> read the AI summary
edit: "less sleep sounds" / "more energizing mornings" / "explore 90s alt rock"
lands: Home refreshes within hours, playlists and Wrapped follow
limit: no follow actions, no always or never rules, 18 plus Premium mobileNew Zealand lessons worth copying
Taste Profile debuted at SXSW in March via co-CEO Gustav Soderstrom, then beta tested in New Zealand before the US rollout. Staged geography plus a beta label gave Spotify real correction data before its largest market touched the feature: the same rollout shape works for any preference model you ship.
What should builders steal from this?
Show the profile, not just results
Users trust recommendations more when they can read the summary behind them. Expose your model's model of the user in plain words before asking for feedback.
Accept vibe edits, not only filters
More energizing is not a genre checkbox, yet Spotify handles it. Let users correct direction and energy, then translate to your own weights behind the scenes.
Keep exclusions separate from taste
Excluding a playlist or track reduces its influence without rewriting identity. Offer both: soft steering for direction, hard exclusion for contamination like shared accounts.
Label the limits up front
Spotify states it cannot follow or enforce always and never rules. Naming what the control cannot do prevents the exact support tickets a silent launch would create.
What is YouTube Music doing in response?
YouTube Music shipped Ask Music: describe what you want in ordinary language and get a personalized listening queue. Spotify counters with persistent profile editing rather than one shot queues. Both point the same way: recommendation becomes a conversation, and the service that remembers corrections wins the next session.
Does this connect to the agent week?
Why does a music feature close an AI agent series?
Because it is the permission pattern made friendly. Muse and CC gate what agents may touch; Taste Profile gates what the model may assume. All three move control toward the user: scope access, show memory, accept corrections. Agents that do work need the same visible profile before they spend or post.
Should I worry the AI misreads me?
That is the point of shipping the summary visibly. Flag when the profile misses, ask for more or less of a vibe, and check Home after a few hours. A misread you can see and fix beats a misread baked silently into every recommendation.
As of September 30, 2026: the best recommendation interface is an editable one. Read your summary, correct it in plain words, and exclude what was never you. Next, read agents that do work to see the same user control pattern applied to agents that book, buy, and file.