Spotify has reported on its clean-up of spam and low-quality generated tracks in recommendations. Since the start of the year more than 50 million tracks have been removed from recommendations, and another 75 million AI tracks deleted from the platform entirely for breaking the rules.
The company chose not to rely on a single detector. It weighs listener reactions, complaints, metadata and verdicts from several third-party AI detectors at once. The reasoning is simple: detectors make mistakes, and accidentally dropping a real musician from recommendations is a bad outcome.
What changes on the platform
- An AI Persona label for AI artists: by default their tracks stay out of recommendations and editorial playlists, but followers still see them
- Artists can state exactly where they used neural networks — vocals, instrumentals, lyrics or production. This shows up under Song Credits
- By the end of 2026 Artist Profile Protection arrives, letting musicians track releases under their name and remove wrong attributions
That last item answers a popular scheme: releasing generated music under well-known artists’ names to rack up plays faster.
Our take
This is not only about music. Any platform with user content — a marketplace, a catalog, a blog or a news feed — faces the same flood of generated junk. Spotify’s approach is worth remembering: do not ban everything AI-made, label it honestly, watch how people actually react, and never trust a single detector. We write our own news with AI help too, which is why every piece retells the source in our own words and links to the original. The story ran on Habr.