Skip to content

YouTube Music will tell you if a song was made by a machine: the label is here, but the money for AI songs keeps flowing

1 min read
Share
YouTube Music will tell you if a song was made by a machine: the label is here, but the money for AI songs keeps flowing

YouTube Music is introducing labels for songs created with generative artificial intelligence. The label sits on the screen where the song is playing, and by tapping it the listener can find out how much the machine was involved - in the composition, in the vocals, or both. The same labels will also come to the main YouTube app.

The system has three levels: "Fully Gen AI", "Partly Gen AI" and "No Gen AI". The label is declared by music partners, and YouTube says it also checks it with its own detection tools. Vocals ironed out with Auto-Tune and similar pitch-correction tools officially count as "No Gen AI". So Auto-Tune doesn't count as a machine - at least not under this classification.

The key detail is what isn't changing: AI involvement on its own doesn't exclude a song from monetisation. YouTube chose disclosure over restriction. The listener gets information, and the platform keeps earning from every play. On top of that, the same company is experimenting with Dream Track - a tool that lets selected creators make music with AI versions of the voices of artists who have agreed to take part. So the platform is labelling goods it is helping to produce.

The competition is stricter. France's Deezer announced in July that it receives almost 90,000 fully generated songs every day. There, detected AI music is labelled and removed from algorithmic recommendations and editorial playlists, and no royalties are paid for fraudulent streams. Deezer is also working on a filter that will let users switch it off entirely. TIDAL already allows that and leaves fully generated recordings out when distributing royalties. Spotify has labels and a music-spam filter, and Apple Music has announced a "Made With AI" label by the end of the year.

Almost 90,000 songs a day on a single platform is no longer a trend, it's a flood. The label tells listeners what they're hearing, but it doesn't stop the algorithm from serving up the next authorless "song". The question is whether anyone will even look at it while scrolling. And one closer to home: how long before the first AI hit in Balkan turbo-folk style racks up millions of views, and nobody asks who is actually singing?