Spotify Pulls AI Songs From Its Recommendation Algorithm: The Fight Against Spam Music

Spotify has begun filtering low-quality AI-generated music and spam tracks out of its recommendation system. But this is not a ban. The company is not deleting the tracks from its catalog; it is stopping the algorithm from recommending them to listeners. The difference may look small, but it is exactly where the real debate over AI music is centered.

What happened?

Spotify announced that over the past year it has removed more than 50 million tracks it considers AI-generated "slop" or spam from user recommendations. The company also said it has changed its recommendation system to give more prominence to music made by human artists, and stressed that its editors apply the same standard.

Why doesn't Spotify ban AI music?

The problem isn't new; only its scale has changed. Spam music existed before AI. White noise recordings, 31-second tracks and uploads imitating the titles of popular songs have circulated on platforms for years. What they all had in common was that they exploited a loophole in the pay-per-stream model. Music generation tools like Suno and Udio, however, have pushed the cost of production close to zero. Now a single person can upload hundreds of tracks in a day.

This is where you need to understand royalty economics. Spotify pays out largely through a "pro-rata" model. All subscription and advertising revenue is collected in a single pool and distributed to each artist in proportion to their share of total streams. In other words, every stream a spam track earns shrinks a real artist's share of the pool. That makes this not an aesthetic question but a direct question of revenue distribution. The annual minimum stream threshold Spotify introduced in 2024 was a measure aimed at the same problem: tracks with very few streams no longer earn royalties.

So who defines slop? Spotify's signals most likely rest on listener behavior. These could include rapid skips, low completion rates and mass-upload patterns. That raises an uncomfortable question. Treating a track listeners skip as "low quality" is not an artistic judgment; it is engagement optimization. A human-made experimental album could get caught by the same filter. Conversely, a well-crafted AI track can slip through easily. The fully AI-generated band The Velvet Sundown, which passed one million monthly listeners in the summer of 2025, was an example of this.

How do rivals label AI music?

Platforms are approaching the same problem in different ways:

Deezer's model is more transparent: listeners can see what they are listening to. Spotify's is less visible. Users don't know what is being filtered; they only sense that their recommendations have changed. YouTube, for its part, leaves responsibility largely to the creator. That model is weak against bad-faith uploaders who don't disclose.

What changes for human artists?

For independent musicians, the short-term impact will probably be positive. Discovery playlists and autoplay playlists are the most important growth channel for small artists. When those spaces fill up with spam, they are the first to suffer. Cleaning up recommendations could boost their algorithmic visibility.

There is a risk, however. If the rules aren't transparent, an artist who is filtered by mistake may have no clear mechanism to appeal. Musicians who use AI as an assistive tool also remain in a gray area. A producer who uses AI for mastering shouldn't be lumped in with an account that generates an entire song from a single prompt. How that distinction will be made, we don't yet know.

What's next?

There are three things to watch. First, whether AI disclosures coming through distributors become mandatory. Second, how much Spotify will reveal its filtering criteria to artists. Third, the licensing deals major record labels are striking with AI companies. If those deals create a new category of "legitimate" AI music, the boundaries of the filter will have to be redrawn.

Our assessment: Spotify is targeting the right problem, because spam directly erodes artist income. But as long as it leaves the definition of quality to engagement data and keeps its criteria secret, this will remain an optimization setting rather than a curation policy. A clear, Deezer-style label would offer a fairer footing for both listeners and artists.