More than half of the tracks uploaded to Deezer each day, roughly 90,000 songs, are now fully AI-generated, and the company says up to 85 percent of the streams those tracks collect are fraudulent. The threat to working musicians is not mainly that listeners will pick a robot's song; it is that a hidden economy of synthetic tracks and bot streams is draining the shared royalty pool everyone gets paid from. When making a song costs nothing and a fake stream pays something, the flood only rises.

The Song Nobody Wrote: What It Means That Half of All New Music Is Now Made by a Machine

Deezer says more than 50 percent of the tracks uploaded to its platform every day are now fully AI-generated, roughly 90,000 songs that no human sat down to write, and up to 85 percent of the streams they collect are fraudulent.

By The Daily Reflection · July 27, 2026 · 7 min read

---

Somewhere in a data center, about 90,000 new songs will be born today. Nobody will have written them. Nobody will have played an instrument, hummed a melody, or argued in a studio about whether the bridge is working. A prompt will go in and a finished track will come out, complete with a title, a fake artist name, and cover art, and it will land in the same upload queue as every human musician's life work.

That is not a projection. It is Deezer's count of what is happening right now. The streaming service reported this month that fully AI-generated tracks have crossed 50 percent of its daily uploads for the first time, up from 44 percent in April and a small fraction the year before. TechCrunch, Billboard, and Music Business Worldwide all confirmed the figure. Half of everything new, made by nobody.

Read that again, because the number is doing something quietly radical: the majority of new music arriving on a major platform every day is no longer made by people.

---

The Flood Was Always Going to Happen

None of this should be a surprise, and the speed still is.

Two years ago, generating a convincing song from a text prompt was a novelty. Now it is a button. Tools that turn "melancholy lo-fi track about rain" into a polished two-minute recording have gotten good enough, and cheap enough, that the marginal cost of a new song has collapsed to almost nothing. When the cost of making a thing falls to zero, the quantity of that thing rises to fill every container you give it. Deezer's upload pipe is just the container.

The growth curve tells the story. Deezer only started tagging AI-generated music in June of last year. When it first published a number, fully synthetic tracks were a modest slice of uploads. By April they were 44 percent. This month they passed half. That is not a trend line; it is a wall going up.

Here is the part that gets lost in the "AI is coming for music" headlines. The flood is not mostly people trying to make art. It is people trying to make money, and the money is not coming from listeners who love the songs.

---

The Fraud Hiding Inside the Flood

Strip away the culture-war framing about robots replacing artists and a grubbier story emerges, and it is the one that actually matters.

Streaming services pay royalties out of a shared pool. Subscribers pay a flat monthly fee, the money goes into a pot, and the pot is divided according to what gets streamed. Every play is a tiny claim on that pool. This is the mechanism that fraud exploits, and AI is the perfect fuel for it.

The scheme is simple. Generate thousands of cheap tracks. Upload them. Then use bot networks to stream them over and over, each fake play siphoning a fraction of a cent from the same pool that pays real musicians. The songs do not need to be good. They do not need a single genuine listener. They only need to register plays.

Deezer has said that up to 85 percent of the streams its AI-generated tracks collect appear to be fraudulent. Read that number slowly: on the synthetic side of the catalog, roughly six of every seven listens are not listens at all. They are machines paying machines, with a human fraudster skimming the royalty in the middle.

This reframes the whole panic. The threat to working musicians is not primarily that listeners will choose a robot's song over theirs. It is that a hidden industry of synthetic tracks and fake streams is quietly draining the pool that everyone gets paid from, whether or not anyone ever chooses to hear a note of it. The money leaks out the back while the front-of-house argument about taste rages on.

---

Why Catching It Is a Race You Lose

To its credit, Deezer has done more than most. It tags fully AI-generated tracks, keeps them out of its algorithmic and editorial recommendations, and strips fraudulent streams out of its royalty math so that the fakes are not paid. It has even started selling its detection tool to the rest of the industry. This is the responsible playbook, and it still runs into a wall.

The wall is that detection is a moving target and generation is not. Every improvement in the tools that make synthetic music makes it harder to tell synthetic music apart from the real thing. The forger and the inspector are in an arms race, and the forger gets a new weapon every few months for free.

Worse, much of the current system leans on self-reporting. Platforms ask uploaders to disclose when a track is AI-generated, and the honest ones do. Fraud actors, by definition, do not. Reporting this month noted that this self-reporting gap is exactly where the bad actors operate: the label is voluntary, so the people running the scam simply decline to apply it. A rule that only the honest follow is not a rule; it is a suggestion with good manners.

So the platforms are left doing forensic detection at massive scale, trying to catch tens of thousands of new tracks a day, knowing the tools they are chasing improve faster than the tools doing the chasing. Deezer is winning enough of these battles to demonetize most of the fraud it finds. The open question is how much it is not finding.

---

The Question Listeners Have Not Answered

There is a comfortable version of this story where none of it touches you, the ordinary listener, because you would obviously never fall for a machine's song. The data is not so flattering.

Blind-listening tests keep showing that most people cannot reliably tell a competent AI track from a human one, especially in the genres built for background listening: lo-fi, ambient, "focus" playlists, the sonic wallpaper of modern life. And background listening is exactly where the streaming economy lives now. When the point of the music is to not be noticed, the bar the machine has to clear is not "make great art." It is "be inoffensive for three minutes." That bar was cleared a while ago.

This is the uncomfortable middle of the whole debate. The purists insist listeners will always crave the human touch. The technologists insist nobody will care once the songs are good enough. Both are partly right and both are missing the mechanism. Most listeners will not choose synthetic music, and most listeners will not avoid it either, because most listeners are not choosing at all. They are pressing play on a playlist and letting an algorithm decide. In that world, provenance is invisible, and what fills the silence is whatever the system serves.

The people who lose in that arrangement are not the listeners, who mostly get what they wanted, which was some sound. The losers are the working musicians whose slice of a fixed pool shrinks every time a synthetic track, fraudulent or not, claims a play that used to be theirs.

---

The Bottom Line

It is tempting to file "half of new music is now AI" under novelty, the kind of statistic you share for a raised eyebrow and forget by lunch. That undersells it. The number is a marker of a structural shift in how creative work gets made, distributed, and paid for, and the most important part of the shift is the least visible one: not the songs you might accidentally enjoy, but the fraud economy running underneath them, quietly rerouting money away from people and toward whoever can generate and stream the most noise.

Deezer is doing the unglamorous work of labeling, filtering, and demonetizing, and it deserves credit for transparency most platforms have avoided. But one company policing its own pipe is not a solution to an incentive problem this large. As long as making a song costs nothing and a fake stream pays something, the flood will keep rising, and the honest musicians sharing that royalty pool will keep paying for a party they were not invited to.

The machines are not coming for the music. They are already here, 90,000 a day, and the real fight is not over who makes the better song. It is over who gets paid when nobody can tell the difference.

---

The Daily Reflection cuts through the noise to find the stories that actually matter. Follow for thoughtful takes on politics, technology, and whatever's shaping our world.