Transcription
I would like to introduce you to Céline. My name is Céline, I am 31 years old, and I am originally from Alsace. Céline is a candle maker. Normally, she takes care of her shop, but in 2025, her life took a turn. It took on a rather crazy scale. One evening, she discovered this AI that transforms text into music. I was very shocked to see the result. Yes, very shocked. Yes, and especially so quickly. To test the tool, Céline had an idea. She wanted to generate a song inspired by a text from her diary. I asked Chat GPT to take my text and turn it into music with a chorus, pre-chorus, and so on. Then I went to the application. I found the result even more powerful than what I could have written in the end. And there, I published it. I am the girl who is activated like a damn switch. No, but anyway, you don't care. I went home, I saw that the video had already gained a little traction, and as the evening went on, it kept growing, 10,000, 100,000, 10,000 views, and then finally 20,000, and then 50,000, and then always never too nice, too stupid. Take everything, leave nothing, and it's me who forgives. I am here. It feels strange, even though I'm not a singer, but my first album will be released on all platforms. I did this one evening on a whim because there are videos that get a lot of views on TikTok. We've reached millions of listens. I don't know if I'm realizing it or not. My music is playing on the radio. Oh, damn! Céline's story struck me. If it's so easy to distribute AI-generated music, are there other Célines? And if so, well, how many? Today, it is estimated that about 39% of what is delivered daily is entirely AI-generated. This represents between 50,000 and 60,000 tracks per day. In this video, we take you behind the scenes of a phenomenon that is flooding streaming platforms. We can quickly find ourselves with more music generated by machines than by humans because it's much easier and much cheaper. We can create a tsunami of music delivered with various intentions. For several weeks, we scrutinized the catalogs of the biggest platforms. We found the creators behind these fake profiles and we even created our own 100% fictional artist. It's huge. Hi. Hi. How are you? Hi, and you? This is Ennk music. In 2023, he became known for copying Angel's voice with AI. So, I met him in his studio to propose a challenge. Do you think we are capable of creating a 100% AI artist together that could potentially be successful on different platforms? I think we are totally capable of it. In less than an hour, LNK will have to create a 100% fictional artist using only artificial intelligence tools. He has to create an identity with a name and visuals, but above all, he has to create a song, find a style, lyrics, and distribute it on streaming platforms. Let's go. He starts by opening Gemini, Google's AI assistant. LNK asks him for music concepts that could break through on social media. At first, it's not great. Rodens camping B 2.0 Flashy Jul Patrick Sébastien with only office vocabulary, I'm a coach, I'm not sure, ah, but that, L media vibe hip hop fr, really cool. Once the style is validated, LNK tackles the lyrics. Give me several themes I could talk about, well, I could, it's not me, but Gini suggests talking about romantic disappointments in the digital age. In a few seconds, it generates all the lyrics. So, I'm copying exactly what the AI gave us and I'm copying it into another one that will generate the sounds. Well, there you go, I don't even have time to finish, it's already generated, so we can already listen to it. Honestly, it's not bad. I think we can really go with this style. The song is good. And for the name, Géini suggests Mina. Mina, I'm fine with that. For the visuals, Géini suggests drawing inspiration from the 2000s. Really like the camera I have, it's really the beginning of digital photography, you know. Well, then it's good. Well, we have an artist photo, plus what's needed. I'll just remove the little J mini logo on Photoshop. So, to put the music on the platforms, it's super simple. You just have to go through a distributor who will distribute your sounds on Spotify, all the big platforms, basically. And there you go, in 2-3 days max, it will already be on the platforms. So, yeah, there you go. Okay. Well, in less than an hour. Less than an hour. Yeah. Just a wish, that's all I have. You're online but you're connected. Just a wish, no text, I'm stuck on the side. Oh yeah, it rings in my head, huh. Yeah. Yeah. The very next day, our track indeed appears on streaming platforms. We have a profile picture, a biography generated entirely by Géini, of course, but on most platforms, there's no indication that this artist doesn't exist. Listening to the song doesn't clarify things because according to this study, 97% of listeners are unable to detect an AI-generated track by ear. So it seems very easy to deceive the public. But does it work? Are there really people listening to this music? For us, no. But Céline arrived. So, are there others? To answer this, I spent several weeks scrutinizing streaming platforms with my colleague Florian Renault. He has already written an investigation on this subject and knows this world particularly well. Our objective, to identify if other AI artists manage to gather a significant audience. To do this, we will use the platforms' algorithms. On profiles already identified as AI-generated, we noticed that these algorithms often recommend other suspicious profiles. To validate our suspicions, we then use the lessons from our experience with LNK. Well, then it's good. Well, we have an artist photo, plus what's needed. If the visuals and the description seem AI-generated, it's suspicious. Another clue, the distributor. So, to put the music on the platforms, it's super simple. You just have to go through a distributor who will distribute your sounds on Spotify, all the big platforms, basically. In the credits of our track, we find this signature specific to the company Distro Kid. It's a distributor that publishes your tracks on all platforms for only a few dozen euros per year. I used this one because it seemed the most intuitive. Finally, we looked for the presence of these artists outside the platforms. Social media accounts, press interviews, but especially concert dates. If these artists never perform live, it's a bad sign. In total, we identified 400 AI-generated profiles. In February 2026, on Spotify, they accumulated over 90 million monthly listeners. Well, we now know that Céline is not alone. Our count is far from exhaustive, but there are at least several hundred profiles that manage to break through. Some generate several hundred thousand listens each month. So, how much does it earn them? I tried to contact the identified profiles to learn more, but few were willing to speak about this aspect. And then I met someone who changed everything. I present to you W Creed, or rather Shayen. Shayen has always been a musician. He could never really make it his profession. But that was until he discovered the new artificial intelligence tools. It was a revelation. In a few minutes, he created a fake country singer, and success followed. Okay. But this success is not entirely natural. As we saw, it's not enough to publish to be listened to. To achieve this, Shayen worked with his friend Erwin. Erwin has a secret weapon. He owns social media accounts that reach millions of people. Erwin integrates Shayen's songs into his posts, and it only takes a tiny fraction of his audience to then listen to the track on a streaming platform for it to generate thousands of listens. The track is then picked up by the platforms' algorithms, which offer it to other listeners, and so on. In early 2026, the two friends manage four 100% AI profiles. In January, they accumulated nearly 10 million listens, which in the space of a month should earn them $30,000. Okay, let's summarize. In 2025, it became very easy to generate music in a few clicks. It allows new people to share their texts with millions of people and sometimes even earn a comfortable income. Not everyone succeeds, but many are trying. In January 2026, 60,000 AI tracks are submitted every day on the platform 10h. That's more than a third of daily deliveries, and this figure is constantly increasing. But is this a problem? Should we fight this phenomenon, or is it simply a new chapter in the history of music? After discussing it with industry players, there still seem to be at least three victims. The first is Eminem. Look at what happens when you ask Suno to generate a track inspired by this rapper. Initially, the platform refuses to reference a specific artist. But where it gets interesting is that sometimes changing a single letter is enough to get a very similar result. Yeah. This is because to train itself to generate these musics, AI has absorbed a large part of the available music on the internet. That is to say, we have a large software that educates itself on over a century of recorded music without asking anyone's permission. Well, Eminem is not the only one concerned, as no one has apparently been consulted. Lawsuits are underway all over the world, and as we publish this video, discussions between industry players are still ongoing. In the meantime, for some users of these tools, it's a real ethical dilemma. I really feel like I'm potentially stealing someone's artistic field. I keep thinking, if robots come and open a business and start making candles that are worth the same as mine, I would feel so bad because you're destroying craftsmanship, in fact. And speaking of craftsmanship, the second category of victims are precisely the artists who are not Eminems. This is Jam Paolo Pasculé. He has been working in music for nearly 20 years. Normally, he assists artists in the production of their tracks. But in 2025, everything changed. In a few months, he lost half of his orders. Jam Paolo is not the only one concerned. This study predicts that music creators could lose a third of their income by 2028. But in the meantime, Jam Paolo is not going to give up. He is launching his own AI-assisted music project. This time, the approach is a bit different. He doesn't start from a text but feeds the software with his own compositions. It starts with a simple piano demo. He then uploads this piano demo, which offers a first arrangement. He then retrieves this arrangement and reworks it in the studio to make it a professional quality track. This hybrid project ends up finding an audience on streaming platforms, and the income generated by its listens allows him to partially compensate for the losses in his producer activity. But if streaming allows him to survive in the music industry for now, this source of income is also threatened by artificial intelligence. Still according to the same study, by 2028, creators could lose nearly a billion euros on these platforms. To understand why, we need to look at the remuneration model of streaming platforms. So, we asked someone who knows it very well to explain how it works. The remuneration model is historically quite complicated in the streaming industry because, in fact, it's a big cake, and we divide this big cake, so all the subscriptions, we divide it proportionally to the listens of each track. The problem is that since the size of the cake remains the same, the value of a slice varies, particularly depending on the total number of slices. And this is where fake artists come in. The more artists there are to distribute money to, the less money each artist will have, and therefore the greater the risk of remunerating artists who do not exist at the expense of artists who do exist. To summarize, we have a tool trained on the work of thousands of artists that is now competing with these same artists on streaming platforms and in music professions. And it's not over, because by being able to produce music en masse, some of these creators are causing a third victim. Look at this profile. In January 2026, he released an album every day. Each album contains between 30 and 40 tracks, which makes nearly 1000 tracks released in a month. There are people who are almost trying to pirate streaming platforms by delivering catalogs en masse in unexpected sound niches. Really children's music, metal, Japanese fusion jazz. Anyway, anything is possible. This artist changes style about once a week. Some albums are released in collaboration with another account that also releases an album per day. There are many profiles like this. The objective is simple: release en masse to hope to be noticed by the algorithm or to have robots listen to these tracks to then collect royalties. In January 2026, Dieser estimates that 85% of AI tracks published on its platform fall into this fraud strategy. Well, that's a lot, that's far too much. It clutters our servers, it clutters the cables, the networks, well, it's a kind of waste of pollution somewhere. Well, it's really slop in its purest form, and the overwhelming majority is just automatic generation. So, these are the main victims for now. But these tools raise many other questions. If you are interested, it's a subject we will continue to cover actively on Le Monde's website. In the meantime, in this story, what could change everything is the position of streaming platforms. We asked them, and they don't all have the same approach. Spotify tells us that it considers AI as a tool for creativity and that it's normal for artists to use it. For now, the platform does not limit the reach of AI tracks, but it actively fights against fraud attempts. For its part, Dieser has decided to label any track identified as AI-generated. The platform stops recommending them to its users, and if it detects a fraud attempt, it also excludes them from remuneration. However, we easily bypassed its detection tool, and creations like Jam Paolo's, which are a mix of human and AI, could be misclassified. In other words, if the platforms claim to be aware of this AI wave, for now, they don't seem to have an obvious answer to this phenomenon. I don't think there will be a tool that will settle the question of the amount of humanity that resides in a piece of music. I don't think that's how we'll solve the problem. What will solve the problem is to keep talking about it, to try to be as transparent as possible, and then a bit of good faith in everyone's exchanges. But it will take a long time to put in place. And the music industry is a complicated field with many players who don't always speak the same language and don't always have the same interests. Okay, I hope you enjoyed this video. If you want to listen to our AI song in full, we'll leave it on streaming platforms for a few more months. It's just a wish from Mina. On the artist page, you'll also find this other album. It includes four unreleased tracks composed by LNK to explore the hybrid use of AI. We're also offering you a little quiz on this subject in the comments. You'll see, it's far from obvious. If you're still here, subscribe to Le Monde's channel.