Listening with the label on: what changes when you know a song was made with AI

Play a song to someone. Now play it to someone else, but first tell them: "this was made with artificial intelligence." Same song, same sound, same seconds. And yet they will not hear the same thing.

This is not just my hunch. For a few years now, music psychology has been measuring it in the lab, and large listener surveys back it up. As a lyricist who produces his music with AI, I find it fascinating. Not to complain, but to understand what happens in the head of the person listening to me.

The label plays before the song

Listening is not just hearing. When a song starts, we already carry an idea of who made it, where it comes from and what it wants to tell us. That idea colours everything else.

It happens with anything. Wine tastes different when you know it is expensive. A painting looks different when you are told it hangs in a museum. Music is the same: what we know beforehand enters through the ear along with the notes.

My song «Vas a escuchar un disco muy raro» has a warning about exactly this: «Si vienes con prejuicios, échate a un lado» ("If you come with prejudice, step aside"). I wrote it as a joke, but science largely backs the joke up.

What the studies say: the AI composer bias

One of the most cited papers was published in 2023 by a team at Missouri University of Science and Technology in the Journal of Experimental Psychology: Applied. They called it, literally, the "AI composer bias". Across several experiments, people rated the same piece lower when they believed an AI had composed it.

There is an interesting twist. With electronic music, the effect almost disappeared: the style already "sounded like a machine", so the label did not change much. With classical excerpts, which sound more human, the label did matter. In other words, the prejudice shows most when it clashes with what we expected.

In 2026, Sarah H. Wu and Kevin J. Holmes published a study in Cognitive Research: Principles and Implications that goes even deeper. They measured something lovely: whether people imagine a story while listening to instrumental music. With almost 400 participants in the United States, they found that the "AI" label reduced those imagined stories, and the ones that did appear were less absorbing. It did so even when the piece had actually been composed by a person. The authors suggest an explanation: if we believe there is nobody behind it, we stop looking for an intention, a message, someone trying to tell us something.

That idea strikes me as key. When we listen, we are looking for someone. If we are told there is nobody there, we stop looking.

But without a label, almost nobody can tell

Now the other side. In November 2025, the platform Deezer published a survey carried out with Ipsos of 9,000 people in eight countries. They played them three tracks, two fully AI-generated and one human-made, and asked them to guess. 97% did not get all three right.

The most telling part came afterwards. 52% said they felt uncomfortable not being able to tell them apart. 80% want fully AI-generated music to be clearly labelled. And even so, two out of three streaming users said they would listen to it at least once out of curiosity.

So the ear does not notice the difference, but the mind wants to know it. And once it knows, the listening changes.

Another study complicates things in an interesting way. A team from the Universitat Autònoma de Barcelona, working with the University of Ljubljana, published an experiment in PLOS One in 2025 with 88 people who watched videos set to either human-composed or AI-generated music. According to the published summary, the AI-generated music produced more arousal, measured through pupil dilation and through what participants reported, while the human music felt more familiar. There, participants did not know where each piece came from. Without a label, the body did not reject it.

What weighs is not the sound, but the idea that nobody is there

Put these pieces together and the picture is fairly clear. Sound alone already fools almost everyone. What changes the listening is the belief about who is behind it.

A preprint by researchers at TU Wien and the University of Innsbruck, posted on arXiv in December 2025, sharpens this further. They found that the "AI" label made pieces less liked than when they were presented as film soundtracks. But once they factored in each listener's prior attitude towards artificial intelligence, the label effect was no longer significant. The best predictor of how much people liked the music was what they already thought about AI before pressing play. And in their free comments, people talked about looking for the "human touch", the artist's personality, a story behind it.

It is a preliminary study, still awaiting peer review, and it should be read as such. But it fits with everything above: we do not listen with our ears alone. We listen with what we believe.

What listeners expect

The large surveys point the same way. In 2023, the international recording industry federation, IFPI, asked more than 43,000 people in 26 countries. 79% said that human creativity remains essential to making music. And 74% rejected using AI to clone or impersonate artists without permission.

In the Deezer survey, 70% believe fully AI-generated music threatens musicians' livelihoods. It is a worry I understand and share. I have written about it in other pieces, such as What is mine in a song made with AI.

Notice the word "fully". Much of the rejection is aimed at songs with no person in them at all, at catalogues churned out in bulk, at voices cloned without permission. That is not the same as a song where someone has written every word and chosen every version.

Why human lyrics change the listening

This is where I come in. I write all my lyrics. I produce the music with Suno, an AI tool that I direct, choosing, discarding and arranging what it offers me. I always say so up front, because I believe the listener has a right to know. I explained why in Telling the listener.

If the studies are right, what the "AI" label switches off is the search for intention. The feeling that someone is telling us something. And in a song with lyrics, that has a very concrete answer: the intention is written down. It lives in the words.

When someone listens to «Discutiendo con la IA» and reaches «¿Por qué decías que como lo mío era IA no valía?» ("Why did you say that because mine was AI it didn't count?"), they are not hearing a statistical pattern. They are hearing a question a person asks himself and asks whoever judges him. The sound may come from a tool. The question does not.

That is why I think human lyrics change the listening in three ways:

• They put someone back on the other side. There is an author with a name, a life and something to say. The search for intention finds something to hold on to.
• They give the sound context. The same melody is heard differently when you know what it is about and why it was written.
• They make trust possible. If you know what the machine does and what the person does, you can decide how to listen, instead of feeling deceived later.

I do not want to sell this as a magic formula. Some people, knowing all of this, will still not want to hear anything made with AI. That is fair. What matters to me is that the decision is made by listening, not just by reading a label.

An experiment for whoever is reading

Here is a suggestion. Next time you hear a song and know it was made with AI, notice the exact moment that information shows up in your head. Before the first note? Halfway through the chorus? When the lyric says something that hits home?

Then ask yourself what you were really listening to: the sound, or the idea you had of it.

The studies say almost nobody can tell the origin by ear. What we can tell, when we pay attention, is whether someone is talking to us. That is what I try to put into every lyric. The rest, as the record says, depends on whether you come with prejudice or with a wish to listen.