Deezer Says 44% of New Uploads Are AI-Generated Music Nobody Streams

AI-generated music has quietly taken over nearly half of everything uploaded to Deezer, and almost no one is actually listening. The Paris-based streaming service said Monday that synthetic tracks now account for 44% of all new daily uploads, a figure that works out to about 75,000 songs per day, or more than 2 million per month. Yet the same tracks pull in only 1% to 3% of total streams. The rest of the traffic around them, according to Deezer, is fake.

Deezer Becomes the First Major Platform to Show Its AI-Generated Music Math

Most streaming companies have kept quiet about how much of their catalog is machine made. Deezer just stopped pretending that was sustainable. The company’s own filings show that more than 13.4 million AI tracks were added to its library during 2025 alone, and the current pace of roughly 75,000 new uploads a day means that pile grows faster than any human catalog ever did.

This is the first time a major streaming service has put real numbers on the flood. Spotify has not. Apple Music has not. YouTube Music has not. Deezer went ahead and did it anyway, and the picture it painted is uglier than most insiders were willing to say out loud. Full disclosure on AI-generated music uploads arrived in a company blog post Monday, with CEO Alexis Lanternier framing it as a call for the rest of the industry to follow.

Starting the same day, Deezer also stopped storing high-resolution masters of AI-generated tracks. It is a small technical move, but a revealing one. The company is drawing a line between music it considers worth preserving in full fidelity and music it considers catalog filler.

AI-generated music is now far from a marginal phenomenon, and as daily deliveries keep increasing, we hope the whole music ecosystem will join us in taking action to help safeguard artist’s rights and promote transparency for fans.

Why Is Nobody Actually Listening to These Tracks?

The short answer is simple: these tracks were never really made for people in the first place. AI uploads account for 44% of new supply but only 1% to 3% of demand on the platform. That gap is not an accident of taste or marketing. It is the business model itself.

Deezer’s detection team says 85% of the streams coming from AI-generated tracks look artificial. Not the songs themselves, the plays. Bots are streaming bot music, at scale, to farm royalties from the platform’s payout pool. The company has demonetized those streams, meaning no money flows out on them, but the uploads keep coming anyway. That tells you something about how cheap it has become to manufacture a song and a listener at the same time.

Think about what that means for a working musician. Every fake stream that squeaks past detection pulls from the same finite royalty pot your actual fans are paying into. The math was already rough. Now it is getting gamed by machines on both sides of the speaker.

• 44% of new daily uploads are AI-generated
• 1% to 3% of total streams go to those uploads
• 85% of streams on AI tracks are flagged as fraudulent
• 13.4 million AI tracks added to Deezer’s catalog in 2025 alone

The Detection Tool That Claims 99.8% Accuracy

Deezer deployed its in-house AI music detection tool in January 2025, a patent-pending system the company says catches fully synthetic tracks with 99.8% accuracy. By June of the same year it became the first major service to stamp AI content with a public tag, so listeners browsing a playlist can see which songs came from a model and which did not.

That tag is the part artists have been begging for. Labels have spent the last two years warning that if platforms do not disclose AI origin, the market quietly floods and human artists absorb the damage without ever seeing a bill. Deezer is now running the experiment in public.

The accuracy number deserves scrutiny. A 99.8% hit rate sounds airtight, but at 75,000 daily uploads, even a 0.2% miss rate still means roughly 150 synthetic tracks slip through every single day, or about 55,000 a year. That is not a rounding error. That is a sub-genre.

A Blind Study Found 97% of Listeners Cannot Tell the Difference

Here is the part that should worry the industry more than the upload numbers. A blind test commissioned by Deezer surveyed 9,000 participants across eight countries and found 97% could not distinguish an AI-generated track from a human one by ear alone.

Read that again. Almost nobody can hear the difference. The idea that listeners will naturally gravitate toward human-made music because it sounds more real is, at least by this data set, a comforting fiction. The reason AI songs lose the engagement war is not taste. It is distribution, discovery, and the fact that most of them were never meant to be heard by people in the first place.

The same survey found that 80% of respondents still want AI music clearly labeled. People cannot tell, but they want to be told. That distinction matters for how regulators end up writing disclosure rules over the next year.

The $8 Million Michael Smith Case Shows Where This Leads

The same week Deezer released its data, a North Carolina man named Michael Smith pleaded guilty in the Southern District of New York to conspiracy to commit wire fraud. Federal prosecutors say he used AI tools and automated accounts to stream his own machine-made catalog and pull in more than $8 million in royalties. He agreed to forfeit the payments and faces up to five years in prison.

Smith is not a one-off. Prosecutors have already charged a separate operator in a $10 million bot streaming case involving artificial songs played by artificial listeners. The playbook is the same every time: generate cheap tracks, generate cheaper streams, collect real money. The only thing changing is the scale.

What Deezer’s numbers do, bluntly, is put a size on the attack surface. If 2 million AI tracks are hitting one mid-tier platform every month, the fraud pipeline has more raw material than any enforcement team can chase. The Smith case is a win for the DOJ. It is also a sample size of one.

What This Means for Artists, Labels, and the Rest of Streaming

For working musicians, the immediate worry is the royalty pool. Streaming services pay out of a fixed share of subscription revenue, and every fraudulent stream that survives detection is money skimmed from real artists. Deezer’s demonetization of 85% of AI streams is a strong move. It is also one platform.

For labels, the pressure now sits on the other majors. Spotify, Apple Music, Amazon Music, and YouTube Music have not released comparable data on AI upload volumes. Deezer has effectively dared them to show their numbers or admit they do not know. Both answers are bad for investor stories that lean on catalog quality.

For listeners, the change is mostly cosmetic. AI-tagged tracks will show up in search results with a label, high-resolution streams will stay human-only on Deezer, and the home feed will keep serving what the algorithm thinks you want. The machine music is already there. You just have not been listening to it.

The bigger question is whether regulators move next. If 97% of listeners cannot tell the difference, voluntary disclosure from one French platform is not going to be the end of this policy conversation.

This article is for informational purposes only and does not constitute investment advice. Every investment and trading decision involves risk. Readers should conduct their own research before making any financial decisions.