spotify’s algorithm is broken and our language is going to change

we're so cooked

Over the past few weeks, I’ve been noticing a bizarre pattern in my Spotify recommendations: no matter what playlist I finish listening to, the next song in my autoplay is always one of four tracks: Espresso by Sabrina Carpenter, Not Like Us by Kendrick Lamar, Million Dollar Baby by Tommy Richman, or Lunch by Billie Eilish. Doesn’t matter whether I’m listening to reggae or classic rock or 90s rnb: one of the four will always be queued up.

Don’t get me wrong, I like the songs, but I don’t want to hear them in every scenario. I’ve only listened to them by my own volition a few times each, and yet they pervasively wait in my recommended songs like they’re Mr. Brightside at a millennial karaoke night. After noticing two friends and a TikTok video independently mention the same phenomenon, I decided it was widespread enough to write about.

It’s not random noise, of course: Spotify’s algorithm works by pushing recommendations likely to keep you on the app. These recommendations are based on the taste profiles of listeners similar to you. If people like me listened to a song, the reasoning goes, I would like that song too. And the logic isn’t bad, but at a certain point the song gets pushed simply because it’s being pushed. Now that I’m listening to Espresso, even though I don’t particularly care for it, it’s going to be recommended to more people, and the cycle continues.¹

You might be wondering why the etymology guy is talking about music recommendations, but at this point language change has become inextricable from algorithms. For one, people are more likely to use words they hear in media—you may have noticed how Espresso has led to a discernible rise in people using the grammatical construction that’s that me, from the lyric that’s that me espresso:

case in point

But this isn’t just about Spotify. Every major social media app does something similar. TikTok creators are well aware that the platform rewards trending audios, and actively incorporate them into their videos to go viral. The example I like to use is the Rizzler song that went viral last year:

Sticking out your gyat for the rizzler

You’re so skibidi

You’re so fanum tax

I just wanna be your sigma

Freaking come here

Give me your ohio

After this audio came out, everyone's For You Page was dominated by a melange of videos using, remixing, or parodying the original song. Why? Because creators knew that the algorithm was pushing the trendy song. As a result, it became even trendier, and the words in those lyrics became mainstream. Now we have millions of children walking around saying the words skibidi, gyat, and sigma, even though those terms were virtually unknown a year ago.

I see Spotify’s broken recommendation system as a harbinger of what is to come (and is partially already here) for both social media and linguistics: a reality where our culture and words are predetermined by what is trending. As soon as platforms have a monopsony on their consumers, they’re increasingly incentivized to prioritize trends because, on average, it helps them retain your attention so you can be advertised to.² That’s that shareholder value.

---

Life announcements

• I’ve officially signed with Knopf to write a book! It’ll be published around fourteen months from now so stay tuned lol

• I just wrote a new article for the Washington Post on the skull emoji - read it here!

• Happy to announce that I’m a recipient of the 2024 LingComm grants for linguistics communication - stay tuned for long-form YouTube videos this fall

1. New Yorker writer Kyle Chayka has an excellent discussion of how this leads to a “flattening” of global culture in his book Filterworld—highly recommend.

2. Canadian author Cory Doctorow calls this phenomenon enshittification, and he kind of ate with that.