Deezer’s Playlist Detector Faces a 50%-Plus AI Upload Surge
Deezer’s free AI Music Detector remains documented as a cross-platform playlist scanner, but the scale surrounding it changed within weeks of release. The official June 11 launch record defines a free service covering 20 streaming platforms and 27 languages, and assigns its 99.8% accuracy claim specifically to detecting fully AI-generated music.
The post-launch update is more consequential than a new feature: synthetic tracks became a majority of music entering Deezer during peak delivery periods. Deezer’s July 28 half-year disclosure puts intake at nearly 90,000 AI-generated tracks per day and above 50% of daily deliveries at peak times; it also identifies the Hungarian rights organization EJI and Dutch collecting society BumaStemra as licensees of the detection technology.
A scanner, not a universal streaming label
The detector examines playlists held on supported services, including Spotify, Apple Music, YouTube Music and SoundCloud. A listener selects a provider, authorizes access to the relevant account and receives track-level findings after the playlist has been analyzed.
This is an external inspection workflow. Deezer does not add labels to the original playlist, remove tracks from another company’s catalog or require that service to accept its classification. A result therefore tells the listener what Deezer’s system detected; it is not a moderation ruling issued by Spotify, Apple Music or another provider.
The distinction also explains what “across 20 platforms” means. It describes the sources from which playlists can be connected to Deezer’s detector, not a shared detection standard adopted by 20 streaming companies. The participating services continue to control their own interfaces, catalog policies and disclosure rules.
The accuracy claim covers a limited category
Deezer’s advertised performance figure concerns recordings generated entirely by AI. It should not be interpreted as equivalent performance across every song that uses an AI-assisted production tool, contains a synthetic instrument or combines generated material with human vocals and composition.
A positive result is a classifier finding, not a complete account of authorship. It does not establish who operated a generator, how much editing occurred afterward, whether training material was licensed or whether the finished recording infringes copyright. Those questions require provenance and rights information that cannot be reconstructed from an audio classification alone.
The detector’s category is nevertheless useful because fully generated recordings can be difficult to identify from playlist metadata. It gives listeners an additional transparency layer where the original service may not display comparable provenance, while leaving unresolved the more complicated spectrum of mixed human and machine production.
More uploads do not mean most listening is synthetic
The headline threshold measures incoming supply, not audience demand. A platform can receive a very large volume of cheaply generated recordings even when listeners spend relatively little time playing them, so the share of deliveries cannot be substituted for the share of streams.
TNW’s independent July 22 account places fully AI-generated music at only 1% to 3% of actual Deezer streams, while describing 85% of streams on those tracks in the previous year as fraudulent and detailing removal of AI tracks connected to manipulated listening.
That gap helps explain why detection matters to streaming economics even when synthetic music has limited organic demand. Generative systems can produce catalogs at a speed unavailable to conventional recording workflows, while automated accounts or streaming farms can manufacture plays across those catalogs. High delivery volume may therefore reveal an attempt to exploit distribution and royalty systems rather than a corresponding shift in listener preference.
Deezer applies stronger consequences on its own service
The public detector can only present information about playlists connected from outside services. Within Deezer’s own catalog, identified fully AI-generated tracks receive a label and are kept out of algorithmic recommendations and editorial playlists, although they remain available for deliberate listening.
Fraud enforcement is a separate layer. A recording’s classification as AI-generated and a finding that its streams were manipulated answer different questions: one concerns how the audio was produced, while the other concerns how plays were obtained. Treating every synthetic track as fraudulent would collapse that distinction, just as treating every play on an AI track as genuine would ignore the manipulation pattern behind Deezer’s policy.
The commercial licenses extend the same core detection capability into professional rights-management settings, but they do not make consumer results legally authoritative. A collecting society evaluating a catalog can combine detection with ownership records, registrations and other evidence unavailable to an individual scanning a saved playlist.
What changed after the release
The detector’s original purpose has remained stable: it offers a free way to check playlists without requiring the music to be hosted by Deezer. What changed is the surrounding intake benchmark, which moved from a large minority of Deezer’s deliveries at launch to a majority during peak periods soon afterward.
That development makes the service more relevant without making it more conclusive. It can expose fully synthetic tracks in a connected library and illustrate the widening gap between the amount of music being supplied and the amount audiences actually choose to hear. It cannot settle mixed authorship, licensing, copyright or intent—and its findings do not bind the platform where the playlist originated.
Also read:
Subscribe to our newsletter
Get the latest Web3, AI, and crypto news delivered straight to your inbox.