Spotify laid out its anti-spam ledger in unusual detail in its October 20, 2026 progress report, "One Year In: Putting Human Artistry First on Spotify": this year alone, more than 50 million tracks judged to be low-quality AI-generated slop or spam have been taken out of recommendations, on top of more than 75 million spammy tracks removed from the platform altogether for serious or repeated rule violations. The contrast between the two numbers is the real story.
Out of recommendations is not the same as deleted
The 50 million were demoted, not erased. The songs stay on the platform: followers can still open them and listeners can still search for them, but algorithmic recommendations and editorial playlists no longer push them to new audiences. Artists labeled with the AI Persona badge follow the same logic: excluded from recommendations by default, visible only when a listener deliberately follows them. Spotify says its editors apply the same standard and never program slop or spam, and that listener complaints about such content in recommendations have dropped markedly since the changes rolled out over the past six months.
That is also why Spotify did not simply delete everything. Mass-generated slop pollutes distribution, so cutting off recommendations cuts off its main route to profit; outright impersonation and fraud belong to the separate 75-million removal tier.
How the platform spots a slop track
No single signal is reliable, and Spotify concedes AI detection has limits. In practice it stacks signals: listener skips, ignores and negative feedback (an internal review earlier this year found heavily skipped tracks overlapped strongly with flagged slop), upload patterns, metadata, and results from several third-party AI detection tools. Mass uploads, duplicates, metadata manipulation for search, spammy playlist networks and artificially short tracks are also hunted explicitly.
In other words, the test is not only "was this made by AI" but also "was it produced and dumped like spam". That line is kinder to musicians who use AI as a tool: AI use itself is not the offense; bulk flooding and fake identities are.
The identity layer on the human side
Tools launched over the past year add up to a disclosure system. AI Credits, launched in April, now receives tens of thousands of submissions a day stating which parts of a song involved AI, visible to listeners in Song Credits on mobile. More than 300,000 artists carry the Verified by Spotify badge, covering over 99% of listener searches, and more than 70% of them are independent. Thousands of AI Personas are labeled, which Spotify says makes it the only major streamer labeling AI-generated artist identities. Artist Profile Protection, which lets artists review releases credited to them, opens to all artists by year end.
The royalty threshold quietly filters too: since 2024 a track needs more than 1,000 streams a year to earn royalties, and tracks flagged as highly likely AI-generated are 10 times more likely to fall below that line, squeezing the economics of mass generation. It is a different playbook from labels going to court, as when Sony Music demanded the removal of 26,000 AI deepfake songs, which targeted impersonation; Spotify is policing distribution.
For listeners the payoff is a cleaner feed. For creators the rules are arguably clearer: disclose AI involvement, keep a real identity, and you are not treated as spam; Spotify is even building an AI remix tool with Universal Music Group and indie publisher Believe. What deserves watching next is the false-positive rate: at a scale of 50 million, even a small error share means real musicians losing reach, and whether appeals keep pace will decide if this system holds up.
