On Deezer, some 75,000 titles generated by artificial intelligence are now uploaded every day, or nearly 44% of new content put online. In just over a year, this volume has multiplied sevenfold. However, these titles only account for between 1% and 3% of actual listening. The gap between the mass produced and the attention actually captured is so striking that it could intuitively suggest that all this is ultimately of no consequence for the artists’ business. Nay: if the rise of AI-generated music has not, at this stage, caused listeners’ preferences to waver, it does, however, profoundly modify the face of the offer. A wave which puts the distribution model to the test, but also risks hampering the visibility of pre-existing titles in the medium term.
The barrier to entry vanishes
The first and most immediate consequence relates to the very structure of the offer. The barrier to entry, already tenuous in the world of streaming, is now tending to disappear. Designing, testing and distributing a title is an almost instantaneous operation, at negligible cost.
In such a context, the growth of the catalog is no longer tempered by the cost of production: it comes from increased technical capacity, freed from any economic constraints. This dynamic establishes a regime of overabundance where the difficulty no longer lies in creation, but in capturing attention within an uninterrupted flow.
The distribution model under tension
Contrary to a widespread idea, according to which each artist would be paid directly based on the plays of their own titles, the dominant economic model of streaming, known as pro-rata, is based on a principle of pooling. Revenue from subscriptions and advertising is aggregated, then redistributed based on the share of each artist’s streams in total listening. This is the system that Spotify, Apple Music and Amazon Music still apply. In this context, the rapid multiplication of available titles leads to a fragmentation of listening: each new content captures a fraction, even if it is tiny, of the available attention. At a constant income envelope, this dispersion mechanically translates into an erosion of unit remuneration, without the titles generated by AI even needing to establish themselves in usage.
Aware of this vulnerability, Deezer switched to a so-called model in 2024. user-centric : each listener’s subscription is redistributed exclusively to the artists they actually listened to. This system mitigates the dilution effect by reestablishing a direct link between individual listening and remuneration. But it does not neutralize all the risks. In both models, catalog saturation weighs on the discoverability of works, and wiretapping fraud remains a structural threat.
Algorithmic recommendation, a blind spot in the debate
The pressure on recommendation algorithms deserves attention. As the catalog grows, automated prioritization of content becomes an increasingly uncertain exercise. Recommendation systems, designed to connect abundant supply and individual preferences, are faced with increasing background noise. The dissemination of micro-listenings over a long tail of titles contributes, even if marginally, to redistributing the revenue pool to the detriment of the most established works.
Added to this is the question of fraud. According to data communicated by Deezer, a notable proportion of listening attributed to AI-generated content is qualified as fraudulent and is subject to demonetization. This observation attests that the problem goes beyond just musical production to affect the very integrity of the distribution model.
AI, an economic arbitrage instrument
The ability to produce huge volumes of content at almost zero cost opens the way to unprecedented optimization strategies. The issue ceases to be exclusively artistic to become resolutely economic. Some players may be tempted to exploit remuneration mechanisms by multiplying titles, algorithmically evaluating their performance and, where appropriate, artificially inflating viewing counts.
A value that migrates rather than disappears
Beyond questions of fraud, the most structuring development is the gradual erasure of scarcity. In an environment where music production can be automated on a large scale, the value of content inevitably tends to shift.
Certain forms of music, particularly functional ones: ambient playlists, utility background sounds, lend themselves easily to automated generation. Others, based on artistic singularity, narration or the bond established with an audience, retain irreducible sources of valorization.
This dividing line outlines a progressive segmentation of the market: on one side, a plethoric and interchangeable offer; on the other, an embodied offer, whose value is based on criteria which transcend the sole counting of listenings.
The determining role of platforms
In this configuration, streaming platforms are given the role of arbiter. Deezer has undertaken several initiatives intended to contain the effects of this change: identification and reporting of AI-generated content, exclusion of these titles from algorithmic recommendations, demonetization of streams deemed fraudulent, and development of detection technology offered under license to the entire industry.
These measures reflect an undertaking of endogenous regulation of the system. They aim to dry up the economic incentives associated with mass production and to preserve, as much as possible, the balance of remuneration.
The general director of Deezer, Alexis Lanternier, places this approach in a collective perspective: musical generative AI is no longer an epiphenomenon and, as uploads intensify, the entire ecosystem will have to mobilize to defend the rights of creators and guarantee transparency towards listeners.
An issue that goes beyond the scope of platforms
The implications of this transformation extend far beyond streaming services. A study conducted with the participation of CISAC suggests a risk weighing on a significant part of creators’ income by 2028. Without prejudging the actual extent of this impact, this estimate highlights the vulnerability of the current model in the face of an accelerated change in production conditions.
At the same time, user expectations are evolving. A large majority of listeners express the wish that content generated by AI be clearly identified as such. Transparency thus emerges as a structuring element of the tripartite relationship between platforms, creators and the public.
A recomposition in progress
The rise of AI-generated music has not, to date, resulted in a direct eviction of artists. It highlights a deeper tension inherent in abundance and the redistribution of value.
In an environment where production is no longer subject to the constraint of scarcity, the capacity to capture attention, to order distribution and to preserve forms of singularity becomes the determining factor. The question is no longer only that of musical creation, but that of the conditions in which this creation is exhibited, valued and remunerated.
The trajectory remains open, we are only at the beginning of generative AI. It will be shaped by the decisions of the platforms, the evolution of the regulatory framework and the capacity of the economic model to absorb a wave which could turn into a tsunami. One thing, however, seems certain: the cardinal issue is no longer to produce more music, but to safeguard the conditions of its value.
