There was a time when finding new music was partly a matter of luck.
You heard something on the radio and had to wait for the DJ to tell you what it was. A friend gave you a cassette and insisted that you listen to the third track. You walked into a record shop looking for one album and came out with something you’d never heard of because the person behind the counter thought you’d like it. Sometimes a song appeared in a film or television programme and stayed in your head long enough for you to start looking for it.
It could take some effort to find music. That was part of the experience.
Today, there is almost no effort involved at all.
Open Spotify, YouTube Music, Apple Music or another streaming service and there is already something waiting for you. A playlist made for your taste. An artist similar to the ones you’ve been listening to. A song that apparently fits the mood you’re in. Another album because you played something else three times yesterday.
We have more music available to us than anyone could possibly listen to in a lifetime. At the same time, we increasingly rely on machines to decide which tiny part of that enormous catalogue deserves our attention.
That doesn’t necessarily feel like a problem. Most of the time, it feels convenient.
But there is something worth thinking about in the fact that the way we discover music has changed so quietly.
Someone — or something — now has a hand in deciding what we hear next.
When someone else used to introduce you to music
Music discovery was never completely free of gatekeepers.
Radio stations decided what got played. Record labels decided which artists received promotion. Music magazines had editors and critics. Record stores had limited shelf space. Even your friends were gatekeepers, in their own small way, because you were more likely to listen to an album if somebody whose taste you trusted told you it was worth your time.
But there was usually a person involved somewhere.
That person might have been wrong. They might have recommended something you hated. They might have been obsessed with a band that made no sense to you. Occasionally, they introduced you to something that changed the way you listened to music.
That human element mattered because recommendation was never only about similarity.
A friend might know that you listen to Nick Cave and give you a record that sounds nothing like Nick Cave because, for some reason, they think you’ll understand it. A record-store employee might notice that you’re looking through jazz records and hand you something from an artist you’ve never heard of. A DJ might play a song simply because they loved it.
There was room for intuition.
Algorithms don’t have intuition in that sense. They have data.
And there is a lot of it.
Every time you save a song, skip one, replay another, follow an artist or spend several minutes listening to something you have never heard before, you give a streaming service another small piece of information about your taste.
Eventually, the service knows quite a lot.
The strange thing about personalised music
Personalisation has made streaming much easier to live with.
There are days when I don’t want to choose an album. I don’t want to search for something. I just want music on. The appeal of a personalised playlist is obvious in those moments.
Spotify has spent years building systems around this idea. Discover Weekly, Release Radar and other personalised features are designed to combine things that are already familiar with music the listener hasn’t heard before. Spotify’s own research describes recommendation as a balance between familiarity, similarity and discovery.
That balance is important.
Because if a service only gives us exactly what we already know, it becomes boring. But if it gives us music that has nothing to do with our taste, we’ll probably skip it.
The ideal recommendation sits somewhere in between.
It sounds unfamiliar enough to be interesting and familiar enough to make us want to stay.
The problem is that this can create a peculiar version of discovery. We may hear hundreds of new songs without actually moving very far from the musical territory we already occupy.
If you listen to indie rock, the algorithm can introduce you to twenty more indie bands. If you listen to 90s trip-hop, it can find artists influenced by trip-hop. If you spend a lot of time with old soul, it can keep finding more old soul.
Everything is new, but the world remains strangely familiar.
New isn’t always the same thing as different.
Now we’re beginning to talk to the algorithm
The next step is more interesting.
Instead of simply allowing the streaming service to observe what we do, we’re starting to tell it what we want.
In 2026, Spotify introduced Talk to Spotify, allowing users to make conversational requests about what they want to listen to. The company has also been developing Prompted Playlist, which allows listeners to describe a mood, situation or musical idea in ordinary language and receive a playlist based on that request.
That changes the relationship slightly.
You don’t have to search for the right genre anymore.
You can describe a feeling.
You could ask for something for a long drive. Something melancholic without being depressing. Music from artists you’ve never heard. Songs that sound like a particular period of your life.
It’s an oddly intimate thing to ask a machine.
Music has always been tied to things that are difficult to describe. Memories, places, people, moods that don’t really have names. We often know what we want to hear without knowing how to find it.
Natural-language AI is becoming surprisingly good at bridging that gap.
And that could be genuinely useful.
It could also make recommendation systems much more powerful than they used to be.
What happens when the machine gets very good at knowing you?
This is the part I find more interesting than the technology itself.
The better an algorithm understands your taste, the easier it becomes for it to give you something you’ll probably like.
But there is a difference between giving someone something they will like and giving them something they will remember.
The first is relatively easy to optimise.
The second is much harder.
Some of the music that stays with us wasn’t immediately comfortable. It didn’t necessarily sound like anything we were already listening to. Sometimes the appeal came from precisely the fact that it was strange.
That kind of discovery can be difficult for a recommendation system because the system is trying to predict your response before you’ve had it.
If something is completely outside your previous behaviour, there isn’t much data to suggest that you will enjoy it.
So the algorithm has a reason to stay relatively close to what it already knows.
That doesn’t mean recommendation systems inevitably make our taste narrower. Research suggests the picture is considerably more complicated. A large study of music recommendation found that algorithmic recommendations can increase short-term novelty while the music people discover may still remain relatively close to their existing preferences over longer periods.
That’s probably closer to the reality of how most of us experience streaming.
We discover things.
We just don’t necessarily wander very far.
And yet, algorithms can also find music we’d never find ourselves
There is another side to this.
A 2026 study examining generative-AI recommendations in music streaming found that recommendations could shift listening towards less prominent tracks rather than simply reinforcing the most popular music. The researchers also found evidence that newly recommended tracks were more likely to enter active listening and remain there for longer.
That’s worth paying attention to because it complicates the usual argument about algorithms.
It’s easy to imagine recommendation technology as a machine that continually pushes the biggest artists towards us.
Sometimes that may happen.
But recommendation can work in the opposite direction too.
A listener doesn’t have to know the name of an obscure artist. They don’t have to follow a particular scene. They don’t have to spend hours reading about new releases.
They can simply describe what they’re looking for and let the system search through a catalogue that would otherwise be impossible to navigate.
For a small artist, being discovered by the right listener may matter much more than being discovered by everyone.
In that sense, recommendation technology could potentially make the enormous size of modern music libraries less intimidating.
The problem of abundance doesn’t disappear.
It just gets a better search engine.
We have quietly turned algorithms into curators
This is probably the biggest change.
A recommendation system isn’t simply helping us find music. It is helping determine which music we notice.
That distinction matters.
There may be millions of songs available on a streaming service, but availability doesn’t mean visibility. A song can be there and still effectively be invisible to most listeners.
What appears on your screen matters.
What plays after the song you’ve just finished matters.
Which artist gets placed next to the artist you already love matters.
Even the decision to put something in front of you at 8pm rather than 10am can affect whether you listen to it.
None of this means that Spotify or another platform is secretly deciding what everyone should listen to. The reality is less dramatic than that.
But these systems have become part of the environment in which musical taste develops.
And once you see them that way, recommendation stops looking like a minor feature of a streaming app.
It becomes part of music culture.
Perhaps the real question isn’t whether algorithms are good or bad
I don’t think there’s much point in asking whether algorithms are ruining music discovery.
They aren’t.
They’ve also clearly improved it.
They can introduce us to artists we would probably never have encountered. They can make enormous catalogues manageable. They can help independent music reach listeners who might otherwise never come across it.
At the same time, they can make listening extremely comfortable.
You can spend an entire afternoon hearing songs that fit perfectly into the edges of your existing taste and never once encounter anything that makes you stop and wonder what you’re listening to.
That’s not necessarily a failure of the technology.
It’s partly a choice.
We ask for music that sounds like what we already like because we know that we’ll enjoy it.
The algorithm gives us exactly that.
And then we complain that nothing surprises us anymore.
There is a certain irony in that.
The good news is that we can still interfere.
We can search for an album ourselves. We can listen to a radio station from another country. We can follow someone’s recommendation without knowing whether we’ll like it. We can buy a record because the cover caught our attention. We can deliberately play something completely outside our usual habits.
We can even tell the algorithm to surprise us.
But I suspect the most interesting discoveries will always involve some degree of uncertainty.
A machine can become extremely good at predicting what you will probably like.
It cannot completely know what a song will mean to you after you’ve lived with it for ten years.
It doesn’t know that you’ll hear an old track one evening and suddenly remember somebody you haven’t thought about in years. It doesn’t know that an artist it recommended casually will become the soundtrack to a particular period of your life.
Those things happen after the recommendation.
And perhaps that’s where the human part of listening still begins.
We may increasingly let machines help us find the music.
What we do with it afterwards is still ours.