Deezer Says ~75,000 AI Tracks Hit Streaming *Every Day*—So Why Can’t Spotify (or You) Prove What’s Human?
Deezer’s AI-upload flood isn’t just a novelty—it’s a stress test for discovery, royalties, and identity. If platforms can’t verify provenance, “what you hear” becomes “what gets gamed.”
By TheMurrow Editorial
April 26, 2026
Key Points
- 1Deezer reports nearly 75,000 fully AI-generated uploads daily (44%), turning “new music” into an industrial firehose.
- 2Track labeling and downranking are becoming discovery infrastructure—because recommendation engines break before catalogs do.
- 3Fraud and identity risks rise as provenance blurs, pushing streaming toward shared standards for detection, disclosure, and amplification rules.
Seventy-five thousand tracks a day is not a music trend. It’s an industrial output figure.
On April 20, 2026, Deezer said it is now receiving “nearly 75,000” fully AI‑generated tracks per day—about 44% of all daily uploads to the platform. If that pace holds, the math is blunt: more than 2 million AI tracks a month flowing into a single streaming service’s intake pipe.
Most listeners won’t notice that torrent directly. They’ll feel it in subtler ways: recommendations that get worse, search results that look suspicious, playlists that start to feel like copy‑paste. The real fight isn’t about whether AI can make music. It’s about whether streaming can still prove, to anyone’s satisfaction, what they’re hearing—and why it’s being promoted.
“When nearly half of new uploads are synthetic, discovery stops being a feature and starts being a gatekeeping system.”
— — TheMurrow
Deezer’s number is a warning flare, not a brag
Deezer’s April 2026 claim—75,000 fully AI-generated tracks per day—lands differently because Deezer isn’t presenting it as growth. Deezer frames it as a trust and catalog integrity problem, the kind that can quietly break the user experience long before a platform admits anything is wrong.
The company also paired the statistic with a positioning statement: Deezer says it is the only streaming platform “transparently tagging” AI‑generated music at scale, in a consumer-visible way. Whether rivals would accept that framing is a separate question. The point is that Deezer is trying to turn labeling into a competitive advantage: the platform that can tell you what you’re hearing, and can show its work.
Deezer’s own numbers show how fast the upload environment has changed. When Deezer announced its AI detection tool on January 24, 2025, it said it was detecting about 10,000 fully AI-generated tracks per day—roughly 10% of daily deliveries. Fifteen months later, Deezer says the volume is up to nearly 75,000 per day, and the share has climbed to 44%.
Those are not incremental shifts. They suggest a world where the default assumption—new music equals new human artists—no longer holds.
The company also paired the statistic with a positioning statement: Deezer says it is the only streaming platform “transparently tagging” AI‑generated music at scale, in a consumer-visible way. Whether rivals would accept that framing is a separate question. The point is that Deezer is trying to turn labeling into a competitive advantage: the platform that can tell you what you’re hearing, and can show its work.
Deezer’s own numbers show how fast the upload environment has changed. When Deezer announced its AI detection tool on January 24, 2025, it said it was detecting about 10,000 fully AI-generated tracks per day—roughly 10% of daily deliveries. Fifteen months later, Deezer says the volume is up to nearly 75,000 per day, and the share has climbed to 44%.
Those are not incremental shifts. They suggest a world where the default assumption—new music equals new human artists—no longer holds.
75,000
Deezer says it now receives nearly 75,000 fully AI-generated tracks per day—about 44% of daily uploads.
44%
Deezer’s reported share of all daily uploads that are fully AI-generated as of April 20, 2026.
2M+
At Deezer’s stated pace, more than 2 million AI tracks a month could enter one service’s intake.
What “44% of uploads” actually means
A percentage can hide as much as it reveals. “Uploads” is not “streams,” and it’s not “listeners.” A flood of synthetic tracks could still earn near-zero attention—unless the flood is designed to game the system.
The deeper implication is operational. If almost half of new submissions are machine-made, platforms can’t rely on volume-based signals as a proxy for cultural vitality. They need mechanisms to separate:
- genuine creative output from bulk content
- artistic experimentation from spam
- audience demand from manipulation
Deezer is signaling that the separation step is now unavoidable.
The deeper implication is operational. If almost half of new submissions are machine-made, platforms can’t rely on volume-based signals as a proxy for cultural vitality. They need mechanisms to separate:
- genuine creative output from bulk content
- artistic experimentation from spam
- audience demand from manipulation
Deezer is signaling that the separation step is now unavoidable.
Key Insight
“Uploads” aren’t culture and they aren’t demand. At high AI volume, platforms must actively separate creativity from spam and manipulation.
How Deezer says it detects AI music—and what it does next
Deezer’s detection posture is unusually explicit. In its January 2025 announcement, Deezer said it had deployed an AI music detection tool and had filed two patent applications in December 2024 related to detection methods. In other words: this is not only a policy stance; it’s an attempt to formalize detection as proprietary infrastructure.
By January 2026, coverage of Deezer’s strategy noted the company had made its detection technology available for licensing—an invitation to distributors, labels, and rival services to adopt something like a shared technical standard. That move matters because detection only scales if it becomes part of the supply chain, not a bespoke afterthought for each platform.
Deezer’s operational policy is equally important: the company has said it tags fully AI-generated tracks and excludes them from algorithmic and editorial recommendations. The goal isn’t to ban AI music outright, but to reduce incentives to mass-upload low-effort content that might otherwise be boosted by discovery systems.
By January 2026, coverage of Deezer’s strategy noted the company had made its detection technology available for licensing—an invitation to distributors, labels, and rival services to adopt something like a shared technical standard. That move matters because detection only scales if it becomes part of the supply chain, not a bespoke afterthought for each platform.
Deezer’s operational policy is equally important: the company has said it tags fully AI-generated tracks and excludes them from algorithmic and editorial recommendations. The goal isn’t to ban AI music outright, but to reduce incentives to mass-upload low-effort content that might otherwise be boosted by discovery systems.
“Deezer isn’t trying to erase AI music. It’s trying to prevent AI music from hijacking the recommendation engine.”
— — TheMurrow
The accuracy claim—and why it needs context
Reporting around Deezer’s licensing push cited Deezer claiming ~99.8% accuracy and an ability to detect output from major generators such as Suno and Udio. That is a bold number, and readers should treat it as what it is: Deezer’s assertion, not an independently benchmarked industry standard.
Accuracy claims in detection are notoriously slippery without shared test sets, clear definitions (what counts as “fully AI-generated”?), and adversarial evaluation (how does the tool perform once people try to evade it?). Still, the practical fact remains: Deezer is acting as if detection is viable enough to shape product decisions and public commitments.
Accuracy claims in detection are notoriously slippery without shared test sets, clear definitions (what counts as “fully AI-generated”?), and adversarial evaluation (how does the tool perform once people try to evade it?). Still, the practical fact remains: Deezer is acting as if detection is viable enough to shape product decisions and public commitments.
99.8%
Deezer has claimed ~99.8% AI-detection accuracy in reported coverage—an assertion that still lacks shared independent benchmarks.
Why 75,000 AI tracks a day breaks discovery first
Streaming services rarely fail all at once. They degrade. The most fragile layer is discovery: recommendations, playlists, autoplay, and search ranking. A catalog can grow endlessly, but the pathways through it—what gets surfaced—have limited attention bandwidth.
Deezer’s policy to exclude fully AI-generated tracks from algorithmic and editorial recommendations is an admission of where the system snaps under pressure. Even if synthetic tracks aren’t “bad,” an unlimited supply of them makes it hard to ensure that discovery reflects listener intent rather than upload volume.
Three effects are likely to be felt by ordinary listeners:
- More noise in search: multiple near-duplicates, generic titles, and high-volume “artist” profiles.
- Weaker recommendations: algorithms trained on engagement can be fooled when content is produced to satisfy the machine rather than the listener.
- Erosion of editorial authority: human curators face a larger screening task as submissions balloon.
None of those are moral claims about AI. They’re structural claims about platforms built to reward abundance and pattern-matching.
Deezer’s policy to exclude fully AI-generated tracks from algorithmic and editorial recommendations is an admission of where the system snaps under pressure. Even if synthetic tracks aren’t “bad,” an unlimited supply of them makes it hard to ensure that discovery reflects listener intent rather than upload volume.
Three effects are likely to be felt by ordinary listeners:
- More noise in search: multiple near-duplicates, generic titles, and high-volume “artist” profiles.
- Weaker recommendations: algorithms trained on engagement can be fooled when content is produced to satisfy the machine rather than the listener.
- Erosion of editorial authority: human curators face a larger screening task as submissions balloon.
None of those are moral claims about AI. They’re structural claims about platforms built to reward abundance and pattern-matching.
A catalog is not a culture
Plenty of human-made music is forgettable. Some AI-made music may be compelling. The problem is scale without friction. When marginal cost collapses, “upload” stops being an indicator of commitment—time, money, risk, identity—and becomes a button press.
That changes what streaming is measuring. In a high-friction world, volume roughly correlates with cultural activity. In a low-friction world, volume often correlates with automation.
That changes what streaming is measuring. In a high-friction world, volume roughly correlates with cultural activity. In a low-friction world, volume often correlates with automation.
Key takeaway: Abundance distorts signals
When upload friction collapses, volume stops signaling commitment and starts signaling automation—warping search, recs, and editorial curation.
Royalty dilution and fraud: the incentive nobody wants to talk about
The most uncomfortable consequence of infinite music is financial. Streaming payouts are already small on a per-stream basis. When the number of tracks explodes, the competition for attention becomes harsher—and the temptation to cheat rises.
AI generation makes one part of the fraud equation cheap: producing massive catalogs quickly. The other part is distribution and manipulation—getting streams that look real enough to be paid.
A real-world case shows how direct the incentive can be. The U.S. Department of Justice charged Michael Smith, a North Carolina musician, alleging a scheme using AI-generated songs and bots to generate more than $10 million in royalties. Regardless of how individual platforms tighten their systems, the playbook is now widely understood: generate content at scale, simulate listening at scale, extract money at scale.
AI generation makes one part of the fraud equation cheap: producing massive catalogs quickly. The other part is distribution and manipulation—getting streams that look real enough to be paid.
A real-world case shows how direct the incentive can be. The U.S. Department of Justice charged Michael Smith, a North Carolina musician, alleging a scheme using AI-generated songs and bots to generate more than $10 million in royalties. Regardless of how individual platforms tighten their systems, the playbook is now widely understood: generate content at scale, simulate listening at scale, extract money at scale.
“AI didn’t invent streaming fraud. It lowered the cost of attempting it until it looked like a business model.”
— — TheMurrow
Why “downranking” is also an anti-fraud measure
Deezer’s approach—tagging and excluding fully AI-generated tracks from recommendation surfaces—reads, at first glance, like a cultural stance. It’s also a practical fraud deterrent.
If recommendation systems are the main funnel for passive listening, then blocking synthetic mass-uploads from those funnels reduces the payoff. A track can exist in the catalog without being handed free distribution by autoplay and personalized playlists. That distinction may become the new battleground: not what’s allowed to exist, but what’s allowed to be amplified.
If recommendation systems are the main funnel for passive listening, then blocking synthetic mass-uploads from those funnels reduces the payoff. A track can exist in the catalog without being handed free distribution by autoplay and personalized playlists. That distinction may become the new battleground: not what’s allowed to exist, but what’s allowed to be amplified.
The deception problem: when listeners can’t tell what they’re hearing
Even if fraud were solved tomorrow, streaming still faces a more intimate problem: identity. Listeners attach to artists, voices, scenes, and stories. When a track is synthetic—or when a human is impersonated—those attachments can be exploited.
Deezer-linked coverage points to a striking figure: a Deezer survey claim that 97% of respondents couldn’t distinguish AI from human tracks in a test. Without the full methodology—sample size, listening conditions, track selection—no responsible reader should treat the number as definitive. Still, it captures a plausible reality: many AI outputs are already “good enough” to pass casual listening.
That reality makes transparency matter even more. People don’t only want good sound. They want to know what they’re supporting, what they’re sharing, and what they’re being sold.
Deezer-linked coverage points to a striking figure: a Deezer survey claim that 97% of respondents couldn’t distinguish AI from human tracks in a test. Without the full methodology—sample size, listening conditions, track selection—no responsible reader should treat the number as definitive. Still, it captures a plausible reality: many AI outputs are already “good enough” to pass casual listening.
That reality makes transparency matter even more. People don’t only want good sound. They want to know what they’re supporting, what they’re sharing, and what they’re being sold.
97%
A Deezer-linked survey claim said 97% of respondents couldn’t distinguish AI from human tracks—methodology not fully provided in cited coverage.
Labels, not bans, are where trust may be rebuilt
Deezer’s stance suggests one possible compromise: keep AI music available, but label it clearly and avoid boosting it through automated discovery. That approach tries to protect three constituencies at once:
- Listeners, who want clarity and less spam
- Human artists, who want a fair chance to be found
- AI creators, who want a place in the market without pretending to be something else
The hard part is that labels are only meaningful if they’re accurate—and if the industry agrees on what exactly is being labeled.
- Listeners, who want clarity and less spam
- Human artists, who want a fair chance to be found
- AI creators, who want a place in the market without pretending to be something else
The hard part is that labels are only meaningful if they’re accurate—and if the industry agrees on what exactly is being labeled.
Editor's Note
This article treats figures like Deezer’s ~99.8% detection accuracy and 97% listening-test result as company-linked claims, not independent benchmarks.
Why Spotify—and everyone else—struggles to “prove what’s human”
Deezer’s messaging invites an obvious question: if one platform can tag AI music, why can’t others?
Start with the simplest answer: there is no universal, enforceable standard for what counts as AI-generated music, and no shared requirement for how that fact should be disclosed. The difference between “fully AI-generated” and “AI-assisted” isn’t philosophical; it’s operational. Consider how many steps in modern production can involve machine help:
- AI-generated vocals versus vocal tuning
- AI-written lyrics versus collaborative editing
- AI-composed instrumentals versus AI mastering
- a human prompt that yields a finished track versus a human who heavily edits the output
Without shared definitions, platforms face an impossible task: label everything, but agree on nothing.
Spotify, for its part, has signaled an approach that leans on policy enforcement and disclosures rather than claiming it can reliably detect all AI output. That posture makes sense: if detection is imperfect, false accusations can be costly, and motivated uploaders will adapt quickly.
Start with the simplest answer: there is no universal, enforceable standard for what counts as AI-generated music, and no shared requirement for how that fact should be disclosed. The difference between “fully AI-generated” and “AI-assisted” isn’t philosophical; it’s operational. Consider how many steps in modern production can involve machine help:
- AI-generated vocals versus vocal tuning
- AI-written lyrics versus collaborative editing
- AI-composed instrumentals versus AI mastering
- a human prompt that yields a finished track versus a human who heavily edits the output
Without shared definitions, platforms face an impossible task: label everything, but agree on nothing.
Spotify, for its part, has signaled an approach that leans on policy enforcement and disclosures rather than claiming it can reliably detect all AI output. That posture makes sense: if detection is imperfect, false accusations can be costly, and motivated uploaders will adapt quickly.
Detection is adversarial—and the adversary is motivated
Even if Deezer’s tool performs extremely well on today’s generators, detection is an arms race. Models change. Watermarks can be stripped. Audio can be post-processed. Generators can be tuned to mimic the artifacts of human production.
That doesn’t mean detection is futile. It means platforms need layered defenses: metadata disclosure, distributor accountability, behavioral analysis (upload patterns), and fraud monitoring—not just a single “AI detector” score.
The deeper issue is that streaming became the world’s largest music library without becoming the world’s clearest music registry. Proving authorship and provenance at scale was never built into the product. Now the bill is due.
That doesn’t mean detection is futile. It means platforms need layered defenses: metadata disclosure, distributor accountability, behavioral analysis (upload patterns), and fraud monitoring—not just a single “AI detector” score.
The deeper issue is that streaming became the world’s largest music library without becoming the world’s clearest music registry. Proving authorship and provenance at scale was never built into the product. Now the bill is due.
What the industry can do next: transparency as infrastructure
Deezer’s licensing move—making its detection tech available—implicitly argues that the industry needs shared rails. Not necessarily one company’s patents as a standard, but the idea that detection and labeling must be interoperable across platforms and distributors.
A workable transparency regime likely needs three components:
A workable transparency regime likely needs three components:
1) Shared definitions that match how music is made
“AI-generated” cannot be a single bucket if the goal is clarity rather than stigma. At minimum, platforms need consistent categories for:
- Fully AI-generated tracks
- AI-generated vocals (especially relevant for impersonation)
- AI-assisted production (mastering, stem separation, mixing help)
Without definitional clarity, a label becomes a vibe—useful for marketing, useless for trust.
- Fully AI-generated tracks
- AI-generated vocals (especially relevant for impersonation)
- AI-assisted production (mastering, stem separation, mixing help)
Without definitional clarity, a label becomes a vibe—useful for marketing, useless for trust.
2) Supply-chain accountability, not only platform enforcement
Uploaders are often intermediated by distributors. If disclosure is required, it must be required upstream, where metadata is created and validated. Platform-only enforcement is like checking passports after people have already boarded the plane.
3) Product design that reduces incentives to flood
Deezer’s decision to exclude fully AI-generated tracks from recommendations is a product choice with economic consequences. Platforms could also consider:
- throttling uploads from accounts that behave like factories
- adding friction for bulk deliveries
- routing flagged content away from discovery surfaces
None of that requires banning AI. It requires refusing to subsidize low-trust content with high-trust distribution.
- throttling uploads from accounts that behave like factories
- adding friction for bulk deliveries
- routing flagged content away from discovery surfaces
None of that requires banning AI. It requires refusing to subsidize low-trust content with high-trust distribution.
A workable transparency regime
- 1.Define categories that match real production (fully AI, AI vocals, AI-assisted).
- 2.Require disclosure upstream via distributors and validated metadata.
- 3.Design discovery to reduce incentives for flooding (throttles, friction, rerouting).
What it means for listeners and artists—practical takeaways
The biggest mistake is to treat AI music as a niche debate among technologists. The consequences land on everyday behaviors: what you hear, what you believe, what you pay for.
For listeners: how to keep your feeds human-shaped
- Lean on intentional listening: follow artists directly, save albums, and build libraries. Passive autoplay is where spam thrives.
- Be skeptical of anonymous abundance: dozens of releases from a brand-new “artist” in a week is a signal, not a flex.
- Look for labeling when platforms offer it: if a service tags AI-generated tracks, use that information to decide what you want in your rotation.
- Be skeptical of anonymous abundance: dozens of releases from a brand-new “artist” in a week is a signal, not a flex.
- Look for labeling when platforms offer it: if a service tags AI-generated tracks, use that information to decide what you want in your rotation.
For artists: why transparency isn’t the enemy
Human artists have the most to lose from a system where identity can be simulated and output can be bulk-produced. Clear labeling, improved detection, and tighter recommendation rules are not anti-innovation; they’re pro-authorship.
At the same time, artists experimenting with AI tools have a legitimate interest in not being treated as frauds. The policy target should be deception and manipulation—not the mere presence of machine assistance.
At the same time, artists experimenting with AI tools have a legitimate interest in not being treated as frauds. The policy target should be deception and manipulation—not the mere presence of machine assistance.
For platforms: trust is now a product feature
Deezer is betting that transparency is a differentiator. If users begin to associate certain platforms with cleaner discovery and clearer provenance, “trust” becomes as important as catalog size.
The open question is whether the market will reward that bet—or whether the industry will drift until regulators, labels, and courts force a standard from the outside.
The open question is whether the market will reward that bet—or whether the industry will drift until regulators, labels, and courts force a standard from the outside.
Conclusion: the future of streaming is a provenance problem
Deezer’s 75,000 AI tracks per day figure is not just about how fast machines can compose. It’s about how quickly streaming’s assumptions are failing: that upload volume is meaningful, that discovery can be trusted, that an artist identity maps to a person, that royalties are hard to steal at scale.
Deezer says it can tag fully AI-generated music and keep it out of recommendation systems. It has filed patents, deployed detection, and offered licensing. Those are real moves, even if the underlying accuracy claims still need independent scrutiny.
The larger truth is simpler. Streaming used to be a question of access. Now it’s a question of proof. And the platforms that learn to verify what they distribute—without strangling legitimate creativity—will decide what music culture feels like in the AI era.
Deezer says it can tag fully AI-generated music and keep it out of recommendation systems. It has filed patents, deployed detection, and offered licensing. Those are real moves, even if the underlying accuracy claims still need independent scrutiny.
The larger truth is simpler. Streaming used to be a question of access. Now it’s a question of proof. And the platforms that learn to verify what they distribute—without strangling legitimate creativity—will decide what music culture feels like in the AI era.
T
About the Author
TheMurrow Editorial is a writer for TheMurrow covering entertainment.
Frequently Asked Questions
What did Deezer actually claim about AI music uploads?
On April 20, 2026, Deezer said it is receiving nearly 75,000 fully AI-generated tracks per day, representing about 44% of all daily uploads. Deezer framed the figure as a platform integrity challenge—too much synthetic volume can distort discovery and weaken user trust if it isn’t clearly labeled and managed.
How does Deezer say it handles fully AI-generated tracks?
Deezer has said it tags fully AI-generated tracks and excludes them from algorithmic and editorial recommendations. That approach doesn’t ban AI music from the catalog, but it limits its ability to benefit from discovery features that can drive passive listening and, potentially, monetization at scale.
When did Deezer start using AI music detection?
Deezer announced it had deployed an AI music detection tool on January 24, 2025. At that time, Deezer said it was detecting around 10,000 fully AI-generated tracks per day, or roughly 10% of daily deliveries—far lower than the share Deezer reported in April 2026.
Is Deezer’s AI detection accuracy independently verified?
Not in the reporting cited here. Coverage of Deezer’s licensing push reported Deezer claiming ~99.8% accuracy and detection of generators like Suno and Udio, but those are company claims. Independent benchmarking would require shared test methods and datasets, plus evaluation against attempts to evade detection.
Why does a surge in AI tracks matter if listeners can ignore them?
Discovery systems don’t ignore volume—they ingest it. If synthetic tracks make up a large share of new uploads, search and recommendations can become noisier, and editorial teams face greater screening pressure. Even if many AI tracks get few plays, the sheer supply can still distort how platforms surface music.
How does AI music relate to streaming fraud?
AI can make fraud cheaper by enabling mass production of tracks. A DOJ case alleged that Michael Smith used AI-generated songs and bots to generate more than $10 million in royalties. The basic risk is that low-cost content plus automated fake listening can siphon money unless platforms detect and deter manipulation.
