Spotify says it pulled 50 million AI slop tracks from recommendations this year

What Spotify announced

Spotify says it has taken more than 50 million AI-generated "slop and spam" tracks out of its recommendations this year. The company gave the figure in a 20 October 2026 post marking one year since it laid out plans to protect artists and listeners from spam, impersonation and deception. It says the 50 million are in addition to more than 75 million spammy tracks it removed from the platform altogether for serious or repeated rule violations.

The tracks taken out of recommendations are not necessarily deleted. They stop appearing in algorithmic features such as personalised playlists and suggestions, and Spotify says its human editors also refuse to program content identified as slop or spam. Since the changes, Spotify says it has seen "a marked decrease in listener concerns" about such content showing up in recommendations, though it gives no number for that decline.

The post frames the work around one idea: "human artistry comes first." Alongside the slop figure, Spotify gave progress numbers for its AI credits, artist verification, AI Persona labels and Artist Profile Protection, and said its 1,000-stream royalty threshold has become a defence against mass-produced AI music. The same day it announced a licensing deal with the distributor Believe for a paid AI remix tool.

How Spotify decides what counts as slop

Spotify says the push started with listener behaviour. Earlier this year it looked at the tracks people were skipping, ignoring or flagging as unwanted, and "a clear pattern emerged": they were consistently ones identified as slop or spam. Over the past six months it says it made numerous updates to its recommendation system to remove that content and surface more music by human artists.

No single test does the job, the company says. It combines listener engagement and feedback, upload patterns, metadata and information from several third-party AI detection tools. Spotify is unusually candid that the last of those is weak on its own: "AI detection technology in particular has limitations, and it will only get harder as generative AI evolves." Using the signals together, it says, lets it identify likely slop or spam with more confidence.

It also looks for tactics rather than just sounds: mass uploads, duplicates, search manipulation, spammy playlist networks and artificially short tracks designed to rack up paid plays. Those are the same behaviours Spotify named in its September 2025 announcement of a music spam filter, which it said would tag uploaders and tracks using these tactics and stop recommending them, rolled out conservatively to avoid penalising legitimate artists.

The labels and controls behind it

Spotify listed four tools it says now back the recommendation changes:

• AI Credits, launched in April, which let artists disclose which roles and instruments were made with AI. Spotify says artists are now submitting tens of thousands of credits every day, visible in Song Credits on the Now Playing view on mobile.
• Verified by Spotify, a badge meaning a real, identifiable person is behind the music. Spotify says more than 300,000 artists are verified, representing more than 99 percent of what listeners search for, and more than 70 percent of them are independent.
• AI Persona badges, which label artists whose identity is AI-generated. Thousands are now labelled, and AI Personas are excluded from editorial and algorithmic recommendations by default unless a listener follows one.
• Artist Profile Protection, which lets artists review releases that list them before they reach their profile. Thousands opted into the beta, and Spotify says it will open the tool to all artists by the end of the year.

Why the AI Persona label is different

The AI Persona badge is about who an artist claims to be, not how a song was made. When Spotify introduced it in August, it said profiles could self-disclose, and that it would also review profiles whose name and imagery appear to show a photorealistic AI-generated identity, starting with artists above set audience thresholds. Labelled artists are notified and can appeal. Spotify now claims it is the only major streaming service that labels when an artist's identity is believed to be AI-generated.

Artist Profile Protection targets a related problem: music landing on the wrong artist page. In its March introduction of the feature, Spotify said misattributed releases had become worse with easy-to-produce AI tracks, and that an artist who declines a release keeps it off their profile, stats and recommendations, though it may still appear under their name on other services.

The royalty threshold as a filter

Spotify also pointed to a rule it introduced in 2024 for a different reason. Tracks must reach 1,000 streams in a year before they earn royalties. The company now says that threshold "has become a first line of defense for human artists against the wave of AI-generated content."

Its evidence is a single ratio: tracks with indicators showing a high probability of being AI-generated are 10 times more likely to fall below the 1,000-stream line. Spotify argues that makes it harder for bad actors to siphon royalties from real artists by generating music at scale, since thousands of low-play tracks earn nothing. It did not say how many tracks fall below the threshold or how much money that keeps in the pool for other artists.

The royalty point matters because Spotify pays artists from a shared pool. In 2025 it said total music payouts had grown from $1 billion in 2014 to $10 billion in 2024, and that big payouts attract spam operations. Every stream an AI farm captures is a fraction of a cent that would otherwise go elsewhere.

A licensed AI remix tool with Believe

On the same day, Spotify and the French music company Believe announced a licensing agreement for an upcoming Spotify tool that lets fans make AI-powered remixes and covers of existing songs. The tool is meant for reworking songs, "not entirely new works," and streams of the remixes will pay royalties to the participating artists and rights holders behind the originals. It is planned as a paid add-on for Spotify Premium subscribers, and every remix will link back to the original track.

Participation is voluntary. Believe and its distribution arm TuneCore will use a two-step consent process: artists first choose whether to be considered for AI-related programmes at all, and those who opt in are then offered the Spotify deal separately. Believe chief executive Denis Ladegaillerie said the partnership turns "an already widespread fan remixing activity into a licensed experience." Spotify did not give a launch date or price for the tool.

The pairing shows how Spotify wants to split the market: keep cheap, mass-produced AI music out of recommendations while selling AI features built on licensed songs from consenting artists. Music AI tools such as ElevenLabs' Instant Soundtracks are making generated audio easier to produce, which is the backdrop for both moves.

How other streaming services compare

Other services have taken different routes. Deezer said on 13 October that it will add a toggle letting listeners hide fully AI-generated music from search before the end of 2026, and that it has excluded more than 42 million detected AI tracks from recommendations since January 2025. YouTube Music began labelling songs made primarily with generative AI on 14 October, combining partner disclosures with its own detection.

Spotify has stopped short of labelling every AI-generated track. Its labels cover AI artist identities, and AI credits depend on artists choosing to disclose. The Verge notes that Bandcamp has gone furthest by banning AI music entirely. Spotify's approach is to keep suspected slop out of the places where most listening happens, its recommendations, rather than to remove or label all of it.

What the numbers do and do not show

The 50 million figure comes from Spotify and has not been independently checked. The company did not say how many of those tracks were confirmed as AI-generated rather than spam made by other means, what share of its catalogue they represent, or how many tracks were wrongly flagged. It also did not say what happened to the 75 million removed tracks over time; in September 2025 it gave that same 75 million figure for the previous 12 months.

Spotify itself warns that detection will get harder as generative models improve, and that bad actors adapt quickly. Its system leans on behaviour and listener signals for that reason, which means some slop may only be caught after people have skipped it, and genuine artists with unusual upload patterns could be caught in the net. The company says it will keep updating its systems.

For listeners, the practical change is in what appears in Discover Weekly, radio and editorial playlists, not in search. Music from AI Personas and suspected slop can still be found and played; it is just not pushed. For artists, the more useful tools are Artist Profile Protection, due to all artists by year end, and the verification badge.