Deezer: 44% of New Music Uploads Are AI-Generated

Deezer has published the sharpest signal yet that generative audio has swamped streaming supply without winning the ear. In a press update on April 20, the French music service said 44% of all new tracks uploaded to its platform are now fully AI-generated, translating to roughly 75,000 synthetic songs a day and more than two million a month.

TL;DR

44% of new uploads are fully AI-generated, up from 28% in late 2025.
Volume: 75,000 AI tracks per day, 2M+ per month, up 650% since January 2025.
AI tracks account for only 1-3% of total streams on Deezer.
85% of AI-track stream activity is detected as fraudulent and demonetized.
Deezer started platform-level AI tagging in June 2025 and has since tagged 13.4M tracks.

The Upload Curve

The trajectory is steeper than any prior figure Deezer has disclosed. The company's own detection tool, live since January 2025, has surfaced a near-tenfold rise in daily AI uploads in fifteen months.

Date	Daily AI uploads	Share of daily uploads
January 2025	~10,000	Not reported
September 2025	~30,000	~28%
November 2025	~50,000	~33%
January 2026	~60,000	39%
April 2026	~75,000	44%

Source: Deezer newsroom disclosures.

Deezer is the first major streaming platform to publish this kind of time-series on synthetic content. Spotify and Apple Music have not released comparable figures; both pursue quieter approaches based on filtering low-quality or spammy content rather than platform-level AI labels.

Supply Without Demand

44% of Uploads, 1-3% of Streams

The upload share is the headline, but the consumption figure is the story. AI-generated tracks account for only 1-3% of total streams on Deezer. That's a structural asymmetry: generative tools are manufacturing catalog at industrial scale while listeners, algorithms, and editorial teams continue to favor human-made music.

The gap is partly engineered. Deezer removes AI-tagged tracks from algorithmic recommendations and editorial playlists, and has stopped storing the hi-res versions of synthetic uploads. Those are all supply-side levers that push AI music out of the discovery paths that drive most streams on any major platform.

Fraud Is the Business Model

What remains of AI-track listening is mostly not listening. Deezer reports 85% of streams against AI-detected tracks are fraudulent - likely bot-driven play inflation aimed at extracting royalties from the pro-rata streaming pool. Those streams are demonetized, meaning no payout flows to the uploader.

The combination of industrial generation, bot listening, and platform demonetization describes a full fraud pipeline, not an artistic movement.

"AI-generated music is now far from a marginal phenomenon. We have shown that it's possible to reduce AI-related fraud and payment dilution in streaming to a minimum."
Alexis Lanternier, CEO, Deezer

Detection as a Product

Deezer has also turned detection into a revenue line. The company filed two patents on its AI music classifier in December 2024, began licensing the detection technology in January 2026, and reports tagging 13.4 million tracks across 2025. Selling detection to rightsholders, labels, and other platforms is a durable play if the upload curve continues.

What It Does Not Tell You

The 44% figure is a measurement, not a judgment on quality. Deezer classifies uploads as "fully AI-generated," which captures end-to-end synthetic production but does not distinguish AI-assisted human work - the broad middle where most professional musicians are quietly operating in 2026. A guitarist using AI stem separation and a ghost-written pop track produced completely by a diffusion model land on different sides of Deezer's definition even when the output differs by degree.

The consumption figure is also platform-specific. Deezer is the fourth-largest music streaming service globally and its user base skews toward curation-driven listening. On short-form video platforms, where AI tracks are frequently used as sound beds without listener attention, the consumption-to-upload ratio could be very different.

The survey data Deezer released alongside the numbers is the most uncomfortable piece. A November 2025 study found 97% of respondents could not reliably distinguish AI-created music from human-made tracks on blind listen, yet 80% wanted clear AI labeling. Listeners cannot hear the line and are asking platforms to draw it for them.

The 44% number is going to be the number that every music executive, regulator, and rightsholder quotes for the rest of 2026. The underlying story is narrower and more interesting: a platform is publishing enough structured data to show that generative audio is flooding the supply side faster than demand can absorb it, and that most of the resulting stream activity is fraud. Those two facts together are the framework the next wave of streaming royalty reform is going to argue over.

Sources: