Spotify has removed more than 50 million tracks that appear to have been created by artificial intelligence from its music recommendations so far this year, as the streaming giant works to prevent AI-generated music from crowding out human artists.
“Spam and AI-generated spam have no place in your recommendations, and we’ve made changes to keep them out and protect the listening experience,” Spotify said in a statement Tuesday.
Last year, the company removed 75 million titles for what it considered serious or repeated violations of its rules, which prohibit duplicating an existing creator’s content or artificially creating media with technology.
The flood of AI-generated creative content online poses a major challenge for streaming services like Spotify, which must identify and eliminate instances of impersonation or deception among tens of millions of tracks.
Here, Spotify said it turned to its listeners for help. The company analyzed song patterns that users frequently skipped or ignored on the platform over the past six months. These tracks systematically overlapped with artificially generated audio, sometimes called “slop”.
By combining listener feedback with metadata and third-party AI detection tools, Spotify was able to update its internal recommendation system to limit how often AI-generated content appeared.
The company acknowledged that AI detection technology has limitations and that identifying suspicious content could become more difficult as generative AI evolves.
For Spotify, the presence of AI-generated music can also have huge implications for artists who are paid based, in part, on how often their music is listened to relative to the total number of streams on the platform.
“If the denominator has increased due to AI negligence, the percentage that each legitimate rights holder or artist will receive will be decreased,” said Morgan Hayduk, CEO of Beatdapp, which sells AI and fraud detection software to music labels and distributors.
To be eligible for royalties on Spotify, an artist must reach a minimum of 1,000 streams, which is much less common for AI tracks than for human-created music.
But not all artists who use AI in their creative process release what Spotify may consider “slop.” For these situations, Spotify launched an AI Credits feature earlier this year that artists can use to reveal which parts of the song creation process involved AI.
The combined use of audience feedback, metadata, and detection tools to identify AI flaws is one of several features Spotify is introducing this year to increase transparency around the use of AI on the platform.
It also labels the artist profiles of musicians it claims are AI-generated with an “AI Persona” badge. AI Persona artists are excluded from user recommendation feeds by default. Only listeners who proactively follow a particular AI artist will have that character recommended to them.
Spotify said these efforts resulted in a “marked decrease” in the number of listeners’ concerns about AI spam appearing in their recommendations.
“AI is evolving rapidly and bad actors are adapting just as quickly,” Spotify said. “But as AI makes music more plentiful, we think listeners will appreciate human artistry even more.”
Gn entert
