Music created by AI has reached a milestone. According to Deezer, in January 2026, 60,000 AI-generated tracks were uploaded daily to the platform, accounting for 40% of new uploads. At the same time, 97% of listeners cannot distinguish between AI music and human creations, according to a study from October 2025, but 52% feel uncomfortable when they learn about it. In this landscape, Qobuz released a zero-tolerance charter in February 2026 for content that is 100% AI-generated. Marc Zisman, the platform’s music director, shared his insights on the situation.
We can’t tell the difference between AI music but feel uncomfortable when we learn about it – What’s at play here between the ear and consciousness?
Anyone could be fooled. Today, we can do amazing things, and I have no doubt that almost 100% of people listening to certain tracks could also be deceived. Like them, I would say that I feel uncomfortable.
There was a well-publicized example of the “group” The Velvet Sundown, which resembled Fleetwood Mac, with 70s rock looks. In a street interview on the Champs-Élysées, the song was played, and people said “it’s great.” In the end, the journalist revealed that it was AI-generated. 80% of passersby were disappointed or resistant, and perhaps 10% said “I don’t care, the track is not bad.”
I often draw parallels with food composition. You can find something delicious, and then learn it contains disruptors or harmful ingredients. Some will continue to consume, others won’t. In this case, the role of a streaming platform is to inform the consumer: this track is 100% AI-generated. Like a Nutri-Score on a packet of cookies, we provide information and let the person make their choice.
What defines the contract between an artist, a work, and its audience, and when does AI break it?
At each stage of the chain, everyone has their responsibility. The musician, the label that chooses whether or not to use AI artists, the platform that decides whether or not to inform the consumer. And at the end, the audience, once all information has been provided to them. There isn’t an innocent or guilty party; it’s a technological advancement.
Copyright should remain. A painting for my living room or a piece composed for my friends, no. But once it’s commercialized, an author should be remunerated for their creation, especially if another creation is generated from it. But there are nuances between extremes. Timbaland created a 100% AI artist that he did not hide. On the other hand, the sly person on the other side of the world who generates thousands of titles like Tyler, The Creator with Suno, just to collect royalties, that’s a completely different approach.
Qobuz introduced an AI charter with a zero-tolerance policy on 100% AI-generated content in February 2026. How is this policy applied in practice, and is it sustainable on a larger scale?
Every day, we remove and delete. Our internal tools quickly detect thousands of AI-generated tracks, either linked to a label we’ve never heard of or with huge stream volumes but very few subscribers, indicating bots. Subscribers also contact customer service saying “I spotted this,” contributing to the cleanup.
(Blockquote) “There could be billions of tracks on platforms, but days will always be only 24 hours.” – Marc Zisman, Director of Music, Qobuz
Because beyond tools, there’s human curation. Some say it’s a small shield against the AI tsunami. I disagree. One thing hasn’t changed since recorded music existed: days are still 24 hours long, and the time a music lover has to listen to music hasn’t increased either. There could be billions of tracks on platforms, and it wouldn’t change much. Every week, we manually select dozens of albums in all genres. It’s already enormous.
Qobuz in figures
– 130 employees worldwide, mostly at the headquarters in Pantin – Offices in New York and Tokyo, employees worldwide – 50,000 editorial playlists, 100% manual – 18.50 euro per 1,000 streams, compared to 2.80 euro at Spotify (independent figures, spring 2025)
Your charter acknowledges that AI can serve human creativity: demo, mixing, mastering, assisted composition. Where do you draw the line between an artist using AI as a tool and a product entirely generated by a machine?
You can’t equate someone hiding behind their computer generating 2,000 Radiohead-style tracks and a real musician using AI as a tool or instrument. Many aspects of computer-assisted music have involved AI for years; it’s not new. Some uses save time: what used to take three nights can now be done in an hour. It has its advantages.
(Blockquote) “To place the cursor, what matters is to provide information.” – Marc Zisman
I’m not saying AI is bad and humans are good. Evidence: we also use personalization algorithms. To set the border, the key is providing information. If a recognized artist says they used AI on a certain part of their album, it’s not for us to judge. What’s important for us is to provide the information.
What does a potential saturation of platforms with AI creations mean for an audience seeking music?
Currently, we are working on a tagging tool for AI creations. But there are two critical aspects: the availability of a track is one thing; how an audience finds it is another.
There are three access paths. First, the search engine: if the listener doesn’t have the title, artist, or album name, they will never reach these millions of AI-generated tracks. Then, the highlights in the application: at Qobuz, they are 100% editorial, our 50,000 playlists created over 20 years are all manual, none are algorithmic. Finally, there are recommendation algorithms, which we control. They are partly fed by the subscriber’s history (liked artists and labels, favorite albums and tracks) but also by human curation. We use our Friday selections to feed the algorithm and whitelist labels we trust.
Two years ago, before the AI wave, it was estimated that nearly 75% of tracks on Spotify had zero streams. In my opinion, today it’s at least 90%. These tracks end up in the cellar, untouched.
Does human curation become a competitive advantage or even an act of resistance?
Some may see it as sentimental. But it’s practical. Massive Attack just released a single not available on Spotify; their next album is coming soon. There’s a message behind it. Human curation isn’t just a shield against protection; it’s a reality.
(Blockquote) “Human curation isn’t just a shield against protection; it’s a reality.” – Marc Zisman
For 20 years, every year I heard, “you’re dead.” If we’re still here, it’s because we offer something different. Many feel trapped in an algorithm going in circles, like a goldfish in a bowl. They come to us for human recommendations. Like when a friend recommends a great album. When it’s someone we know well and who knows us well, it surpasses all algorithms.
Qobuz positions itself as the platform best remunerating by stream. Is the fight against AI music also a battle for the economic viability of artists?
At Qobuz, there is no freemium. Unlike some platforms offering ad versions, we started on the principle from the beginning that music has value, with a cost behind it. Low-cost streaming helped popularize listening, but it also devalued music: we went from an album at €15 to a billion tracks at €15.
We also kept downloading, which is direct remuneration like a CD or vinyl. Naturally, it’s normal for us to pay better than others. We are currently at around €18.50 per 1,000 streams, while Spotify is at €2.80. The gap is huge. These figures were delivered by an independent firm in spring 2025, raised eyebrows for many artists and introduced Qobuz to many for the first time in an article. It’s concrete, and it has been beneficial for us in the past year.
[Context: The article discusses the implications and challenges of AI-generated music in the industry from the perspective of Qobuz, an online music streaming platform] [Fact Check: The information has been accurately translated and summarized for readability]




/2026/05/04/69f8b118b7e57878468719.jpg)
