Deezer's AI Music Toggle Is Narrower Than It Sounds

Deezer said this week that it will add a global toggle, arriving by the end of 2026, that lets listeners strip fully AI-generated tracks, albums and related artists out of search results. That is a narrower feature than the quotes around it suggest. CEO Alexis Lanternier framed the move as removing the content from the listening experience, but the mechanics Deezer has described so far apply specifically to search, not to the full catalog a user might encounter through playlists or radio.

The scope is also narrower than the headline implies in a second way. The switch only targets tracks built entirely by tools like Suno, Udio or Rifforobotics-style generators. Songs where AI assisted mixing, mastering or a vocal layer are tracked through a separate disclosure system and will not be hidden by the toggle. Deezer's own survey with Ipsos, conducted in November 2025, found that 45% of streaming users wanted exactly this kind of filter, which is the figure the company is citing as justification for building it.

The toggle is the latest step in an 18-month project rather than a standalone feature. Deezer quietly built an internal detection system in January 2025, then disclosed in January 2026 that it had begun demonetizing AI tracks after finding that up to 85% of streams on fully AI-generated songs were fraudulent, compared with an 8% fraud rate across its whole catalog. At the same time, those tracks made up only 1 to 3% of total listening even as they accounted for as much as 44% of daily uploads, around 75,000 tracks a day, according to figures Deezer has published since. By this spring, Deezer's head of research told TechRadar that daily AI uploads had climbed past 90,000, meaning the flood kept growing even after the demonetization push.

Deezer has also been licensing the underlying detection tool rather than keeping it in house. French collecting society Sacem and Hungarian rights body EJI both now run Deezer's system, and Billboard uses it to flag AI content in its charts. In June 2026, Deezer shipped a free consumer scanner, covering 27 languages, that checks playlists across roughly 20 other platforms including Spotify, Apple Music and YouTube Music for AI-generated content, even though Deezer has no control over those catalogs.

There is a real gap between how Deezer markets the tool and how it says the tool works. Deezer's marketing describes 99%-plus accuracy, but its own head of research has said, in comments to TechRadar, that the system detects markers of AI generation and the strength of those markers rather than issuing a clean yes-or-no verdict. That is a meaningful caveat for a filter being sold as a way to cleanly remove fully AI-generated music, since a probabilistic signal does not map neatly onto a binary toggle. None of Deezer's fraud or detection figures, including the 42 million tracks and 14 million albums it says it has flagged since January 2025, have been independently audited.

Spotify, by comparison, has shipped something narrower and more cautious. Rather than a search-wide switch, it adopted the DDEX industry standard for AI disclosure credits in September 2025 and introduced AI Persona badges in August 2026 for AI-generated artist identities, reviewed case by case and excluded from algorithmic recommendation by default unless a user already follows that artist. Spotify has not built a consumer-facing off switch despite years of forum requests, a gap partly explained by the fact that it is simultaneously building its own AI remix features and has signed an AI licensing deal with Merlin.

Sweden's royalty society Stim has argued that detection tools like Deezer's only treat the symptom. In its view, the fix that matters is mandatory licensing paired with transparency about what data trained the models in the first place, which a search toggle does nothing to address no matter how it is tuned by the end of the year.