That's the part of this story that keeps getting lost in the headlines. This isn't really about robots making music. It's about what happens when a payout system built for humans gets discovered by machines that can produce infinite content for free.
In this updated guide, you'll see where the 75 million number came from, how fake tracks drain real artists' paychecks, and which new rules (from Spotify, the music industry and the EU) are trying to fix it.
Table of Contents
- What Is Spotify's AI Music Purge, Really
- The 75 Million Number: Where It Came From
- It's Not a Ban, It's a Fraud Crackdown
- How AI Slop Steals Money From Real Artists
- The Deezer Numbers That Exposed the Scale
- Why the Purge Hasn't Changed the Royalty Pool
- The Fall 2026 Spam Filter
- The Case That Started It All: "Heart on My Sleeve"
- When It Happens to an Independent Artist
- The Band That Was Never Real
- What a Compliant AI-Assisted Release Looks Like
- Where Artists Say Enforcement Falls Short
- Spotify's AI Persona Badge
- Why Spotify Built the Badge Now
- The RIAA and IFPI Two-Tag Proposal
- Apple Music, Deezer and YouTube Approaches
- The EU AI Act's Article 50
- Who Actually Has to Comply
- The Grace Period Nobody Should Assume Applies
- Why This Changes Things for Global Platforms
- Where This Leaves the Purge
What Is Spotify's AI Music Purge, Really?
In September 2025, Spotify confirmed something the industry had suspected for a while. According to Spotify's September 2025 announcement on AI protections, the platform had removed more than 75 million spam tracks over the previous twelve months. Not "AI-assisted." Not "experimental." Spam: mass-produced, low-effort, often fraudulent uploads that existed purely to intercept streaming royalties before a real listener pressed play.
By mid-2026, Spotify's global head of artists, marketing and policy, Sam Duboff, was still repeating the same figure on stage in Australia. The cleanup wasn't a one-time event. It was ongoing, and the platform was still taking in about 100,000 new uploads every single day.
Do the math on that intake rate and the purge looks almost inevitable. A year of uploads at that pace adds up to roughly 36.5 million tracks. The 75 million removed didn't just clear a backlog. It cleared close to two years' worth of everything Spotify normally takes in.
"Want to know how this flood of junk content is hitting other platforms too? Read our full breakdown of the $117 million AI slop economy."
The 75 Million Number: Where It Came From
The figure first surfaced through Spotify's own newsroom announcement. The Guardian independently reported it as a crackdown on "vexatious tracks" tied to the rise of accessible AI generation tools. It wasn't a one-off press release either. Outlets including the Los Angeles Times and Rolling Stone confirmed the same figure again in July 2026, when Spotify reiterated that the twelve-month total still stood at roughly 75 million.
A few things about that number matter more than the number itself.
What it means:
- It covers a rolling twelve-month period, not Spotify's entire catalog history.
- It was confirmed by Spotify and repeated months later, not walked back.
- It's described specifically as "spam" and "vexatious" content, which is Spotify's own language, not a media exaggeration.
What it does not mean:
- It does not mean 75 million AI-assisted songs were deleted.
- It does not mean AI music itself is banned from the platform.
That last distinction is the one almost every panicked headline gets wrong.
It's Not a Ban, It's a Fraud Crackdown
If you make music with an AI tool as part of a genuine creative process, Spotify has been consistent on one point: that's not what's being targeted. The platform's policy draws a hard line between two very different things.
What Spotify allows:
- Music where AI was used for vocals, instrumentation, or post-production, as long as the creator owns or licenses the output
- Tracks that disclose AI involvement through the DDEX industry metadata standard, which Spotify began surfacing in its song-credits panel on April 16, 2026
- Genuine creative intent, even when a large part of the production is AI-generated
What gets a track pulled:
- Unauthorized AI voice cloning of a real artist
- Mass-upload spam patterns, such as one operation dumping thousands of near-identical tracks
- Fake or manipulated metadata designed to hijack search and playlist placement
- Artificially inflated or bot-driven streams
The industry shorthand for mass-produced, zero-creative-intent content is "AI slop." It's not a technical term. It's what happens when a tool that can generate a finished song in under a minute meets a payment system that rewards volume.
"Not sure how to spot it? Here are the telltale signs you're looking at AI slop before you hit play."
How AI Slop Steals Money From Real Artists?
This explains why Spotify bothered to act at all. Streaming royalties aren't paid per song the way old-school licensing worked. Spotify, like YouTube, TikTok and X, pays out of a shared pool, split according to each track's share of total streams that month.
That structure means every fraudulent stream on a junk AI track is a real artist's royalty check, redirected.
Industry estimates now put the cost of this kind of streaming fraud at roughly $2 billion a year in stolen royalties, according to reporting that tracked the RIAA and IFPI's joint push for AI labelling standards. IFPI's content-protection leadership has gone as far as calling generative AI "the ultimate enabler" of the fraud economy around streaming payouts.
Spotify has tried structural fixes:
- Since 2024, a track must hit 1,000 streams in a rolling twelve-month period before it earns any recorded royalties, a rule built to choke off micro-payment farming.
- Spotify for Artists has confirmed this reallocates tens of millions of dollars a year that would otherwise be diluted across the long tail.
The catch: none of these fixes change the underlying incentive. More uploads still means more shares of the pool, even fraudulent ones.
"Streaming isn't the only place where fake engagement drains real creators. See how Instagram's great purge of bot accounts played out."
The Deezer Numbers That Exposed the Scale
Spotify doesn't publish daily upload breakdowns by AI origin. Its rival Deezer does, and its numbers are the clearest public window into how fast this problem accelerated.
Deezer reported that fully AI-generated tracks made up 44% of daily uploads by April 2026. Two months later, in June 2026, that number crossed 50% for the first time. More than half of everything uploaded to Deezer in a single day was fully AI-generated, at an average pace of roughly 90,000 AI tracks a day.
The climb, month by month:
- January 2025: around 10,000 AI tracks a day
- September 2025: 30,000 a day
- November 2025: 50,000 a day
- January 2026: 60,000 a day
- April 2026: 75,000 a day
- June 2026: 90,000 a day, crossing the 50% threshold
The most striking figure isn't upload volume. It's what happens once those tracks go live. Deezer found that up to 85% of streams on fully AI-generated tracks were fraudulent, generated by bots rather than real listeners. Strip that fraud out, and genuine human listening to AI-generated music sits under half a percent of total streams. This is a supply-side flood, not a demand-side trend.
Why the Purge Hasn't Changed the Royalty Pool?
Here's the uncomfortable part most coverage skips. Deleting 75 million tracks sounds decisive. It didn't change the size of the royalty pool by a single dollar.
The payout structure, a fixed pool split by share of streams, doesn't get bigger or smaller based on how much content exists. Extra tracks, fraudulent or not, dilute the humans competing for that pool. Removing junk tracks after they've already taken a share of a given month's payout doesn't return that money to the artists it was taken from.
In other words, the purge is cleanup, not prevention. That's why Spotify, Deezer and the wider industry are racing toward something that stops the fraud before the stream ever counts.
The Fall 2026 Spam Filter
Spotify has confirmed it's building a dedicated AI spam filter, expected this fall, designed to catch uploaders and tracks that use AI to mimic existing artists. It tags them and cuts off their access to algorithmic recommendation before they can pile up fraudulent streams. Since the rollout window is now open, check Spotify's newsroom for the latest status before you publish this section.
It's arriving alongside an industry-wide shift. In July 2026, the RIAA and IFPI, backed by the Recording Academy, SAG-AFTRA and the American Association of Independent Music, put forward a two-tier labeling proposal for every major streaming platform: a hard "AI-generated" tag for tracks built entirely by AI, and a softer "AI-assisted" tag for human work that leans on AI tools in places.
There's real public appetite for that clarity. A Deezer and Ipsos study of 9,000 listeners found that 97% couldn't tell an AI-generated song from a human-made one by ear, yet 80% still wanted fully AI tracks clearly labeled.
Layered on top is a hard regulatory deadline. The EU AI Act's Article 50 transparency obligations became enforceable across all 27 member states on August 2, 2026, requiring machine-readable disclosure for AI-generated audio. In Europe, labelling AI music is no longer a platform courtesy. It's law.
The Case That Started It All: "Heart on My Sleeve"
In April 2023, an anonymous producer going by Ghostwriter977 uploaded a track called "Heart on My Sleeve" to Spotify, Apple Music and YouTube. The vocals sounded exactly like Drake and The Weeknd. Neither artist had recorded a single note of it.
The track pulled more than 250,000 Spotify streams and roughly 10 million TikTok views before Universal Music Group filed a claim and the major platforms pulled it down. No human in that recording ever agreed to sing.
It became the reference case for everything that followed, mostly because it involved two of the biggest names in music. But the technology never cared about fame. Voice cloning now takes only a few seconds of source audio, so anyone with a handful of clips of your singing, your livestream commentary, or even a podcast appearance has enough material to generate a convincing fake.
"Voice cloning is only one piece of a bigger problem. Here's how AI deepfake scams are targeting everyday people."
When It Happens to an Independent Artist?
The "Heart on My Sleeve" case got headlines because of who was cloned. Murphy Campbell's case shows what happens to an independent artist instead.
Campbell is a folk musician who records traditional Appalachian ballads. In January 2026, she discovered songs on her own Spotify artist profile that she had never uploaded. Her voice had been cloned with AI, reworked into what she described as a "bro-country" sound, and distributed under her real name through a third-party distributor.
It's widely believed the source material came from her own performance videos on YouTube, was run through an AI voice-cloning tool, and was pushed back out as synthetic covers under her identity. That pipeline requires no permission, no contract and no contact with the artist at any point. Campbell eventually got most of the fakes removed, but only after they had already been live and streaming under her name.
This is the scenario Spotify's impersonation policy exists to catch. Spotify defines "impersonate" and "clone" broadly: any release using a replica of another artist's voice without permission, whether or not the uploader claims to be that artist or brands the release as an "AI version." Vocals "clearly recognizable as the exact voice of another artist" trigger removal.
The Band That Was Never Real
Not every controversial case involves a stolen identity. Some involve an identity that never existed.
The Velvet Sundown became one of 2026's stranger music stories: a "band" that built a following of roughly 1.4 million monthly listeners across three albums before it came out that the whole project, including vocals, instrumentation and promotional photos, was generated using Suno. When the story broke, Spotify didn't remove the project. It attached a disclosure label noting the music was AI-generated, which satisfied almost nobody. People who wanted it pulled were unhappy it stayed up. People who saw it as a legitimate, transparent AI art project were unhappy it got treated as suspect at all.
That reaction gap is the whole tension of this purge in miniature. Spotify isn't deciding whether AI music is good or bad. It's trying to build a system that can tell disclosed from undisclosed, and fraudulent from non-fraudulent, and not everyone agrees where those lines belong.
"Fake artists aren't new to social media either. Read the full story of an AI influencer's rise and fall."
What a Compliant AI-Assisted Release Looks Like in 2026?
If you're releasing music that used AI anywhere in the process, the rules that emerged from Spotify's September 2025 policy come down to a short checklist.
Do:
- Own or hold a clear license for anything the AI tool generated.
- Disclose AI involvement through your distributor's DDEX-compliant metadata fields. Most major distributors, including DistroKid, CD Baby, Amuse and Believe, now build this into the upload flow.
- Never use vocals that are recognizably another named artist's voice, even as a tribute or "AI version," without documented consent.
- Keep accurate contributor and credit metadata. Mismatched or missing credits are now a common cause of manual rejection.
Don't:
- Mass-upload dozens of near-identical AI tracks from a single account in a short window.
- Rely on stream-farming or promo services promising a fast jump toward the 1,000-stream royalty threshold. Spotify's fraud detection is built to catch that pattern.
Spotify added a new trust signal in April 2026 with Verified by Spotify, a badge program that reviews artist profiles for signals of a genuine human artist, such as concert history, merch and linked social accounts, rather than judging the music itself. Spotify has said more than 99% of actively searched artists are expected to qualify. A verified human artist using AI tools in production isn't flagged as "AI music," while a fully synthetic project with well-built metadata could, in theory, still pass if it mimics the signals of a real artist convincingly.
Where Artists Say Enforcement Still Falls Short?
None of this has fully quieted working musicians, and the complaints cluster around three points.
The screening net is wide.
Spotify's automated spam filter now flags roughly 39% of incoming uploads for additional scrutiny. That shows how aggressively the system is tuned, but it also means plenty of legitimate, human-made releases get pulled into manual review alongside the actual spam.
Disclosure doesn't equal detection.
The DDEX AI-credits system depends on someone in the supply chain choosing to check the box. It catches artists who disclose honestly and does very little against an operator who never intended to disclose, which is exactly the group behind most of the fraud.
Removal after the fact doesn't undo the damage.
As NPR and other outlets have reported, Spotify's own forum moderators have confirmed that AI disclosure tags do not currently affect algorithmic recommendations one way or the other. An undisclosed AI track can still get full playlist and discovery treatment for as long as it stays live and undetected.
The industry's own numbers explain why patience is thin. IFPI's 2026 Global Music Report puts global recorded music revenue at $31.7 billion for 2025, spread across 837 million paid streaming accounts, and its language on streaming fraud is blunt: "theft, plain and simple." When a working musician sitting just under the 1,000-stream threshold gets pitched a $30 promo service promising 5,000 plays overnight, that offer is exactly what the industry says it is.
Spotify's AI Persona Badge: Labelling the Artist, Not the Song
On August 11, 2026, Spotify announced something different from anything it had rolled out before. Every earlier AI policy, including the spam filter, the DDEX disclosure fields and the impersonation rules, dealt with how a track was made. The new AI Persona badge deals with whether the artist behind the profile is a real person at all.
According to NPR's report on the announcement, the AI Persona badges started rolling out from mid-September 2026. They appear on the artist's profile, in search results and on track rows in playlists. Spotify's team put the reasoning plainly: "This badge is about the artist's public identity, not about how the music was made."
The mechanics close the loophole that self-disclosure alone would leave wide open:
- Artists can voluntarily self-identify as an AI Persona through Spotify for Artists.
- Spotify also runs its own review and can apply a badge to accounts that didn't disclose but present a photorealistic AI-generated identity. Flagged artists are notified and can self-disclose or appeal.
- Listeners can see whether the artist opted in or Spotify applied the label.
- By default, AI Persona profiles are excluded from editorial and algorithmic recommendations, so no Discover Weekly or Release Radar unless a listener already follows them.
- The badge is explicitly not a judgment on the music. Spotify separates it from AI Credits and SongDNA, its tools for disclosing how a track was produced.
Worth sitting with: a fully human artist who leans heavily on AI production tools still shows up as a real person. A synthetic act with polished promotional material could, in theory, use human-performed vocals recorded to sell the illusion, and the badge targets identity, not the audio itself.
"The same identity problem is hitting social media. See how brands got burned by AI influencer scandals."
Why Spotify Built This Now: The Fake Band Problem
The badge didn't appear in a vacuum. It's a direct response to a string of AI acts that built real audiences before anyone realized they weren't people.
The Velvet Sundown crossed roughly 1.4 million monthly listeners before the Suno reveal. More recently, an AI country act called IngaRose released a single that hit number one on the US and global iTunes sales charts within weeks, after building momentum through more than 300,000 TikTok videos. It drew direct comparisons to an earlier AI country act, Breaking Rust, which had already topped Billboard's Digital Song Sales chart.
None of that was necessarily fraudulent. People were genuinely listening and buying. The problem Spotify is responding to isn't stolen royalties this time. It's that listeners had no way of knowing what they were listening to until a journalist figured it out.
The Industry's Competing Fix: RIAA and IFPI's Two-Tag Proposal
A month before Spotify's badge announcement, the record labels made their own move. On July 10, 2026, a coalition led by the RIAA and IFPI, joined by the Recording Academy, SAG-AFTRA, the American Association of Independent Music and other creator and label groups, proposed a shared, voluntary labeling system for every major platform at once.
The proposal, reported first by The Wall Street Journal, centers on two tags:
- "AI-Generated": a track where AI creates the whole performance, or performs the lead vocal or main instrumental parts, including anything built entirely from a text prompt.
- "AI-Assisted": a track that's substantially human-made, where AI contributed to specific elements without taking over lead vocals or core instrumentation.
It's designed to work like explicit-content labeling: a simple, visible badge next to a track rather than something a listener has to dig for. As SAG-AFTRA's Duncan Crabtree-Ireland put it, transparency is "essential, but it is only the beginning."
Where this differs from Spotify's approach is the target. The RIAA and IFPI tags describe how the music was made. Spotify's AI Persona badge describes who the artist claims to be. A track could carry an "AI-Assisted" tag and still belong to a fully human, verified artist, or a fully "AI-Generated" track could come from a profile that isn't flagged as an AI Persona at all.
Apple Music, Deezer and YouTube: Different Rules, Same Problem
Apple Music moved on a slower timeline, then sped up. In March 2026 it introduced optional Transparency Tags, letting labels and distributors disclose that a track, its artwork, composition or video was "materially generated" using AI. Uptake was inconsistent because nothing required anyone to use them. In August 2026, Apple told industry partners it would require the tags, soon to appear publicly as a "Made With AI" label, wherever AI created "a material portion" of a track. According to Apple Music VP Oliver Schusser, roughly a third of everything on the platform now qualifies, while accounting for less than 0.5% of total listening once bot activity is removed.
Deezer has taken the most aggressive technical approach. Its AI-detection system, running through ACRCloud since January 2025, scans every incoming upload and tags AI-generated tracks automatically rather than waiting for the uploader to self-report. That's how the industry got hard numbers at all.
YouTube sits closer to Deezer than Spotify: it requires labels on synthetic vocals and instrumentals as platform policy, rather than treating disclosure as an opt-in credits field.
Side by side:
- Deezer: automated detection at intake
- YouTube: mandatory labels on synthetic vocals and instrumentals
- Apple Music: moving from optional tags to a mandatory "Made With AI" label
- Spotify: voluntary disclosure backed by manual review, plus an identity-focused badge for AI Personas
No two platforms enforce this the same way, which is why the RIAA and IFPI proposal exists: to get everyone speaking the same labeling language before five systems fragment into five sets of rules.
"YouTube has its own crackdown underway. Read how YouTube's AI slop purge shook creators."
The EU AI Act's Article 50: The First Legal Deadline
Article 50 of the EU AI Act became directly enforceable across all 27 member states on August 2, 2026. It's the first piece of this saga that isn't a platform policy or an industry gesture. It's binding regulation with real financial teeth: non-compliance carries fines of up to €15 million or 3% of a company's worldwide annual turnover, whichever is higher.
Two of the Article's four transparency categories land directly on music platforms:
- AI-generated or AI-manipulated content must be marked in a machine-readable, detectable way, not just described in a press release.
- Deepfakes, meaning content that convincingly resembles a real person, place or event, must be disclosed as artificially generated, unless the content is evidently artistic, creative, satirical or fictional and is disclosed appropriately.
That second exemption matters for music. A track clearly framed as an AI art project doesn't trigger the same burden as one that impersonates a real artist's voice. That lines up almost exactly with the distinction Spotify's impersonation policy has enforced since September 2025, now with EU law behind it.
Provider vs. Deployer: Who Actually Has to Comply
Article 50 splits the obligation into two roles, which is why this doesn't fall only on the companies that build the AI models.
Providers build the generative AI system itself. They carry the technical burden: outputs must be marked in a machine-readable format and detectable by verification tools, using metadata tagging, watermarking or cryptographic provenance.
Deployers put that system to use, including a streaming platform that hosts and distributes the resulting content. They carry the disclosure burden: the transparency information has to actually reach the listener, not just sit in a metadata field nobody sees.
For a service like Spotify or Deezer operating in the EU, the DDEX AI-credits fields and AI Persona badges built for global product reasons now double as EU compliance infrastructure.
The Grace Period Nobody Should Assume Applies to Them
The August 2 deadline doesn't apply evenly to everything.
- Systems already on the market before August 2, 2026 get a grace period, but only for the marking-and-detection obligation under Article 50(2), and only until December 2, 2026.
- Disclosure duties for deepfakes and AI-generated public-interest text took effect immediately, with no grace period.
- Content generated before August 2, 2026 doesn't need to be labeled retroactively. The obligation is forward-looking, not a demand to relabel a back catalog.
The European Commission also finalized a Code of Practice on July 20, 2026. It proposes a standardized EU visual icon for AI-generated content, drafted as an "AI" mark and localized as "KI" in German and "IA" in French, plus a formal split between "fully AI-generated" and "AI-assisted" content that reads almost identically to the RIAA and IFPI two-tag proposal announced three weeks earlier. The Code is technically voluntary, but it's expected to become the practical benchmark regulators measure compliance against.
"Disclosure rules are spreading beyond music, too. See the 10 facts about AI-made ads running without labels before the midterms."
Why This Changes the Calculation for Global Platforms?
Spotify, Apple Music, Deezer and YouTube can't easily run different rules for different regions without real operational cost.
- A voluntary US labeling system can evolve slowly, get renegotiated, or quietly stall if platforms disagree.
- An EU legal requirement with €15 million fines attached offers no such flexibility.
- Building disclosure infrastructure to satisfy EU law tends to become the global default, because maintaining two systems costs more than one.
That's the pattern global platforms followed with GDPR, and there's little reason AI music disclosure plays out differently. The badges and tags built throughout 2025 and 2026 will likely become the de facto global standard, not because every market required them, but because one market did and rebuilding the system twice never made business sense.
Where This Leaves the Purge?
Zoom back out to where this story started: 75 million tracks deleted, with 100,000 new uploads still arriving every day behind them.
The purge was never going to be the ending. It was the moment the industry admitted the old system (upload anything, get paid per stream, sort out fraud later) couldn't survive tools that generate a finished song in under a minute. Everything since, from the impersonation cases to the DDEX fields, the AI Persona badges, the RIAA and IFPI proposal and now a binding EU law, is the slow, uneven work of telling a musician using a new tool apart from a machine gaming a payout formula.
None of it is finished. Spotify's spam filter still flags roughly 39% of new uploads for review. Disclosure still depends on someone choosing to check a box. And a track can still rack up fraudulent streams until detection catches up.
But for the first time since generative AI hit the music industry, there's a floor under all of this that isn't optional. Whatever comes next, whether a US disclosure law, a unified global icon, or better real-time fraud detection, it's being built on a foundation that, as of August 2, 2026, one part of the world has made legally impossible to ignore.