Deezer says AI music now makes up 44% of new uploads and most suspect streams are fraud

AI music is flooding the upload pipeline

Deezer says 44% of all new music uploaded to its platform is now AI-generated, a figure that shows just how quickly synthetic audio is filling mainstream distribution channels. The company says that translates to about 75,000 new AI tracks every day. Yet the more revealing part of the update is not the upload volume itself. Deezer says most of the streams tied to this music appear to be fraudulent and that roughly 85% of AI music streams are being demonetized.

The result is a snapshot of an AI content economy where production is cheap, distribution is easy, and the main business model may be less about reaching listeners than gaming payout systems.

Volume is high, discovery is low

Despite the surge in uploads, Deezer says AI-generated tracks account for only 1% to 3% of total listening on the service. That discrepancy matters. It suggests the platform is being inundated with AI content faster than audiences are actually choosing it. In other words, the supply side has exploded, but genuine demand remains limited.

Deezer attributes that outcome partly to moderation decisions. The company says AI-flagged tracks are excluded from editorial playlists and recommendation surfaces, reducing the chance that users encounter them organically. That policy is effectively a containment strategy: allow uploads, but restrict algorithmic amplification.

Without those measures, the service appears to believe AI music would be more likely to seep into ordinary listening sessions, especially because many users struggle to distinguish it from human-made tracks.

Most listeners cannot tell the difference

According to Deezer, a user survey asked listeners to hear three songs, two of them AI-generated, and 97% could not correctly tell the AI tracks from the human-made one. Even without the full methodology, that result reinforces a central problem for streaming platforms. If synthetic music can reliably mimic generic commercial styles, labeling and detection become more important than listener intuition.

The issue is not simply aesthetic confusion. It is economic. Streaming systems are built to distribute small payments across vast libraries of tracks. If synthetic music can be mass-produced and paired with artificial streaming activity, it becomes possible to dilute payouts or redirect revenue without building any authentic audience.

Detection becomes infrastructure

Deezer says it has developed technology to identify AI-generated uploads and that it licenses that system to third parties, claiming a false-positive rate below 0.01%. The company is also one of the few streaming services explicitly labeling AI content. Those details matter because they suggest synthetic-audio detection is becoming platform infrastructure rather than an optional moderation feature.

The rationale is straightforward. Once AI-generated music approaches half of all new uploads, manual review is no longer plausible. Detection has to operate at scale and has to distinguish between content that is merely AI-made and content connected to fraudulent monetization schemes.

Deezer’s public stance is that the main purpose of many AI uploads is fraud. That is a strong claim, but it aligns with the gap between massive upload numbers and small organic listening share. If the music were primarily being made for fans, one would expect more discovery-driven listening growth to accompany the output surge.

A warning for the rest of streaming

Deezer’s figures may not automatically describe every platform, but they expose a pressure point the entire music industry is likely to face. Streaming services built for frictionless distribution are especially vulnerable to synthetic content floods because the cost of making and uploading tracks is collapsing while the cost of monitoring authenticity remains high.

That creates several overlapping risks. Recommendation systems can become cluttered. Human artists can face payout dilution. Fraudsters can simulate listening activity. And audiences may find it harder to know what they are hearing, especially in genres or moods where generic production values are already common.

The company’s response shows one possible model: detection, labeling, recommendation restrictions, and aggressive demonetization of suspicious streams. Whether that framework becomes industry standard will depend on how quickly rivals confront the same scale problem.

The platform era enters its synthetic phase

The deeper significance of Deezer’s update is that AI music is no longer a novelty issue. It is becoming a structural feature of digital media supply. Once nearly half of new uploads can be synthetic, platforms have to decide what counts as acceptable participation, what deserves recommendation, and what looks like manipulation.

For now, Deezer’s data suggests the real battle is not over whether AI music exists, but over whether automated production and automated listening will distort the economics of streaming faster than platforms can adapt. That is no longer a theoretical question. The numbers say it is already happening.

This article is based on reporting by Ars Technica. Read the original article.