A Duke University pilot study found that nine artists identified as AI-generated through Deezer’s tagging system were recommended through Spotify’s Release Radar. Five researchers used new genre-focused accounts over 28 days and encountered 728 artists. They assessed 76 unfamiliar artists using public signals, then cross-checked 64 against Deezer. All nine confirmed cases surfaced through the personalized new-release playlist, and the researchers reported no AI disclosure from those artists. Duke’s Tech Policy Program discussed the findings in an October 12 post.
The study illustrates the limits of asking listeners to investigate an artist’s identity themselves. Its public-signal method produced confirmation through Deezer for roughly 14% of the cross-checked artists, while an absence of a Deezer tag did not establish human authorship. The pilot does not measure the proportion of AI music across Spotify’s catalog or recommendations, and its listening dates are not specified. The researchers call for clearer mandatory disclosures and listener controls. Their evidence supports a focused concern about transparency in the sampled recommendations, rather than a platform-wide estimate or an assessment of later policy changes.
Join the discussion