How recommendation actually works
Modern music recommendation runs on three layers stacked on top of each other. The first layer is acoustic. A neural net listens to your song and places it in a high-dimensional space alongside every other song it has heard. The second layer is collaborative. The system watches who else listens to you and what else those listeners listen to, then expands or narrows your audience cluster. The third layer is contextual. Time of day, device, mood, listening history shape which songs surface to which listener at which moment.
How genre is decided now
Genre is no longer determined by what the artist or label calls the music. It is determined by where the recommendation system decides the song lives, based mostly on the acoustic layer. A song uploaded as “indie pop” can find itself recommended to listeners who like dream pop or alt-R&B if its acoustic fingerprint sits closer to those clusters. The label on the file is a hint to the system, not a verdict.
The metadata identity trap
Most artists set their metadata at distribution and never look at it again. That is a mistake. Your primary genre, sub-genre, and similar-artist tags influence which playlists you are eligible for, which Daily Mix slots you appear in, and how the algorithm describes you to listeners who have never heard your name. Identity, in algorithmic terms, is metadata maintained over time. Identity set once is identity by accident.
By the time you know what you are to the algorithm, you are already that thing.
Working with the algorithm, not against it
Three principles that compound:
- Be consistent. The algorithm rewards artists who release in a recognisable lane. Sudden lane changes confuse the system and reset the audience cluster.
- Be deliberate. Choose a primary genre tag and similar-artist tags with intent, and revisit them every six months as your sound evolves.
- Be patient. Algorithmic identity takes 90 to 180 days of consistent listening data to stabilise. Releases inside that window benefit from the previous one.
What artists can actually control
You cannot control the recommendation system. You can control five inputs that shape what it concludes about you: the production palette of your songs (the acoustic fingerprint), your release cadence (frequency and consistency), your metadata (the explicit instruction set), your collaborators (who you appear next to), and your audience-acquisition strategy (which listeners encounter you first). Each of those signals tells the system what kind of artist to make you.
The longer game
The artists who succeed in the algorithmic era are not the ones who try to game the system. They are the ones who understand that they are training a model that will make decisions about them for years. The music, the metadata and the audience are all training data. Treating them as such, with the seriousness that implies, is the difference between an artist the algorithm understands and an artist the algorithm only half-knows.
