44% of New Songs on Deezer Are AI. Nobody's Actually Listening.

44% of new songs on Deezer are AI-generated. Here's the part that should be in every headline but isn't: nobody's actually listening to them.

Deezer announced last week that AI-generated tracks now represent 44 percent of all new music uploaded to its platform. That works out to roughly 75,000 tracks per day — more than two million per month. The numbers are real, the growth is consistent, and the story writes itself: the robots are taking over music.

Except they're not. AI-generated tracks account for 1 to 3 percent of total streams on the platform. Listeners are uploading nearly half of all new songs using AI tools and then ignoring those songs almost completely. The robot band is playing to an empty room.

The machines are talking to themselves

Of the streams AI tracks do collect, Deezer flags 85 percent as fraudulent and demonetizes them. So we're not just talking about bots uploading music. We're talking about bots uploading music and then bots streaming that music to game royalty systems. The machine is mostly talking to itself.

This is the part that gets lost in the "AI is taking over music" discourse. The supply side has been transformed. The demand side has not moved.

Deezer's arms race

To their credit, Deezer has been running harder than most platforms to keep up. In June 2025, they became the first major streaming service to explicitly tag AI-generated music at the platform level — not just in some internal database, but in the actual listener-facing interface. In 2025 alone, they tagged more than 13.4 million AI tracks.

They've also removed detected AI tracks from algorithmic recommendations, excluded them from editorial playlists, and stopped storing high-resolution versions. CEO Alexis Lanternier's statement from the announcement is worth reading carefully: "AI-generated music is now far from a marginal phenomenon... we have shown that it's possible to reduce AI-related fraud."

He said "reduce." Not "stop." Not "solve." The growth curve makes that choice of words look smart.

In January 2025, the daily AI upload count was 10,000 songs. April 2025: 20,000. September 2025: 30,000. January 2026: 60,000. April 2026: 75,000. Whatever you think the detection systems are doing, the upload numbers are not responding to them.

The 97% number and what it doesn't mean

Here's a statistic from Deezer's own research that sounds alarming and isn't: 97 percent of survey respondents couldn't distinguish AI-generated music from human-made music in a blind test.

This gets used as proof that AI music has already won. If nobody can tell the difference, why would listeners choose differently?

But listeners are choosing differently. They just can't explain why.

The 44 percent upload share converting to 1 to 3 percent of streams suggests that whatever listeners are using to sort music — artist history, context, recommendations, the simple fact of having heard something before — it's not a forensic analysis of audio waveforms. People can't identify AI music when you put it in front of them with no context, but they're clearly not gravitating toward it in the environment where they actually choose what to play.

There's an argument in there that creativity isn't just about the output. It's about the context the output lives in: who made it, why, what they were trying to do. Strip that context and put the audio in front of a listener with no information, and sure, they can't tell. Put it into a streaming platform with a fake artist page and no history, and apparently they can.

What this is a preview of

The Deezer numbers are a picture of what happens when the generation cost for content hits zero in a specific domain. It's not music that's uniquely vulnerable. Text, images, and video are all heading the same direction. The question for every platform is the same one Deezer is navigating: what do you do when supply is infinite?

The AI backlash we've been seeing in other spaces tends to focus on outputs — on what AI is producing and whether it's good. The Deezer data suggests a different frame: the bottleneck isn't production anymore. It's attention.

Deezer is betting on detection and labeling. That might work. What's already clear is that 75,000 AI songs per day are hitting one platform and generating almost no legitimate listening. The robots are not, in fact, taking over music — at least not from the listener's side.

Whether that holds is a different question.