Spotify has removed more than 50 million tracks that appear to have been created by artificial intelligence from its curated music recommendations so far this year, as the streaming giant works to keep AI-generated music from crowding out human artists.

“AI-generated slop and spam don’t belong 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 tracks for what it said were serious or repeated violations of its rules, which prohibit replicating an existing creator’s content or creating media artificially with technology.

The flood of AI-generated creative content online poses a major challenge to streaming services like Spotify, which need to identify and weed out instances of impersonation or deception from among tens of millions of tracks.

Here, Spotify said it turned to its listeners for help. The company analyzed patterns within songs that users frequently either skipped or ignored on the platform over the past six months. These tracks consistently overlapped with the artificially generated audio sometimes referred to as slop.

By combining feedback from listeners with metadata and third-party AI detection tools, Spotify was able to update its internal recommendation system to limit how often AI generated content was surfaced.

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 played relative to the total number of streams on the platform.

“If the denominator has grown because of AI slop, the percentage that each legitimate rights holder or artist will receive is diminished,” said Morgan Hayduck, 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 far less common for AI tracks than it is for human made music.

But not all artists who use AI in their creative process publish what Spotify may consider “slop.” For these situations, Spotify launched an AI Credits feature earlier this year that artists can use to disclose which parts of a song’s creative process involved AI.

The combined use of audience feedback, metadata and detection tools to identify AI slop is one of several features that Spotify has introduced this year to increase transparency around AI use on the platform.

It also labels the artist profiles of musicians it believes are AI-generated with an AI Persona badge. AI Persona artists are excluded from user recommendation streams by default. Only listeners who proactively follow a particular AI artist will see that persona recommended to them.

Spotify says these efforts have resulted in a “marked decrease” in the number of concerns from listeners about AI spam showing up in their recommendations.

“AI moves fast and bad actors adapt just as fast,” Spotify said. “But as AI makes music more abundant, we believe listeners will value human artistry even more.”