AI Music Indistinguishable from Human Creations, Say Listeners

97% of listeners can't tell AI music from human.

According to a Deezer–Ipsos study reported by Reuters, 97% of listeners cannot differentiate between songs created by artificial intelligence and those composed by humans.

The survey, conducted among 9,000 individuals from eight countries including the US, UK, and France, revealed that 73% of respondents desire transparency regarding the use of AI in music creation.

Additionally, 71% expressed surprise at their inability to distinguish between human and synthetic tracks.

Music streaming service Deezer, with 9.7 million paying users, reported over 50,000 AI-generated tracks uploaded daily—about a third of the total. In April, this figure was 18%.

The platform has implemented labeling and excluded AI tracks from editorial playlists and algorithmic recommendations.

“We firmly believe that creativity is born from humans and should be protected,” said Deezer CEO Alexis Lantier.

He noted the challenges in implementing a differentiated payment structure for AI-created music. Deezer has also begun excluding fake streams from royalty calculations.

Lawsuit Against OpenAI

In early 2025, a scandal emerged involving the group The Velvet Sundown, which attracted millions of listeners on Spotify. The synthetic origin of its music was uncovered.

Another issue is the training of neural networks on songs without obtaining the necessary licenses.

The Munich Regional Court ruled that the startup OpenAI violated German copyright laws by training its models on the lyrics of Herbert Grönemeyer and other artists, including hits like Maenner and Bochum.

The case was initiated by the German copyright protection society GEMA, which includes composers, poets, and publishers.

Presiding Judge Elke Schwager ordered OpenAI to compensate for the use of copyrighted materials. The amount was not disclosed.

The developer of ChatGPT claims that language models do not store or copy specific training data. They reflect what they have learned from all the information received.

OpenAI argues that since the result is generated after a prompt is entered, users are responsible for it.

However, the judge believes that memorization in LLM and subsequent reproduction of song lyrics constitutes a copyright violation.

Precedent

The court’s verdict may set a precedent in Europe regarding the use of copyrighted materials for training AI models.

“The internet is not a self-service store but a collection of human creative achievements. These are not free templates. Today we have set a precedent that protects and clarifies authors’ rights: even operators of AI tools like ChatGPT must comply with copyright law,” stated GEMA CEO Tobias Holzmüller.

An OpenAI representative noted that the company disagrees with the decision and will consider further steps.

“The ruling concerns a limited set of song lyrics and does not affect the millions of people, companies, and developers in Germany who use our technology daily,” he said.

In September, Suno introduced the “world’s first” generative DAW—Suno Studio, which “radically reimagines the music creation process.”