Deezer discloses that AI music accounts for 44% of new uploads, but actual plays are only about 3%

French music streaming platform Deezer announced today, April 21, the most comprehensive AI music infiltration data to date: AI-generated songs currently make up 44% of the platform’s daily newly uploaded content—up to 75k tracks per day, and over 2 million tracks per month. Despite an influx of AI music at scale, its real plays account for only 1–3% of the platform’s total plays, and among those, 85% of AI-generated tracks have their eligibility for earning profits canceled.

Unprecedented scale: more than 2 million AI music tracks per month

The figures disclosed by Deezer reveal how deeply generative AI tools have penetrated streaming platforms. With a monthly upload volume of over 2 million AI songs, it marks the first time a major streaming platform has fully disclosed specific data on the rampant spread of AI content—while competitors like Spotify have largely remained silent on the issue so far.

In January 2025, Deezer deployed a patent-pending AI music recognition tool with an accuracy rate of 99.8%. By June 2025, it became the world’s first major streaming service platform to clearly label AI-generated content. In just the year of 2025 alone, this technology identified more than 13.4 million AI-generated tracks in the platform’s catalog.

A severe disconnect between uploads and plays: 85% of play counts are false

There are few real listeners to AI music, and this disconnect is closely tied to fraudulent activity. Deezer’s detection technology has identified 85% of AI-generated tracks’ play counts as fraudulent streams and canceled the monetization eligibility of those play counts. Previously, reports indicated that a man pleaded guilty to admitting that he used AI music to scam more than $8 million in streaming royalties, and another AI music producer was also federally indicted for a $10 million fraud case. The main purpose of大量 uploads of AI music appears not to be attracting real listeners, but extracting royalties through fraudulent plays.

97% of listeners can’t tell the difference; 80% demand clear labeling

Even though AI music is widespread, listeners still have difficulty distinguishing it by ear. A blind listening study commissioned by Deezer investigated 9,000 participants across eight countries. The results showed that 97% of respondents could not distinguish AI-generated songs from human-made creations. However, 80% of respondents believe that music generated entirely by AI should carry clear labels—reflecting users’ strong demand for informed choice, even when they can’t identify it independently from a technical standpoint.

Frequently Asked Questions

How does Deezer identify AI-generated music?

In January 2025, Deezer deployed a set of patent-pending AI music recognition tools with an accuracy rate of 99.8%. The system can automatically scan music uploaded to the platform and identify AI-generated characteristics. In June 2025, Deezer became the world’s first major streaming service to clearly label AI-generated content on the platform interface.

Why is the AI music upload volume so high but real play counts so low?

The main purpose of mass AI music uploads does not seem to be attracting real listeners, but rather stealing streaming royalties through fraudulent streaming. Deezer’s detection data shows that 85% of play counts for AI-generated tracks are fraudulent streams, strongly pointing to organized streaming fraud activities rather than natural audience growth.

What impact do fraudulent AI music streams have on royalty earnings for human artists?

Fraudulent AI music streams dilute the overall streaming pool and indirectly reduce royalty income for real artists—because royalties are typically distributed in proportion to play counts. Deezer’s cancellation of monetization measures is one of the most aggressive responses in the industry so far, but other major platforms have not yet released detection data of a similar scale, making the real global impact difficult to quantify comprehensively.