Nearly half of the tracks uploaded to Deezer in recent months are generated by AI — about 75,000 new AI tracks added each day, the company has told partners — and it says its detector flags such uploads with a false positive rate below 0.01%.

Deezer's numbers and detection claim

Deezer has told partners that about 44% of the tracks uploaded to its platform in recent months are produced by generative AI systems. That figure translates to roughly 75,000 new AI tracks being added every day. Deezer also says it can spot those uploads automatically and labels them for listeners.

The firm is licensing the detection tool to third parties and says the system has a false positive rate below 0.01%, meaning it very rarely tags human-made music as AI-made.

That combination of rapid growth and machine detection is central to Deezer's pitch: AI music is exploding in volume, but the firm believes it can identify and flag synthetic works at scale. Few other major streaming services are as explicit about labelling AI content, the company adds.

Mastheads

Why AI tracks are flooding platforms

Generative audio models have improved rapidly and can produce short songs, backing tracks or vocal imitations with little human input, making it easy to create large amounts of material quickly.

  • Models can produce polished, generic pop-style tracks that fit playlist formats used by streaming services.
  • With the right prompts, creators or automated systems can churn out many tracks that blend into catalogues and playlists.
  • Deezer's internal listening test found about 97% of respondents failed to identify the AI-produced tracks when presented with two AI tracks and one human-made track, which the company says shows AI content can be indistinguishable to average listeners.

Fake plays: bots are listening to bots

Deezer reports that most of the streams for these new uploads are generated by automated listeners—bots or scripted accounts—rather than real people. That matters because streaming numbers feed algorithms and royalty calculations.

  • Automated plays can push tracks into playlists and charts without reaching genuine listeners.
  • Inflated counts distort discovery systems and the payouts that follow.
  • Deezer says this creates a feedback loop: models generate lots of music, automated listeners boost visibility, and more uploads follow as creators chase apparent plays.

Detection, labelling and the limits of tech

Deezer's detector looks for patterns indicative of synthetic audio and has been tuned to avoid mislabeling human work. The firm is licensing the technology to other companies to help scale detection across the wider music ecosystem.

Labelled content isn't necessarily banned; labelling gives listeners context about how a track was produced and can act as a transparency measure for platforms seeking to distinguish human-made art from computer-generated output.

However, detection alone does not stop fraudulent plays: botnets and automated streaming schemes adapt quickly. Platforms need both detection at upload and ongoing monitoring of play sources to address artificial amplification.

Related Articles

Deezer estimates about 75,000 AI tracks are uploaded to its service daily and says its detector’s false positive rate is under 0.01%, while warning that most plays for those uploads appear to come from automated listeners.

This article was created with AI assistance.