In one year, Spotify removed more than 75 million tracks it considered spam. The platform's entire music catalog is somewhere around 100 million songs. Read those two numbers next to each other and the math gets uncomfortable: in twelve months, Spotify culled an amount of audio approaching the size of its entire surviving catalog and called it garbage. Yet the same company keeps insisting AI-generated music has "minimal" engagement and does not meaningfully redirect royalties away from human artists.
That contradiction — a platform-scale spam emergency that supposedly nobody is listening to — is the actual story. AI music has not flooded your headphones. It has flooded the back end. And the response Spotify is building, piece by piece, looks less like a ban on robot bands and more like a new operating system for artist trust: Verified by Spotify badges, Artist Profile Protection, SongDNA, About the Song, expanded credits, a music spam filter, and an AI-impersonation policy that finally puts voice clones in the takedown bucket.
For independent artists, none of this is academic. The risk is not that one synthetic band steals one fan from your release this Friday. The risk is slower and more corrosive: listeners stop trusting what is real, distributors get stricter, low-volume tracks stop earning anything, your name shows up on a song you did not record, and the ambient pollution makes it harder to prove a real fan is real. The good news is that the response is boring and powerful. The bad news is that nobody is going to do it for you.
Spotify is not banning AI music — it is targeting spam, impersonation, and deception
The single most useful thing to understand here is the taxonomy. "AI music on Spotify" is not one thing, and Spotify is not treating it as one thing.
There is AI-assisted music, where a human writes, produces, and signs off, and a model is somewhere in the chain — vocal tuning, mastering, stem separation, a synth patch trained on a producer's own samples. Spotify has no policy against this, and it would be unenforceable if it did.
There are fully synthetic artists with no human performer, like The Velvet Sundown, the AI band The Guardian reported had crossed a million streams by mid-2025, prompting Ivors Academy and BPI figures to demand clearer labeling and consent rules. These are allowed but increasingly fenced off — the new Verified by Spotify program explicitly excludes "profiles that primarily represent AI-generated artists or AI personas" at launch.
Then there are unauthorized voice clones, which Spotify will remove on valid claim, and identity-hijacking uploads — fake AI tracks delivered under a real artist's name, the kind Rolling Stone documented showing up in smaller artists' catalogs. A2IM's Lisa Hresko told Rolling Stone the problem is "incredibly prevalent."
Finally, the volume problem: industrial spam. WIPO Magazine describes the modern playbook plainly — fraudsters use generative models to upload millions of fake songs, then stream each one a few thousand times to slip under old red-flag thresholds. IFPI's Melissa Morgia, in the same piece, calls AI an "ultimate enabler" of streaming fraud. Deezer says it now receives over 30,000 fully AI-generated tracks per day, more than 28% of its total daily delivery. Spotify has not published the equivalent number, but the scale of its 75 million removals suggests it is not smaller.
So when someone asks, "Is AI music allowed on Spotify?" the honest answer is: it depends on which kind, and the rules are tightening from the spam end inward.
The 75-million-track number is huge, but it needs context
Spotify's own newsroom post confirms the figure: more than 75 million spammy tracks removed in 12 months, alongside three new pillars — an impersonation enforcement policy, a music spam filter, and AI-disclosure support via industry-standard credits. The Guardian's coverage put it next to the platform's nearly 700 million users and roughly $10 billion in 2024 royalty payouts — a useful reminder that "spam" here means industrial-scale extraction attempts, not bedroom demos.
Two things to hold in your head at once. First, the cleanup is real and probably overdue. Second, "spammy" is a precise word and not a synonym for "AI-generated." Some of those 75 million were AI-generated, but Spotify is careful not to claim all of them were, and the takedown bar is behavior — duplication, near-duplicates, mass uploads, manipulated streams, content that exists to game the royalty pool — not the production tool.
A platform with 700 million users and a 100-million-track catalog just deleted enough music to nearly fill another Spotify-shaped one. That is not a content moderation problem. That is an economic infrastructure problem.
The number gets weirder when you set it next to Spotify's claim that AI-generated music drives "minimal" engagement. Both can be true. A flood of low-listened tracks can still poison the system: each one chips at the royalty pool, smears the recommendation graph, and dilutes the signal a real artist needs to reach a real listener. BBC's explainer on AI on Spotify cites a community-built blocker tracking 4,700 suspected AI artists. That is the kind of project that does not exist when a problem is theoretical.
The money problem is not one fake band; it is the royalty pool
The real fight is over how a finite pool of subscription revenue gets sliced. Spotify's 2024 royalty modernization introduced three changes that matter here:
A track must clear 1,000 streams in the previous 12 months to generate recorded royalties. Below that, the per-stream amount Spotify says was "demonetized at the track level" gets redistributed to artists who do clear the bar. Billboard's breakdown notes that only about 0.5% of the total royalty pool actually shifts under the change — small in aggregate, large for the individual hobbyist, and large in incentive design.
There is a minimum-listener requirement Spotify has not publicly disclosed in detail, confirmed in support docs. It is meant to stop a single account or a small bot cluster from dragging a track past the 1,000-stream line.
There are fraud penalties and "noise" track rules. Functional white-noise tracks chopped into 31-second loops, ambient-sleep mass uploads, and fake-band catalogs designed to scrape sub-threshold streams — these now get penalized or removed.
So when artists ask, "How does Spotify detect fake streams?" the answer is a moving combination of velocity anomalies, listener concentration, geographic patterns, IP signals, and partner data from distributors. Spotify does not publish the exact heuristics, partly because publishing them would brief the next generation of fraudsters. WIRED's reporting on the Michael Smith case describes a federal indictment for nearly $10 million in fraudulent streaming royalties powered partly by AI-generated music — "a billion streams and no fans" is the line worth keeping. IFPI's 2026 Global Music Report puts global recorded music revenue at $31.7 billion in 2025 with 837 million paid streaming accounts and calls streaming fraud "theft, plain and simple."
The practical takeaway for the working musician: if your release is hovering near the 1,000-stream threshold and you are tempted by a $30 promo service promising 5,000 plays, you are not buying growth. You are buying a flag. If you want real streams, the only durable path is the unglamorous one — playlist pitching, real ads, real fans. We have a guide on how to get more Spotify streams that walks through the legitimate version.
The scariest version for indie artists is identity hijacking
Most artists I have spoken to do not lose sleep about The Velvet Sundown. They lose sleep about waking up to find a song they did not record sitting on their artist page.
The Guardian profiled jazz pianist Jason Moran discovering an AI-flavored release uploaded under his name. Rolling Stone's piece documents the same pattern at smaller scale. Music Business Worldwide catalogs AI-born accounts like Aventhis and The Devil Inside racking up massive monthly listener counts and sliding into algorithmic recommendations. The Top Music Attorney podcast walked through hijacked artist pages in detail, and a follow-up episode dissected an AI band scam at length. The mechanism is the same in each case: a distributor accepts metadata that points an upload at the wrong Spotify URI, and suddenly a stranger's track is on your profile.
Why does this hit indie artists hardest? Common names, low support leverage, slower takedown processes, fewer legal resources, and — most painfully — fan confusion. The wrong upload also corrupts everything algorithmic: your stats, your Release Radar, your "Fans Also Like," and your Discover Weekly seed all start treating you as a different artist than you are.
Spotify's response is Artist Profile Protection, currently in beta, which lets eligible artists review and approve releases before they appear on a profile. It is a good policy and not a complete solution. Coverage is uneven, eligibility is gated, and it does not retroactively scrub past mistakes. The pragmatic move is to claim and actively monitor your Spotify for Artists profile, set alerts on every release date, and do a metadata audit at least monthly. If you have a common name, audit weekly during release windows.
Verification helps, but it is not the same as AI labeling
Verified by Spotify, launched in April 2026, is Spotify's biggest trust-layer move yet. The program reviews artist profiles for "signals of a real artist" — concert dates, merch, linked social accounts, off-platform presence — and badges the ones that pass. Spotify says more than 99% of actively searched artists will qualify at launch. AI-persona profiles are explicitly out.
Two things to read carefully. First, Verified is not the same as AI-labeled. BBC News covered the launch with a useful caveat: a track produced with AI assistance by a verified human artist will not be flagged as AI music; a fully synthetic project may slip through if its metadata performs the moves of a real artist. NPR reported that Spotify still does not take the same labeling approach as YouTube, Meta, or TikTok, despite community feedback in its own forum — and a moderator response confirming AI disclosures will not affect recommendations.
Second, the criticism that verification favors artists who are already winning is not paranoid. If your eligibility signals are tour dates, merch sales, social account graphs, and press mentions, the brand-new bedroom artist with three Bandcamp followers and zero shows is going to look more like a synthetic project than a real one to a system that cannot tell the difference. The point is not that the criteria are wrong; it is that the criteria push artists toward demonstrable off-platform existence.
That has implications for how you build a career, not just how you upload music.
Verification is not telling you which artists are real. It is telling you which artists left enough public evidence to look real. Those are not the same thing, and the gap is exactly what AI exploits.
Spotify's quieter trust layer: SongDNA, About the Song, and the credits build-out
Most coverage stops at verification. The more interesting work is happening on the song side, and indie artists should pay attention because it is where credibility now lives.
SongDNA, in beta, surfaces inside Now Playing for supported tracks. It maps the writers, producers, and collaborators on a song; samples; interpolations; and covers — and lets fans click through the relationships between songs. It is powered partly by Spotify-delivered and team-submitted data, partly by community sourcing, and eligible artist or label teams can review and manage components in Spotify for Artists.
About the Song, also beta, generates short story cards in Now Playing, summarized from third-party sources — context about inspiration, recording, behind-the-scenes moments. It is being framed as fan context. It is also, quietly, a provenance system: a song with About the Song content has been written about somewhere by someone in public. AI-spam tracks almost never clear that bar.
These features change which off-platform signals matter. Verified looks for tour and merch. SongDNA looks for clean, recognizable credits and creative relationships. About the Song looks for press, interviews, and articles. None of this is vanity. It is the public record that lets a recommendation system tell you apart from a content farm.
If you have ever wondered whether a press feature, a podcast appearance, or a small blog write-up is "worth it," this is the answer. Yes. Not for the traffic. For the provenance.
What independent artists should actually do this week
Skip the panic. The list is short and almost insultingly practical.
Claim and actively maintain your Spotify for Artists profile. Keep bio, photos, social links, and tour dates current. If you have shows, list them. If you have merch, link it. These are now signals, not extras. Our Spotify for Artists tips guide walks through it.
Turn on Artist Profile Protection if you are eligible, especially if your stage name is common or shares letters with another artist. Approve every release manually for the next quarter.
Audit your catalog monthly. Open your discography, scroll. Anything you did not release? Submit a takedown through your distributor immediately and report through Spotify's impersonation flow. Screenshot suspicious traffic spikes — sudden listener concentration in unfamiliar markets, weird hourly patterns — before they get cleaned up by Spotify's anti-fraud system. You may need the receipts.
Avoid stream-selling services. All of them. Even the ones that swear they use "real users." Spotify's anti-fraud tooling is now better than the services trying to game it, and a flagged catalog is harder to recover than slow growth. If you want streams, pay for legitimate ads, work playlist curators directly, and read our streams guide before you buy anything.
Build off-platform proof. Press, interviews, podcasts, video, tour dates, merch, mailing list, website. Verified by Spotify, SongDNA, and About the Song are all looking for the same thing — evidence you exist outside the feed. Build a small, real one.
Be wary of Discovery Mode as a substitute for owned audience. Trading royalty rate for algorithmic exposure can make sense in narrow cases. It does not build a fanbase you keep when the algorithm changes — and it always changes.
Stop outsourcing trust to the algorithm
Here is the part of the article that is going to sound like advice from someone who has watched this play out before, because it is.
The artists who do best in environments like this — where platforms add filters, badges, and verification because the underlying feed is getting polluted — are the ones who already have a route to their listeners that does not go through anybody else's recommendation system. A mailing list. A Discord. A small website. A Smart Link they share once and reuse forever.
This is where NotNoise sits in the workflow, and I am going to say it once and not pad it. A Smart Link page lets you put one shareable URL between your fans and the streaming/store/video destinations of their choice. You see who clicked, where they came from, what they chose, and who came back. Run a campaign and you get a real proof trail — not "12,000 streams from somewhere," but "417 clicks from this Instagram post, 38% chose Spotify, 19% chose Apple, 23% chose YouTube, and most came from the four cities I targeted." That is the texture an algorithm-only career never gives you, and it is exactly the kind of evidence verification systems quietly reward. Our smart link comparison walks through the options if you want to see how the category compares.
You do not need a tool for this. You need the habit of treating your audience as something you talk to, not something the algorithm rents to you.
The uncomfortable truth: AI did not create the problem; it made the loopholes cheaper
The framing the industry keeps reaching for — "AI is coming for music" — flatters the technology and lets the platforms off the hook. The actual story is more boring and more useful. Streaming royalties were always gameable. Stream farms existed before generative models. Identity confusion existed before voice cloning. The 1,000-stream threshold was a response to fraud that existed before The Velvet Sundown. Spotify's enforcement gaps existed before any of this.
What AI did was collapse the cost of producing plausible-sounding tracks at scale, which made every existing loophole cheaper to exploit. That is the real shift. Not "AI music will replace human artists." It is "AI music will industrialize streaming fraud and catalog pollution unless platforms build provenance, and provenance now includes you."
The artists who will be fine on the other side of this are not the ones who panic about AI, and not the ones who pretend nothing is changing. They are the ones who treat their catalog like a small business, their fanbase like a relationship, and their off-platform presence like infrastructure.
If you are building that infrastructure now, start a NotNoise account, put one Smart Link between your next release and the platforms it lives on, and keep the click data. You will not solve the AI music problem. You will solve the only part of it you can actually control: knowing who your listeners are and proving they are real.