Spotify announced on Tuesday that over the past year it removed more than 50 million songs it considers AI-generated slop or spam from user recommendations. It also says it has retuned its recommendation system to surface more music made by human artists, and that its editors won’t program anything flagged as slop. According to the company, listener complaints about this stuff showing up in recommendations have dropped noticeably.
The key word is recommendations. Engadget points out that pulling a track from the algorithm doesn’t mean wiping it from the service. So this was a cleanup of the front window. The back room is still stocked.
Out of the feed, still in the catalog
Spotify is quietly defining the problem as “this showed up where I didn’t want it” and not “this exists on the platform.” That framing lets it shrink complaints without making any decision about what belongs in its catalog. If you never hear the slop, you stop complaining about the slop. Meanwhile it can stay up, findable through search, sitting on playlists someone else built.
I think that’s the wrong call. Fifty million is a huge number, and it’s the second huge number in two years: Spotify said last year it had taken down more than 75 million spam tracks outright. A pipeline that steady will keep producing, and treating the output as a ranking problem means fighting it forever, one tweak at a time.
Detection is the other soft spot. Spotify says it identifies slop using a mix of listener engagement and feedback, upload patterns, metadata, and third-party AI detection tools. That’s a reasonable stack for catching spam farms that upload thousands of near-identical tracks. It’s much shakier for judging any single song, and Spotify isn’t publishing how often it gets that wrong in either direction.
Everyone else picked a side
The rest of the market has been more willing to commit. YouTube Music now labels AI songs. Deezer lets users switch AI slop off. Bandcamp banned AI music entirely. All three put the information, or the decision, out in the open.
Spotify has stopped short of mandatory labels. What it offers is an “AI Persona” badge for artists using an AI identity that doesn’t represent a real person, and it won’t recommend those personas. That covers fully synthetic acts. It says nothing about a human artist who leans heavily on generation tools, and it gives listeners no switch to flip.
There’s an obvious reason for the caution. Spotify is partnering with Universal Music Group and independent publisher Believe on a tool for making AI remixes. A company building its own generative music feature has little appetite for a blanket label that would mark its own product’s output as second-class. The line it’s drawing is basically licensed AI music versus unlicensed AI music, presented as human versus slop.
If Spotify trusts its classifier enough to keep 50 million tracks out of everyone’s recommendations, it can show that verdict to users. Put a label on flagged tracks and add a filter like Deezer’s. Let people see what the system decided.
So far the only public output of Spotify’s slop detector is a single count. How many of those 50 million tracks are still sitting in the catalog, one search away? Spotify hasn’t said.