Deezer will let listeners hide AI music. The difficult part is deciding what counts

Deezer plans to give every listener a switch that removes fully AI-generated tracks, albums and related artists from search. The global feature is scheduled before the end of 2026 and can be reversed at any time. AI tracks are already excluded from the service’s algorithmic recommendations and editorial playlists; the new control extends that policy into a user’s active search experience.[1]

A preference switch sounds simple because the interface is binary. The system behind it is not. Deezer must decide whether a recording is fully generated, partly assisted or human-made, attach that classification to albums and artists, and handle disputes when the label changes what a listener can discover. The detector becomes part of the catalogue’s editorial infrastructure.[1][2]

The toggle changes visibility, not the existence of the music

Accuracy percentages turn into catalogue decisions at scale

Deezer reports 99.8% accuracy for its detector and estimates that it may miss 0.2% of AI-generated tracks—two in every thousand. It also says the false-positive rate on human music is below one in ten thousand. Those are company-reported figures and their practical effect depends on the mix of music, the generators represented in the test set and how performance changes as new systems appear.[2]

An illustrative calculation shows the scale. If the reported 0.2% miss rate applied unchanged to 90,000 fully generated deliveries in one day, about 180 could pass undetected. That is our arithmetic on Deezer’s stated volume and error rate, not an observed daily count. Conversely, even a very small human false-positive rate can matter when applied across a vast catalogue, especially if one label hides an artist as well as a track.[1][2]

“Fully generated” is the policy boundary

The feature does not promise to remove every song that used an AI tool. Deezer says it focuses on music created entirely through generative platforms and treats limited AI-assisted elements differently. That distinction is important for modern production: pitch correction, stem separation, mastering, synthetic instruments and generative composition can occupy very different places in a human workflow.[1]

The cleanest user promise is therefore narrower than “human-only music.” It is “music the service has classified as fully AI-generated is hidden.” A musician can still use automation; a human performance can still be mislabeled; an undetected generated track can still appear. The interface should make that scope understandable without forcing listeners to read a classifier paper.

User choice and royalty protection are different objectives

Deezer’s survey found that 45% of streaming users wanted a way to filter fully generated music, 80% wanted clear labels and 52% did not want such songs in the main charts. Those preferences support a choice mechanism. The separate fraud problem supports detection and demonetization. Combining the two rhetorically can make a listener preference sound like proof that every generated song is deceptive, which the company’s own policies do not claim.[1][2]

Deezer’s move is significant because it converts an invisible classifier into a visible cultural control. Listeners gain a direct choice, artists gain a new reason to care about detector governance, and the platform assumes responsibility for defining the line it asks users to switch on and off. The button may be simple. Its credibility will depend on how honestly the service explains what lies beneath it.

Report an error or suggest a correction