AI Music Is Now Half of Deezer’s New Uploads. That Does Not Mean Half of Listening.

TL;DR

• Deezer says fully AI-generated tracks passed 50% of its new daily uploads at peak in June, averaging about 90,000 tracks a day.¹
• The crucial counter-number: those tracks account for only 1–3% of streams on Deezer, according to the same company.¹
• Deezer will remove AI tracks linked to streaming fraud and tracks unplayed for at least six months; detected AI tracks were already excluded from its recommendations and editorial playlists.²
• The threshold is real and consequential, but it is a Deezer measurement, produced by its proprietary detector—not an independently audited estimate for all music streaming.

Deezer’s number is startling because it describes a reversal in the shape of music supply. At peak in June, more than half of the new music delivered to the French streaming service was, by its detection system, fully AI-generated. The platform puts the average at roughly 90,000 tracks a day—up from 75,000 daily tracks and 44% of deliveries it reported in April.³

But the number most likely to be misunderstood is the one that makes the story interesting. This is not a report that listeners have switched to synthetic music. Deezer says fully AI-generated tracks remain only 1–3% of listening. The flood is entering the catalogue; it is not winning the audience.

The supply curve has gone vertical

A streaming catalogue used to be constrained by recording time, distribution arrangements and a modest amount of administrative friction. Generative music removes much of that friction. An operator can now create large volumes of plausible, inexpensive audio and deliver it at industrial scale.

That changes the bottleneck. The scarce resource is no longer the ability to submit a track. It is trustworthy attention: recommendation slots, search results, royalty-pool allocation and listener confidence.

Deezer’s response is therefore more important than the headline figure. It says it will systematically take down generative-AI tracks used for streaming fraud, plus those with no streams for six months. It has already excluded detected fully AI-generated tracks from algorithmic recommendations and editorial playlists.¹

The editorial call: this is an anti-abuse and catalogue-governance event before it is a creative-replacement event. Music services are being forced to treat content ingestion as an adversarial system.

What happened—and what did not

On 21 July, Deezer announced that detected fully AI-generated uploads exceeded half of its daily new-music deliveries at peak during June. It also said that up to 85% of streams on fully AI-generated tracks in 2025 were fraudulent, and that it removes detected manipulated streams from royalty calculations.¹

Independent music-industry coverage by Billboard, Music Business Worldwide and Music Ally corroborates the announcement, its 90,000-track figure and the removal policy.²³⁴

What those reports do not provide is an independent audit of Deezer’s detector. The service describes it as patent-pending, claims 99.8% accuracy, and says it can detect signatures associated with models including Suno and Udio. Those are company claims. They may be directionally correct; they should not be treated as settled industry-wide measurement.¹

The hype correction: “half of uploads” is not “half of music”

The phrase is doing a lot of work.

“Over half of uploads” measures incoming volume. It says little by itself about listening, cultural relevance, artist earnings or the share of music that consumers actively choose. A service can receive an enormous amount of low-demand material while its audience barely encounters it—especially when the service keeps that material out of recommendations, as Deezer says it does.

That distinction also explains why the policy response is not a blanket ban. The immediate issue is a combination of spam, fraudulent stream generation and database clutter. A legitimate creator using generative tools is a different problem from a network submitting disposable tracks to harvest royalties. The technical task is to distinguish them reliably enough that enforcement does not become arbitrary.

Who is exposed to the new bottleneck

The quieter technical story: detection becomes infrastructure

The obvious debate is whether AI music is “real music.” The operational question is narrower and more urgent: can platforms classify origin, verify accounts, detect coordinated playback, and settle royalties without rewarding manipulation?

That makes audio provenance, model-output signatures, account reputation and stream-anomaly detection parts of the same stack. A classifier alone cannot establish fraud. A suspicious track still needs to be connected to behavioural evidence—accounts, payment routes, playback patterns or coordinated uploads—before removal or demonetisation is defensible.

Deezer’s policy hints at this separation. It has not announced that every detected AI track will be removed. It is targeting fraud-linked material and stale, unplayed material. That is a governance design choice: punish harmful behaviour and catalogue pollution, rather than attempting to settle authorship through a single binary label.

What this means for you

For listeners: there is little useful action beyond preferring services that disclose how they label or recommend synthetic music. The material is a major upload phenomenon, not yet a major listening phenomenon.

For independent musicians and managers: keep dated project files, stems, contributor records and distribution metadata. They are not merely good housekeeping; they are evidence if attribution, impersonation or royalty disputes become more common.

For distributors and music-tech builders: build checks before delivery, not just takedown queues after it. At minimum, combine verified uploader identity, rate limits for bulk submissions, metadata validation, duplicate-audio detection and post-release stream-anomaly monitoring. Do not assume an “AI-generated” label alone identifies misconduct.

For policymakers: resist using this one platform’s metric as a universal market-share statistic. Ask for transparent methodologies, false-positive/false-negative testing and consistent definitions of AI-generated, AI-assisted and fraudulent.

Uncertainty ledger

• Detector validity: Deezer’s 50% figure depends on a proprietary system. There is no public independent audit in the reporting reviewed here.
• Scope: the figure is for Deezer, at peak in June—not a claimed average across all services or the global music market.
• Fraud estimate: Deezer’s “up to 85%” figure is also company-reported. It is evidence of a serious problem, not an independently established industry rate.
• Policy effects: removing unplayed or fraud-linked tracks may reduce clutter, but its effect on attempted fraud will depend on whether bad actors can move to other services or adapt their submission patterns.

Bottom Line

AI music has crossed a meaningful operational threshold: it can now overwhelm a major platform’s intake without commanding much listener attention. Deezer’s response shows where the real fight sits—fraud prevention, recommendation integrity and royalty accounting, not a referendum on whether machines can write songs. The platforms that win will be the ones that make abundance cheap without making trust scarce.

Sources

• Tier 1 / primary: Deezer Newsroom, “Deezer: AI music has surpassed 50% of new music uploads for the first time” (21 July 2026).¹
• Tier 2: Billboard, “Deezer Says Daily Delivery of Fully AI Music Reaches More Than 50% for the First Time” (21 July 2026).²
• Tier 2: Music Business Worldwide, “90,000 AI tracks flood Deezer daily – passing half of new music uploads for the first time” (21 July 2026).³
• Tier 2: Music Ally, “AI-generated music is now more than half of Deezer’s uploads” (21 July 2026).⁴
• Tier 2: heise online, “Erstmals ist mehr als die Hälfte neu hochgeladener Songs auf Deezer KI” (21 July 2026).⁵