// Contents
Press play on January 1 on every AI song Deezer receives in a single day, never fall asleep, and the last one ends on July 7. The service takes in about 90,000 fully AI-generated tracks a day, more than half of all the new music uploaded to it. And the main audience for them is bots.
On October 13, 2026, Deezer promised that before the year is out its app will get a toggle that removes such music from search. A good excuse to work out how much of it there really is, and who it is for. As for how anyone makes money from it, the best explanation comes from a man sentenced to 18 months in prison on October 6: he laid out the scheme in an email to himself.
What we start with
// The inputs
Every one of these figures comes from Deezer itself , with no independent audit. The service only counts tracks its detector classifies as fully AI-generated. Songs where AI merely helped a human musician are not included.
Deezer does not disclose the average length of an AI track. So we borrow a benchmark from ordinary pop: the average song on Spotify’s charts in 2024 ran about three minutes . That is the key assumption here, and we will come back to it.
One day of uploads plays for six months
90,000 tracks at three minutes each make 270,000 minutes. That is 4,500 hours, or 187 and a half days. Hence July 7: start at midnight on January 1 and the last track ends around noon.
Nobody listens without sleep, so here is a kinder version. On a regular 40-hour work week it would take 112 weeks, more than two years with no vacation. Add the music recorded by humans, and a single day of uploads plays for almost a year.
And all that time Deezer would keep receiving another 90,000 tracks every day.
42 million tracks, or a concert since 1786
Now the whole archive. Since January 2025, Deezer’s detector has flagged more than 42 million fully AI-generated tracks. At three minutes each, that is 126 million minutes, or 2.1 million hours. Divide by 24 hours and 365 days and you get about 240 years of continuous sound.
Had someone started listening in 1786, the year Mozart’s “The Marriage of Figaro” premiered in Vienna, they would be finishing right about now. And this archive piled up in less than two years. In January 2025 the service was getting around 10,000 generated tracks a day; by June 2026 the flow had grown ninefold.
Where the estimate could be wrong
The 90,000 figure is the daily average for record-breaking June. In spring the flow was smaller (75,000 a day in April), and in October Deezer still cites nearly 90,000. Other services do not report in this much detail, so the sum cannot be transferred to them directly: there could be more there, or less.
The order of magnitude survives any correction, though. Even if we are off by half, one day of AI uploads plays for months, and the whole archive plays for more than a century.
Who is actually listening
Here the scale flips. Fully AI-generated music accounts for only 1–3% of streams on Deezer. In January 2026 the service reported that in 2025 up to 85% of streams on such tracks were fraudulent. Bots generated them, and Deezer stripped those streams from royalty payouts.
Take a thousand random streams on Deezer. Between 10 and 30 of them go to AI tracks. If the upper fraud estimate holds, a real person chooses to play an AI song somewhere between one and a half and four and a half times per thousand. More than half of all new music gets a fraction of a percent of genuine attention.
It is not that people hear the fake and switch it off. In a survey Deezer ran with Ipsos in November 2025, 97% of 9,000 participants across eight countries failed to tell two AI tracks from a song recorded by humans. The reason seems to lie elsewhere. Deezer keeps these tracks out of algorithmic recommendations and editorial playlists, and without them a song is almost impossible to stumble upon.
Why upload music no one hears
The case of American musician Michael Smith answers that neatly. From 2017 to 2024 he uploaded hundreds of thousands of AI-generated songs under invented names such as Calm Baseball, Calorie Screams and Camel Edible, and pointed up to 10,000 bots at them. On October 6, 2026, a federal court in New York sentenced him to 18 months in prison, though prosecutors had asked for at least 46. “I betrayed the music community I care so much about,” Smith told the court . He has to forfeit $8.09 million.
Smith laid out the math himself, in an October 2017 email to himself that later ended up in the indictment. Fifty-two rented cloud servers, in effect 52 computers nobody ever switches off, with 20 bots each, 1,040 “listeners” in total. Each bot could play 636 songs a day, one every two and a quarter minutes, with no sleep and no weekends. Together that came to 661,440 streams a day. At his own estimate of half a cent per stream, roughly $1.2 million a year.
That is why the volume matters. A million plays on one track would trip anti-fraud systems at once. In a 2018 email Smith explained that in order to not raise any issues with the powers that be , he needed a ton of content with small numbers of streams. Spread 661,440 daily streams over 300,000 tracks and each one gets about two plays a day, around 800 a year. Such a track disappears among millions of real ones.
This is exactly the kind of scheme targeted by a rule Spotify introduced in April 2024: it stopped paying for tracks with fewer than 1,000 streams in the previous 12 months. Under that rule, each track would fall about 200 streams a year short. To get paid, Smith would have had to concentrate his bots on fewer tracks, which would make them easier to spot.
Which leaves the question of whose money this is. A streaming service does not pay out of its own pocket: it sets aside a share of revenue in a common pool and splits it by share of streams. Every fake stream shrinks the slice of every real one. Songs without a single real listener took that money out of payouts to musicians whom people actually hear. Deezer says 8% of all streams on its platform in 2025 were fraudulent. Left unchecked, almost one dollar in twelve from the pool would have gone to bots.
A button for the people still listening
Deezer’s toggle will hide fully AI-generated tracks, albums and the artists linked to them from search. In the same Ipsos survey, 45% of streaming users said they wanted such a filter. Deezer CEO Alexis Lanternier says it will let users remove fully AI-generated music from their streaming experience completely.
A machine writes the song, a machine plays it, and a human remains at only one end of this chain: the one where the payouts arrive. In July, Deezer promised to delete AI tracks caught in streaming fraud, along with those nobody has played for at least six months. Which makes for an almost perfect rhyme. Listening through one day of uploads takes more than six months, and a track nobody has played for six months will be erased.
