A number from Deezer this April has stuck with me. They released a report that said that about seventy-five thousand of the tracks uploaded to it every day are AI generated. That is 44% of everything new arriving on the service. More than two million synthetic tracks a month, in one place.
The total is striking enough on its own. What actually worries me is how quickly it got there. Deezer only switched its AI detection on a little over a year earlier, in January 2025, and back then it was catching around ten thousand AI tracks a day. Thirty thousand by September, fifty thousand by November, sixty by the new year, seventy-five by spring. Every time anyone has checked, the figure has gone up, and there is no obvious reason to think forty-four percent is where it settles.
Almost nobody is listening, yet
If you want some reassurance, here it is: hardly anyone actually plays this stuff. AI tracks are somewhere between 1 and 3% of streams on Deezer, and the platform reckons around 85% of those plays are fraudulent in any case, bots farming royalties rather than people listening. So the catalogue is filling up, but most of what pours in goes nowhere.
That word, fraudulent, is where the real harm sits. Streaming pays out of a shared pot: everyone's plays go in, and the pot is divided by share. So a fake stream that does get paid takes money straight out of the same pot that pays everyone real. The fraud-detection firm Beatdapp puts the cost at around two billion dollars a year, and reckons at least a tenth of all streaming activity is fake. Cheap AI tracks suit that perfectly: you can generate ten thousand of them for nothing and aim a bot farm at the lot. The World Intellectual Property Organization has started writing about exactly this loop, generate, upload, bot, withdraw.
The platforms have noticed, to be fair. Spotify now charges distributors for tracks caught with fake streams, and will not pay a track anything until it passes a thousand plays. That helps. But it is aimed at the symptom, the fraudulent stream, once the track is already up.
The quieter problem
Bots are crude, and catchable in the end. The thing that nags at me more is that the appetite for cheap, faceless music was built into streaming well before AI showed up. In her 2025 book Mood Machine, the journalist Liz Pelly documented Spotify's Perfect Fit Content programme: production companies paid to turn out anonymous, mood-matched tracks that then get seeded into big playlists, because library music nobody is attached to is cheaper for the platform than the artists people actually follow. She put that together from more than a hundred interviews with insiders and former staff.
Put that next to the upload numbers. A platform that already had a taste for cheap, faceless filler has just watched the cost of producing it fall to roughly nothing. The forty-four percent is not only scammers. A good chunk of it is the natural endpoint of a system that was already steering listeners towards music made by nobody in particular.
The bill, projected
The figure I keep in my head comes from CISAC, the body that represents more than five million songwriters and composers worldwide. Its 2024 economic study put 24% of music creators' income at risk from generative AI by 2028: a cumulative ten billion euro across the period, and four billion a year by the end of it. The same work expects AI to make up roughly a fifth of streaming platforms' revenue by then.
It is not only economists raising the alarm. In 2025 IMRO, the body that collects performance royalties for Irish songwriters and composers, posted record revenue of €57 million and, in the same breath, warned that AI is now an 'existential threat' to the country's music creators. That is a striking thing for a rights body to say in a year its own income went up. Its chief executive, Victor Finn, pinned the worry on whether the AI Act gets implemented in full, with real opt-outs, transparency and fair payment for the music these models train on, rather than quietly watered down. When the people whose whole job is collecting money for music reach for the word existential in a good year, it is worth taking the figure at the top of this piece seriously.
We have already had a dress rehearsal for what that feels like. In 2025 an AI band reached more than a million Spotify listeners before anyone could prove it was not a band at all.
Forecasts you can argue with, and you probably should. But the upload count is not a forecast, it is a measurement, and it already reads forty-four percent. So the real question is no longer whether genuine music gets harder to pick out of the pile. It is what a real artist can actually hold up to show they are not part of it. At the moment, the honest answer is close to nothing.