Ninety thousand songs a day, and almost nobody listening

On 21 July Deezer published a number that had not existed before: in June, more than half of all tracks uploaded in a day were AI-generated. On average, around ninety thousand of them every twenty-four hours. For comparison, across the whole of 2025 the platform detected and tagged 13.4 million.

The second row of numbers is the interesting one. All that music accounts for between one and three percent of streams. And of even that sliver, up to 85 percent of the streams in 2025 were fraudulent, against eight percent across the catalogue as a whole. Which means most of the audience for this music is not human.

Deezer puts its detection tool at 99.8 percent accuracy: it misses roughly two tracks in a thousand and mislabels a human recording fewer than once in ten thousand. Since June such tracks carry a visible tag, drop out of recommendations and editorial playlists, earn nothing on fraudulent plays, and face removal if untouched for six months.

That leaves the listener. In an Ipsos survey run in November 2025 across eight countries with nine thousand respondents, 97 percent could not tell machine from human. Eighty percent want clear labelling; fifty-two percent do not want these songs in the main charts.

"Now that half of all daily uploads are AI-generated tracks, we are taking additional steps to safeguard the rights of artists and songwriters," says chief executive Alexis Lanternier.

Ninety-seven percent could not tell — on headphones, between other tasks, on a phone. That is not the same as failing to tell while sitting three metres from the instrument. The question is less about how good the technology is than about how much attention is left over for music at all.