#11: How Spotify names its genres
And why you should never let it tell you what you're listening to
Spotify Wrapped dropped today, the big domino in a data market that started when eugenicist Francis Galton, at an 1884 exhibition in London, discovered that people were willing to part with their personal information — and pay — as long as he shared his findings. It’s the 23andMe business model. It’s the Wrapped business model, too.
Though to be fair, Wrapped is fun, harmless, and asks nothing of its users. It’s even available to non-premium users of the app, if indeed any exist. Spotify is collecting all that data anyway. Whatever else you think about the company, it’s nice to be clued in once a year.
On the other hand, when you’re the biggest music streaming monolith in the universe, even your most innocuous initiatives can change the culture. Until today, I was unaware of my fifth favorite genre, pov: indie — just like listeners of Lizzo and Anderson .Paak in 2016 had no idea that they were big fans of the “escape room” genre. They’d never even heard of it before. I hadn’t. You hadn’t, either. Spotify invented the label.
Music writers balked and still balk at what they call micro-genres, made-up labels that sort Spotify’s catalog at the atomic level. Spotify’s conception of genre — which is really the peculiar opinion of one guy — now serves as the de facto arbiter of how audiences think about music, self-replicating through playlists, metadata, and, yes, Spotify Wrapped. Good, bad, or ugly, something in your end-of-year roundup will confuse you.
If you go looking for a rock song, you expect that it’ll include the electric guitar. If Hatsune Miku started doing screamo, you’d probably be surprised. Genre sets your expectation for what a song will sound like. It can also tell you the cultural context surrounding a song, like how punk challenges authority or hip hop originated in Black communities in New York. A working definition of genre is a set of shared musical conventions.
Within that definition, genre labels are perpetually open to debate. Two artists making similar sounding music may self-describe completely differently. An artist may be in conversation with their own genre, like Taylor Swift defecting to pop from country, or Björk doing whatever Björk does. A music historian once described to me the difference between country and western: “Country music is about what happens indoors. Western music is about what happens out of doors.” An exact science, genre is not.
One thing we do know is that artists never got sole ownership of their own genre labels. They were negotiated by publishers and journalists. Oftentimes — with jazz, funk, and rock ‘n’ roll, for example — genre labels just started as descriptors and caught on through repeated use. Some writers noticed musicians “jazzing up” marching band music. Funk is funky. Etc.
Whatever the particulars, the conventional idea about genre is that it describes something about music. Call it “descriptive genre.” This is how most people think about music.
Descriptive genre is fluid and organic, but also imprecise and impractical. That’s the problem Glenn McDonald ran into while attempting to troubleshoot a genre ID program for a startup called The Echo Nest. Growing out of an MIT Media Lab music identification project, one of the things The Echo Nest wanted to do was develop a program that could tell you the genre of a song automatically.
Except, even human music critics won’t agree on a song’s genre all of the time. Even if you narrowed down one definitive set of characteristics for a genre, there are some things computers just aren’t trained to understand. A computer may be really good at determining the tempo of a song, but really bad at figuring out if it has vocals or not.
To solve this problem, McDonald had to change the way he thought about genre. His argument was basically statistical: If you and I both love Radiohead, Simon & Garfunkel, and Bon Iver, and then you suddenly fall in love with The Velvet Underground — well, there’s a pretty good chance I’m going to like them, too. By aggregating the habits of a significant number of listeners, you could map out preferences based on “listening-based clusters.”
In other words, what if genre describes a group of listeners, not a group of songs or artists? That’s a pretty fringe way of thinking about genre. But it solves McDonald’s problem. An algorithm may not know how sad a song is, but it can track listener data precisely.
Glenn McDonald’s philosophy of music genre might have stayed fringe if Spotify didn’t acquire The Echo Nest in 2014. Suddenly, listener-based clusters powered a streaming giant. When your Wrapped tells you you’re in the top 0.1% of melancholia listeners — the Spotify genre including those artists I listed above — it’s saying that you’re at the center of a massive Venn diagram of people with music taste like yours. You can track it on Every Noise, a fun and distracting scatter plot of all these Venn diagrams.
Listener-based clusters explain how unheard-of genres like “escape room” make it into your Wrapped, but they only tell half the story — and the second half is maybe the most interesting.
Spotify’s “genre miner,” the program that hunts clusters, is prone to returning false positives. If you think about it, there are lots of reasons a cluster might appear, from record labels to viral TikTok playlists. To make sure nothing gets erroneously added to the catalog of genres, Glenn McDonald’s team does their research. They listen. Every single label in the Every Noise catalog — 6,000 classifications — was added by a person, at least as of the 2022 interview I’m pulling this information from.
One thing I haven’t said about these people: they love music. Some genre labels are no-brainers (e.g., these German house artists are all from the 90s, we’ll call it 90s German house), but others are named creatively based on subject, sound, and audience (e.g., escape room).
A lot of coverage you’re going to see in the coming weeks will focus on the explosion of micro-genres, their merits and demerits. My favorite so far comes from Matt Daniels and Michelle McGhee with The Pudding, partially because their visualizations are siiiiick, and partially because they make a point that these micro-genres are taxonomically valuable. More music is being published than ever before. We need more genres to name them, otherwise something like “indie pop” is so encompassing as to be meaningless.
But I argue that all of these arguments take Spotify on its own terms. Spotify itself has tried to explain its genres as a fact of nature. They say, “Maybe you were fascinated by a genre name you didn’t know before,” as though they discovered it when they invented it.
There’s a useful parallel in linguistics: descriptivism versus prescriptivism. Descriptivism attempts to describe the rules of language as they organically appear in a community. Prescriptivism attempts to prescribe rules of language, like a grammar school penalizing you for using a preposition to end a sentence with. Descriptive genre arises from a conversation between artists, audiences, and writers. A team of people naming genres in a closed room, even really smart and music-loving people? That’s prescriptive genre.
When we treat Spotify’s genres as something natural to the culture, we further bend our ideas about art to what Don Norman called “the peculiar demands of the machine.” When we let Spotify prescribe genre, we change ourselves to make things easier on computers.
The issue with prescriptive genre is that music doesn’t exist by itself in a community. For example, emo music is just one point in a constellation that includes fashion, philosophy, movies, and aesthetics. Prescriptive genre removes the community by pulling our listening habits out of a blind box, correlating our music taste without letting us meet each other, like learning language through a textbook rather than a conversation.
Some micro-genre labels are going to catch on. I might not start telling people my favorite genre is pov: indie, but other labels — chillwave, stomp and holler — I didn’t even know were invented before researching this piece. Some, like bedroom pop, are perfectly useful. Smart people make these labels!
The trick for defeating prescriptivism is to listen and resist correction. When Spotify tells you what genre is your favorite, smile and nod. Let your Wrapped be just one data point in a sea of identities, self-descriptions, think pieces, and adjectives. Remember, you have as much say as Spotify does. No matter how good its playlists are.
Thanks for reading!
My top artists: They Might Be Giants, The Mountain Goats, Vampire Weekend, Gorillaz, and Darren Korb — the composer for Supergiant Games who always sneaks his way onto my Wrapped. Better than two full albums of seas shanties.
DR
Two amazing lines in this. First, “Björk doing whatever Björk does” is exactly how to describe them. Second, ending “with” in the sentence about ending sentences with prepositions is a sin was phenomenal 🔥