Why Is Deezer Deciding What Counts as “Real” Music?

Deezer wants the public conversation about its AI policy to begin with a reassuring idea: protect human artists, protect listeners and stop fraudulent streams from contaminating the music economy. Those are difficult objectives to oppose. The problem begins when a streaming company moves from detecting suspicious behaviour to becoming an authority over the nature of the music itself.

Deezer says it detects fully AI-generated recordings, labels them and removes them from algorithmic recommendations and editorial playlists. It has gone further by licensing its detection technology to other organisations and positioning the technology as a commercial capability. That means Deezer is no longer simply operating a music catalogue. It is building a classification system capable of influencing how music is described, discovered and commercially treated. That should concern every independent artist, including artists who have never used generative AI, because the underlying question is not simply whether Deezer can detect synthetic music.

The question is whether an increasingly powerful platform should be trusted to decide what category your creativity belongs in, what your audience is told about it and how much visibility it receives after the machine has made its judgement.

The old music industry gatekeeper was visible. It was the record executive who rejected a demo, the radio programmer who refused a track, the retailer that decided which records deserved shelf space or the publisher who decided which songwriter was commercially useful. Streaming was supposed to break some of that architecture apart by making distribution dramatically cheaper and giving independent creators access to global audiences.

But gatekeeping did not disappear. It migrated into software. Deezer’s Flow recommendation system already determines what listeners are presented with based on their listening behaviour, and Deezer now says AI-generated tracks are excluded from algorithmic recommendations and editorial playlists. The modern gatekeeper therefore does not necessarily tell you that your music is forbidden. It can simply make your music harder to discover. That is arguably more sophisticated because the artist may never receive a dramatic rejection. There is no executive saying no. There is simply an invisible reduction in opportunity produced by a system the artist cannot see, cannot inspect and cannot directly control.

The justification for this system becomes easier to understand when you look at the numbers Deezer itself publishes. In January 2025, Deezer said its detection system was identifying roughly 10,000 fully AI-generated tracks per day, around 10% of daily deliveries. By April, the figure had risen above 20,000 per day, representing more than 18% of uploads. By September, Deezer reported more than 30,000 fully AI-generated tracks arriving each day. By April 2026, it was reporting almost 75,000 per day, roughly 44% of daily deliveries, and by July it reported peak levels of approximately 90,000 AI-generated tracks per day, exceeding half of all new uploads at that level.

Those numbers describe a genuine industrial transformation. The problem is that an industrial-scale problem naturally encourages an industrial-scale response. Once tens of thousands of tracks arrive every day, human examination becomes impossible. Automation becomes inevitable. But the fact that automation is necessary does not mean its decisions should become unquestionable.

There is an important distinction between detecting a technical characteristic and defining artistic legitimacy. Deezer can say that a recording contains characteristics associated with fully AI-generated production. That is a technological claim. But once Deezer attaches a visible label to the recording, excludes it from recommendations and excludes it from editorial playlists, the technological classification becomes a cultural and economic decision. The platform is effectively saying that this category of music should be treated differently from other music. That may be entirely defensible when the goal is to prevent fraudulent streaming operations from exploiting royalty systems.

But it becomes much more complicated when the same infrastructure is applied to questions of artistic identity. A detector does not understand why an artist created something. It does not understand the emotional reason behind a production choice. It does not understand the artist’s philosophy. It analyses the recording and produces a classification. The corporation then decides what that classification means inside its ecosystem.

That second step is where the power lives.

Deezer itself generally speaks in terms of “fully AI-generated” music rather than literally defining all other music as “real.” That distinction matters. Yet the practical effect of the system can create precisely that cultural binary in the mind of the listener: AI on one side, human on the other. Deezer’s own research found that 97% of respondents could not distinguish fully AI-generated music from human-created music in its blind test, while 80% said fully AI-generated music should be clearly labelled. The consumer therefore increasingly relies on the platform to tell them what they cannot reliably determine themselves.

That gives Deezer extraordinary interpretive authority. The listener does not merely hear a recording. They hear a recording accompanied by the platform’s explanation of what it believes that recording is. Once millions of people accept that explanation without questioning it, a corporate classification begins functioning like a cultural fact.

Deezer says its detector identifies fully AI-generated music with 99.8% accuracy and estimates that it may miss around 0.2% of AI-generated tracks. That is an impressive claim, but an accuracy percentage does not answer every question an artist should ask. It does not tell the public how individual disputes are investigated. It does not tell an artist what evidence led to a particular classification. It does not establish that every borderline creative situation will be interpreted correctly.

Most importantly, it does not explain what happens when the artist and the system disagree. A detection model can be highly effective at identifying patterns while still being unsuitable as the final authority over a person’s creative identity. Those are separate questions. The industry should stop pretending that “the detector is accurate” automatically means “the platform’s decision is beyond challenge.”

The fundamental weakness of automated classification is that the machine sees the finished recording while the artist knows the production history. Those are radically different sources of information. A musician can know exactly which instruments were recorded, which performances were played, which samples were used, which software processed the signal, which sections were edited, which sounds were synthesised and which parts were created manually.

The final waveform does not contain every piece of that history in a form a detection system can necessarily reconstruct. A serious dispute therefore cannot be settled simply by pointing at the detector. If an artist challenges a classification, the platform should be prepared to examine contextual evidence rather than treating the algorithm as a judge whose verdict automatically overrides the creator’s account.

Otherwise, the artist is effectively being asked to prove their humanity to a machine.

For a multinational platform, an incorrect classification can be one more entry in a database. For an independent artist, it can become part of the public identity of a release. An artist may have spent months developing a record, building artwork, writing lyrics, recording performances and establishing a distinctive production style. The platform can then attach a classification that the audience encounters before they have any opportunity to understand the creative process.

Deezer’s system is designed to label albums containing fully AI-generated tracks, while detected AI recordings are removed from recommendations and editorial playlists. The consequence is not merely descriptive. It can affect visibility. That distinction is critical because independent musicians do not have enormous promotional budgets to compensate for algorithmic invisibility.

Deezer’s stated reasoning is that removing fully AI-generated tracks from recommendations prevents them from diluting the royalty pool and protects artists. There is a legitimate economic argument here. Deezer says that although fully AI-generated music represents only around 1–3% of total streams, up to 85% of streams associated with such tracks were identified as fraudulent in 2025. If people are generating enormous volumes of synthetic tracks specifically to manipulate streaming systems, allowing those operations to compete for recommendation traffic and royalties would be absurd.

But the policy reveals something deeper about platform power: once Deezer decides that a category is associated with undesirable economics, it can change the economic conditions under which that entire category exists. The platform does not merely observe the market. It actively reshapes the market through recommendation and visibility rules.

This becomes even more interesting because Deezer is not simply spending money on AI detection as a defensive measure. It has begun monetising the technology. In 2026, Deezer announced licensing arrangements for its AI detection system and described AI-detection monetisation as one of its strategic priorities. That does not establish wrongdoing. It does, however, create a commercial incentive around the technology. The detector has become an asset.

Deezer has patents pending or published around its detection methods, and the company is selling access to the capability beyond its own platform. This means the industry should ask uncomfortable questions about governance, transparency and conflicts of interest. When a platform both classifies music and commercially licenses the technology responsible for that classification, who independently audits the system? Who challenges its assumptions? Who represents the interests of artists who disagree?

Those are not anti-technology questions.

They are accountability questions.

The danger is not that Deezer has an AI policy. The danger is what happens when the policy becomes infrastructure across the music ecosystem. Deezer has already made its detection technology available to outside organisations, including collective management organisations.

Once multiple organisations begin relying on the same classification technology, an error or limitation can propagate. One platform’s judgement becomes another organisation’s input. One database becomes another database’s evidence. Eventually, an artist may face several institutions repeating the same classification because they all rely on related technological systems. At that point, the original judgement becomes difficult to challenge precisely because it appears to have independent confirmation. But repetition is not the same thing as independent verification.

Music has always been obsessed with categories. Genres, demographics, formats, markets, scenes and commercial classifications have repeatedly been used to determine which artists receive attention and which artists are pushed aside. The difference now is that classification can happen at machine speed. An executive could once misunderstand an artist.

Now a model can misunderstand thousands of artists at once. The potential scale of the mistake is therefore fundamentally different. Technology can eliminate human inconsistency, but it can also industrialise institutional assumptions. If the underlying category is simplistic, automation does not make the category wiser. It simply makes the category faster.

There is another uncomfortable issue that the music industry needs to confront: the word “AI” can easily become a moral label rather than a production description. Music has never had a clean boundary between technology and human creativity. Electronic instruments, samplers, drum machines, digital workstations, pitch correction, software synthesis, automated mastering and algorithmic tools have all changed the production process. Some musicians embrace them.

Others reject them. Neither position determines whether the resulting work has artistic value. Generative AI introduces genuinely serious questions about training data, consent, copyright, employment, artistic displacement and fraud. Those questions deserve aggressive scrutiny. But collapsing every discussion into “human versus machine” is intellectually lazy. The industry needs more precision, not more slogans.

This is where the discussion should become much more critical. The central problem is not simply that Deezer uses AI. The problem is that a corporation operating a private platform can increasingly determine the context in which music is presented to the public. It decides what gets recommended. It decides what gets editorial exposure. It decides what gets labelled. It decides which classifications affect discovery.

It can commercialise the technology behind those classifications. That is an extraordinary concentration of cultural power. The music itself may belong to an artist, but the pathway through which millions of listeners encounter that music belongs to the platform. Ownership of the recording is therefore only part of the battle. Control over visibility and interpretation has become another form of power.

Listeners should not assume that a streaming interface is a neutral window into music. It is a curated environment built by a corporation with commercial objectives, technical limitations and strategic priorities. What you see is selected. What you hear is filtered. What is recommended is calculated. What is labelled is classified. What is excluded is governed by policy. Deezer’s AI system makes this unusually visible because the company openly tells users that detected AI-generated recordings are removed from recommendation systems and editorial playlists. That may be useful for consumers who want to avoid fully AI-generated music. But consumers should also understand the broader implication: the platform is actively shaping the cultural marketplace rather than merely providing access to it.

Independent artists have lived with this problem for years in different forms. Platforms demand metadata, artwork specifications, genre information, release dates and other classifications because databases require standardisation. The artist adapts. But AI classification introduces a different level of intrusion because the platform is no longer simply asking how you describe your work. It is potentially telling the audience what your work is. That is a fundamental shift. The creator becomes the subject of a machine-generated statement that may be more visible than the creator’s own explanation.

An independent artist should never underestimate how powerful that reversal is.

This is precisely why maintaining a direct relationship with listeners matters to projects such as UNIDARK. When the artist’s music exists only inside a streaming ecosystem, the platform can become the dominant source of information about the release.

When the artist maintains an independent destination, that balance changes. The Official UNIDARK Hub provides a direct route to the project, its releases, videos, lyrics and wider identity, while the Official UNIDARK Store gives listeners a way to support the music directly rather than making a streaming platform the sole economic bridge between creator and audience. That matters even more when a platform places a disputed or unexplained classification around a release. The artist needs somewhere outside the platform where the work can be presented in the artist’s own words.

Direct sales are not merely about making a few extra pounds from a download or physical product. They change the relationship between artist and audience. The platform is no longer the only place where the listener can find the release. The artist can control the presentation, the explanation, the catalogue and the commercial relationship. That does not make streaming irrelevant. Streaming remains useful for discovery and listening. But there is a profound difference between using a platform and depending upon one. The independent musician should be building an ecosystem in which Deezer is one distribution channel among many rather than the institution through which the artist’s legitimacy is established.

If Deezer wants to become an authority on AI-generated music, it should accept the responsibilities that come with authority. Artists should be able to understand the basis for consequential classifications. They should have a meaningful mechanism for challenging decisions. Disputes should not disappear into automated support systems. The company should explain how detection technology is independently evaluated and how false classifications are handled. The more the detector influences visibility, recommendation and commercial treatment, the stronger those safeguards need to become.

A powerful system should require powerful accountability.

The industry should resist the temptation to worship technological certainty. AI detection is useful. Automated fraud detection is useful. Recommendation systems are useful. None of them should be treated as infallible. The history of technology is full of systems that worked impressively under normal conditions and failed spectacularly in unusual ones.

Music is an especially difficult environment because creativity is deliberately full of unusual cases. Artists combine technologies. They manipulate sounds. They experiment with processes. They create things that do not fit neatly into categories. A platform designed to classify enormous quantities of content will inevitably encounter music that challenges its assumptions.

The artist should not be punished simply because the art is complicated.

There is an important contradiction at the heart of the platform’s messaging. Deezer says its AI policy exists to safeguard artists and songwriters, protect remuneration and create transparency for fans. Those goals are worthwhile. But protecting artists cannot mean removing agency from artists. If the platform can decide what category a recording belongs to, change how that recording is discovered and potentially influence how the audience perceives it, then artist protection must include the right to challenge the platform’s interpretation. Otherwise, “protection” becomes paternalism. The corporation decides what is best for the creator, and the creator is expected to accept the decision.

That is not independence.

That is dependency with better branding.

The AI debate is therefore only the beginning. The deeper battle is over who gets to define music in the digital age. Artists create it. Listeners interpret it. Platforms classify it. Algorithms recommend it. Corporations monetise it. Rights organisations regulate parts of it. AI companies increasingly generate it. The boundaries between these roles are becoming increasingly blurred. Whoever controls the classification systems gains an enormous amount of influence over what the public believes it is seeing and hearing.

Deezer has chosen to step directly into that territory.

Artists should be watching very carefully.

An artist appearing on Deezer is not necessarily an independent artist in the meaningful sense of the word. Distribution may be open, but the surrounding infrastructure remains controlled by someone else. Your recording can be yours while your discovery is mediated by another company’s algorithm. Your identity can be yours while your categorisation is determined elsewhere. Your audience can love your music while never actually having a direct relationship with you.

That is the uncomfortable reality of platform-based independence.

You may own the art.

But someone else may own the doorway.

The answer is not to abandon every streaming service. It is to stop treating streaming services as permanent foundations. Platforms change. Companies change ownership. Algorithms change. Policies change. Moderation systems change. Commercial priorities change. An artist who builds everything around one platform is effectively outsourcing part of their career to a corporation whose priorities can change without their permission.

Build your own archive.

Build your own website.

Build your own store.

Build your own audience.

Keep your production records.

Keep your registrations.

Keep your original files.

Keep your evidence.

And make sure your listeners know where to find you if the platform changes the rules tomorrow.

The most dangerous sentence in modern music is not “AI generated.” It is the unspoken assumption that because a platform has classified something, the classification must be true in every meaningful sense. Deezer’s detector may be extremely sophisticated. Its published accuracy claim may be impressive. Its fight against fraudulent streaming may be necessary. Its desire for consumer transparency may be legitimate. None of those facts give the company unlimited authority over artistic identity.

A detector can identify.

A platform can label.

A recommendation system can promote.

A corporation can monetise.

But none of them should become the final judge of what your music means.

Deezer has built an impressive technological system around a genuinely difficult problem. The volume of synthetic music entering streaming services is enormous, and the company has documented serious evidence of fraudulent streaming associated with fully AI-generated catalogues. But the seriousness of the problem is precisely why the industry cannot afford lazy governance. The answer to industrial-scale manipulation cannot be industrial-scale unquestioned authority. If Deezer wants to classify music at unprecedented scale, it must accept unprecedented responsibility for explaining and defending those classifications.

Because once the platform becomes the institution that tells the audience what a recording is, it has crossed a line.

It is no longer simply distributing music.

It is interpreting culture.

And whenever a corporation gains the power to interpret culture at scale, artists should be asking one question above everything else:

Who watches the machine?

UNIDARK, also known as Morning Star, is a UK-based independent extreme metal producer and the creator of Blackdeathgrin Metal — an original extreme metal genre combining elements of black metal, death metal, deathcore, and grindcore.

Explore the Official UNIDARK Hub

The central directory for music, releases, streaming platforms, videos, lyrics, Blackdeathgrin Metal information, and all official project links.
→ Official UNIDARK Hub

Explore Official Lyrics & Song Meanings

Discover the complete UNIDARK lyrics archive, song interpretations, and the concepts behind the music.
→ Lyrics Archive

Watch Official UNIDARK Music Videos

Experience the visual side of the project through official releases and video content.
→ Video Archive

Explore Metal Insights

Read more deep articles covering extreme metal genres, underground music, industry topics, production, platforms, culture, and the evolution of heavy music.
→ Metal Insights

Explore Sync Licensing

Read more deep articles covering music licensing, sync rights, music for film, television, games, advertising, trailers, independent production, hybrid music production, catalogue clearance, and the evolving relationship between music and visual media.
→ Sync Licensing

Explore the UNIDARK FAQ Archive

Find answers about UNIDARK, Morning Star, Blackdeathgrin Metal, releases, the creative process, and the philosophy behind the project.
→ UNIDARK FAQ

Support UNIDARK Directly

Support independent extreme metal creation through the official UNIDARK store.
→ Official UNIDARK Store

Join the Official UNIDARK Transmission

Subscribe for new releases, Metal Insights articles, official announcements and updates directly from UNIDARK. Follow the evolution of Blackdeathgrin Metal and the continuing development of the UNIDARK project.