How to discover new music without relying on an algorithm
Automated recommendation is good at finding the similar and bad at finding the different
Automated recommendation works well, and it is important to start by admitting that. It succeeds when the goal is finding more of what you already like.
The problem is that it is optimised for exactly that. A system trained to predict what you will play to the end learns to avoid what bothers you. And a good share of the music that changes someone's mind is bothersome on first listen.
The result is a familiar feeling: an infinite catalogue and an ever narrower listening habit.
Swap similarity for a person
The most effective alternative is not a better algorithm, it is a person.
When you follow someone whose taste you respect, you do not receive what resembles what you already play. You receive what that person is playing, which includes their strange choices, their phases and their mistakes.
That is less comfortable, and that is the point. A recommendation that never displeases you is not taking you anywhere.
On ChartFM that is the central mechanism: weekly charts made by people, not by systems. And the suggestions of who to follow use genres in common, with no invented score, precisely because the intent is to find people with partial overlap, not identical ones.
Use blind listening when you can
The fastest way out of your own bubble is listening without knowing whose it is.
An artist name carries expectation. You have already decided whether you will like it before the first bar, and the decision comes from everything you associate with that name.
That is why Push, the site's discovery contest, hides the authorship of nominations until the end. It is not mystery for its own sake, it is the only way to make the track compete on its own.
If you do not use Push, the principle applies anywhere: ask someone for a recommendation with no context, or play a record before reading anything about it.
Give it a deadline, not a chance
The practical rule that changes results most: stop giving a chance and start giving a deadline.
A chance is thirty seconds and a decision. A deadline is three listens across a week and only then a decision.
Almost every song that became somebody's favourite was met with indifference on first listen. If your rule is to decide on first contact, you are filtering out precisely the music that needs time, which tends to be the music that stays.
Go back to full records now and then
A playlist is great for discovering a track and terrible for discovering an artist.
A loose track tells you that person made one good song. A whole record tells you how they think, what they repeat, what they are afraid of. It is the difference between knowing a sentence and knowing someone.
If you found a track you loved, the next step with the best return is not searching for similar songs. It is playing the record it came from.
A method that fits in one week
Putting it together, one possible routine:
- Follow three people whose taste you respect, not three playlists.
- Take one recommendation a week that you would not have chosen yourself.
- Listen three times before deciding.
- If you like it, go to the full record instead of hunting for similar tracks.
- Write it down somewhere, because listening that is not recorded disappears.
That last item is the dullest and the most decisive. A discovery you do not record goes back to being forgotten within two weeks. A weekly chart serves that purpose before it serves any ranking.
Frequently asked questions
- Do I have to abandon automated recommendation?
- No. It stays good at filling time and at finding the similar. What it does not do is move you, and for that you need another source.
- What if I do not know anyone with similar taste?
- Start from genres in common and adjust from there. Partial overlap works better than total overlap, because discovery lives in the difference.
- How many new songs per week is reasonable?
- Few. One or two you actually listened to are worth more than thirty that played in the background.
