In June, more than half of everything uploaded to Deezer on a peak day was made by a machine. The streaming service reported that fully AI generated tracks crossed 50 percent of new daily uploads, running at roughly 90,000 tracks per day. That is not a warning about the future of music. That is the current inventory.
The Number That Should Stop You Cold
Ninety thousand tracks a day. For scale, that is more music arriving every twenty four hours than most independent labels release in their entire history, and it is arriving with no artist, no session, no room and no take.
Deezer’s own tracking shows the curve steepening rather than flattening. Daily AI uploads climbed from tens of thousands earlier in the year to that June peak, which means the interesting question is not whether this continues but what breaks first.
Almost Nobody Is Listening
Here is the detail that reframes the panic. Deezer says fully AI generated songs account for only about 1 to 3 percent of total streams on the platform. Half the incoming catalog is competing for a sliver of actual listening.
So this is not an audience revolt. Nobody is abandoning human musicians for synthetic ones. The flood is arriving and then sitting there, which raises the obvious question of why anyone would bother uploading it.
The Answer Is Fraud, Not Art
Deezer’s figure is that 85 percent of the streams these AI tracks do receive are fraudulent. Not listeners. Bots, farms and scripts, generating micro payouts on tracks that cost nothing to produce.
That is the business model, and it is depressingly rational. If a track costs zero to make and a fraudulent stream pays a fraction of a cent, then volume becomes the entire strategy. This is not a creative movement. It is a royalty pool being skimmed.
Who Actually Loses
Everyone in the pool. Most streaming services divide subscription revenue pro rata, meaning payouts are calculated from a share of total streams. Fake streams do not create new money, they redistribute existing money away from the artists who earned it.
A working musician with 40,000 real listeners never sees the fraud happening and simply notices that the same numbers pay less than they used to. That is the harm, and it is diffuse enough to be easy to ignore.
What Deezer Is Doing About It
Two things worth noting. The company built detection to flag fully AI generated uploads, and it excludes them from editorial playlists and algorithmic recommendations. It is also stripping fraudulent streams out of royalty calculations.
Notice what it is not doing. It is not deleting the tracks. Detect, label, decline to promote, refuse to pay. That is a more defensible position than a ban, and it draws the line at monetization rather than at existence.
The Part Where I Defend AI Slightly
A blanket “AI music is garbage” position does not survive contact with how records are actually made now. Producers use machine learning for stem separation, mastering, drum replacement, pitch correction and vocal cleanup constantly, and have for years.
The category Deezer is measuring is narrower and clearer: fully generated tracks with no human performance. That distinction is the one that matters, and it is worth insisting on it, because collapsing every use of the technology into one villain gives the actual bad actors somewhere to hide.
How to Tell What You Are Listening To
There is no reliable tell anymore, and anyone claiming a foolproof ear test is selling something. The obvious artifacts of a couple of years ago, the smeared consonants and the drum fills that never quite land, have largely been engineered out of the better tools.
What still works is context rather than audio. Check whether the act has any live footage, any credited producer, any interviews, any history. Fully synthetic catalog uploads tend to arrive as dozens of releases from an artist with no photographs, no touring and a name generated to sit near a popular search term.
Genre matters too. The heaviest concentration lands in functional background music, the lo fi study mixes, sleep loops, ambient piano and generic workout instrumentals, because that is where listeners are least likely to care who made it. If you mostly stream playlists rather than artists, you have almost certainly heard some already and had no reason to notice.
Where This Is Heading
Disclosure is coming, whether the platforms like it or not. Once a listener can see that a track is fully synthetic, the market gets to decide, which is a far healthier outcome than a quiet algorithmic filter nobody can audit.
The bigger context is that generative tools are being pushed into every consumer surface at once, the same pressure we described when three pop titans jammed the release calendar into one month and attention became the scarcest thing in the industry. Music does not have an oversupply of songs problem. It has an oversupply of things demanding to be heard.
Frequently Asked Questions
How much AI music is uploaded to streaming services?
Deezer reported that fully AI generated tracks exceeded 50 percent of its new daily uploads at peak in June 2026, averaging about 90,000 tracks per day.
Are people actually listening to AI generated songs?
Barely. Deezer puts fully AI generated tracks at roughly 1 to 3 percent of total streams, and says 85 percent of even those streams were fraudulent rather than genuine listening.
Why upload music nobody streams?
Because the streams can be manufactured. Tracks cost nothing to generate, so bot driven plays across enormous catalogs can collect royalty micro payments at scale.
Does stream fraud reduce what real artists earn?
Yes, indirectly. Most services split revenue based on each track’s share of total streams, so fraudulent plays pull money out of the same pool that pays legitimate artists.
Is Deezer removing AI generated music?
No. It detects and tags fully AI generated tracks, keeps them out of editorial playlists and recommendations, and excludes fraudulent streams from royalty payments. The tracks remain available.
Is all AI use in music a problem?
No. Machine learning is routine in mastering, stem separation and vocal processing on records people love. The specific concern is fully generated tracks with no human performance, uploaded at industrial scale to farm royalties.
