Music and AI : Are algorithms taking over streaming?

 

In less than two years, AI-generated music has transformed from a technological curiosity into an industrial phenomenon shaking the foundations of the sector. Between new creative uses, economic abuses, and threatened artists, music streaming is undergoing a complete reshuffling. And somewhere at the end of this chain are we, the listeners, who deserve to know what we are truly listening to.

A surge in numbers

The figures are dizzying, and their progression is breathtaking. In September 2025, Deezer reported over 30,000 entirely AI-generated tracks delivered daily, representing 28% of its total daily deliveries. By early 2026, this flow had reached 60,000 AI-generated tracks per day, accounting for 39% of deliveries. These figures come from two distinct communications from Deezer: their direct comparison provides a clear order of magnitude, even if the counting methods may slightly differ between publications.

The main reason for this explosion? Creative barriers have fallen. Tools like Suno or Udio (text-to-audio generators, meaning AI capable of producing a complete track from a simple text description, without any musical skills required) now allow anyone to compose and publish in a few minutes. The profiles of those publishing these tracks are extremely varied: enthusiasts without technical training, authors looking to bring their texts to life, curious individuals attracted by novelty, and more rarely, actors who exploit it in indirect ways.

But who actually listens to this music? For now, its impact on audiences remains limited: it represents between 1% and 3% of listens on Deezer depending on the month, a variable range rather than a fixed measure. However, this does not mean that the phenomenon is without consequences.

The Velvet Sundown: the AI music that revealed everything

Summer 2025 marked a symbolic turning point. In just six weeks, an indie rock band with a vintage sound called The Velvet Sundown released three albums and attracted, according to several specialized media outlets, over a million monthly listeners on Spotify. Guitar riffs, a slightly raspy voice, a polished seventies aesthetic: nothing initially betrayed an AI creation. The band even provided detailed biographies for its four fictional members: Gabe Farrow, Lennie West, Milo Rains, and Orion "Rio" Del Mar.

It was eventually Reddit that lifted the veil, after users noticed visual inconsistencies in the band's photos and a complete lack of presence on social media. On July 5, 2025, The Velvet Sundown admitted to being "a synthetic music project guided by human artistic direction, composed and illustrated with the support of artificial intelligence."

Whether it was an assumed artistic approach or a clever communication operation, the affair primarily highlighted the system's flaws. The shock caused in the music industry led Spotify to announce, according to several specialized media sources, the removal of 75 million problematic tracks and the deployment of a new generation anti-spam filter.

When some play with the rules of the game

While the vast majority of those who publish AI-generated music do so out of curiosity or passion, a minority has organized to exploit platform mechanisms for purely lucrative purposes.

The most common scheme: publishing massive volumes of tracks to capture micro-royalties on a large scale. Some identified profiles in early 2026 published up to one album per day, changing styles each week to reach a maximum of niches: children's music, jazz fusion, metal...

• Deezer found that up to 85% of AI-generated track listens were fraudulent in 2025 depending on the month, compared to 8% fraud across the entire catalog. Non-authentic streams are excluded from royalty calculations.
• In the United States, Michael Smith was indicted for unduly collecting over $10 million in royalties by publishing hundreds of thousands of AI songs and having them listened to by bots.

Identity theft: the most serious abuse

The most worrying form of abuse does not concern AI artists, but real artists whose identity is stolen. Two cases, widely documented by specialized media, concretely illustrate this risk.

Paul Bender denounces a system where it is enough to declare "it's me" to add a song to any artist's profile. A gaping flaw in the platforms' verification processes.

The listener, often forgotten in the debate

In all these discussions, one dimension is often overlooked: that of the listener. Yet it is we who, every day, inadvertently suffer the consequences of this revolution, without having our say.

According to a study conducted by Deezer among 9,000 people in eight countries:

• 97% of respondents could not tell the difference between AI-generated music and human music.
• When confronted with this misconception, 50% of them said they felt "uncomfortable."

This is a matter of basic consent. One might be open to AI-generated music, curious to explore it, or conversely, prefer to stick to entirely human creations, out of artistic conviction or simple personal preference. In both cases, the choice should be ours. Allowing an algorithm to silently slip synthetic tracks into our discovery feed, without a label or filtering option, takes away this freedom.

Deezer has taken a first step by labeling content detected as AI-generated and excluding it from its recommendations. But the real solution would lie elsewhere: a setting that each user could activate or deactivate according to their wishes, just as one chooses today whether or not to display lyrics. A simple and transparent filter, which would return to each what belongs to them.

Human artists on the front lines

These abuses have very real economic consequences. According to projections from a study commissioned by the International Confederation of Societies of Authors and Composers (CISAC), artists and composers could lose up to 24% of their income by 2028, representing $4.64 billion per year. These figures are prospective estimates, not established facts.

The mechanism is relentless: platforms distribute their revenue proportionally to total listens. The more the volume of tracks in circulation increases, the less each artist receives per listen. An overrepresentation of automated productions also reduces the visibility of human artists in recommendation algorithms.

Faced with this reality, some professional musicians choose adaptation. Artists like D4N1EL, cited by RTS in a March 2026 investigation, integrate AI as a complementary tool, spending several weeks on each song to maintain their authenticity. A hybrid approach that also raises questions: how to distinguish an "assisted" work from an entirely synthetic one?

Mobilization is also reaching institutions: Stéphane Laick, president of FÉLIN (National Federation of Independent Labels) and co-founder of the AT(h)OME label, brought the voice of independent labels directly to the National Assembly, alongside other industry stakeholders.

Platforms with diverging responses

Please note: no detection tool is 100% infallible. AI models are evolving rapidly, and some generated content may still slip through the net, especially when mixing human and synthetic elements.

A legal vacuum that complicates everything

Legally, the situation remains unclear. A track composed entirely by AI, without creative input from a human artist, may not benefit from copyright protection depending on the jurisdiction, and may therefore fall into the public domain upon its creation. However, if an artist uses AI merely as a tool in a creative process they lead, they can claim authorship of the final work. The line remains delicate to draw, and legislation is evolving.

At the end of October 2025, Universal Music announced an agreement with Udio, settling a copyright dispute and planning the launch in 2026 of a joint platform, a first that could pave the way for other compromises between majors and music generation services.

Nevertheless, encouraging legislative signals are emerging:

• In France, the Senate has received the green light from the Council of State to establish a presumption of use of cultural content by AI providers, a first concrete legal lever for creators.
• At the European level, the European Parliament adopted the Voss report, which imposes full transparency on works used to train AI models and guarantees fair remuneration for artists and journalists via a licensing system.

A matter of choice, above all

Figures like Elton John, Dua Lipa, and Paul McCartney have publicly pleaded for stricter restrictions, fearing that the proliferation of these projects could jeopardize the diversity and originality of musical creation.

Music has always adapted to new technologies: from electric guitars to synthesizers, from sampling to home studios. AI is the next step, and it would be naive to think we can escape it. If you want to understand how it is concretely transforming the work of creators in the studio, we have dedicated a comprehensive article to the impact of AI in music production.
What the most discerning artists have understood is that it's not about choosing "for" or "against," but about adapting by setting clear rules: transparency on the origin of content, protection of artistic identities, and fair remuneration for those who create, regardless of the method. Because at the end of the chain, there are us, the listeners. And we deserve, at a minimum, to know what we are listening to, and to be able to choose.