Episode Transcript
This is a computer-generated transcript. While our team has reviewed it, there may be errors.
Morgan Sung: Last year, Sabrina Carpenter released Espresso.
Sabrina Carpenter: Now he’s thinkin’ ’bout me every night, oh/ Is it that sweet? I guess so / Say you can’t sleep, baby, I know / That’s that me espresso
Morgan Sung: I think I first heard the song on TikTok, or maybe it appeared on one of my Spotify daily mixes. Either way, I thought it was fun, so I put it on my summer playlist. And then it was like there was no escaping espresso. For months, every time I finished listening to a playlist or album, it seemed like the next song that Spotify auto-played was almost always espresso. I remember listening to Charli XCX’s Brat, and then…
Sabrina Carpenter: Say you can’t sleep, baby, I know / That’s that me espresso
Morgan Sung: Brat is like super electro pop club music that sounds nothing like espresso. But I guess it made sense in my head because they were both trending last summer. But it kept happening with totally different genres. And music that had been out for years.
Here’s the one that really got me. I was listening to my playlist of instrumental movie scores, right? It was on shuffle and the last song was the one from the cornfield scene in Interstellar. Imagine you’re me, listening to this devastatingly beautiful song. The music swells into these ethereal arpeggios, building on each other until it hits a crescendo. And then it comes to an abrupt stop. And while you’re still lingering on the melody of that song, you hear the opening bars of another. And at that point, I knew exactly what was coming.
Sabrina Carpenter: Say you can’t sleep, baby, I know / That’s that me espresso
Morgan Sung: I wasn’t the only one being haunted by espresso. This was so widespread that there was a whole internet conspiracy theory claiming that Sabrina Carpenter’s team had struck a deal with Spotify to force its algorithm to incessantly autoplay the song in order to boost streaming numbers. Deals between streaming platforms and musicians do exist, but it’s not as simple as that. There’s no single algorithm controlling all of Spotify. What’s more likely is Spotify knew I had listened to the song before, so it just kept recommending it to me.
Liz Pelly: I always try to differentiate, too, the difference between streaming services show you music that you like versus streaming services, show you a music that you seem statistically likely to not hit skip on because basically what streaming services have is a pool of data on you.
Morgan Sung: Liz Pelly is a music critic and journalist who’s been covering Spotify for years. She’s also the author of Mood Machine: The Rise of Spotify and The Cost of the Perfect Playlist. It’s all about how Spotify’s playlists and recommendation algorithms have changed the way we listen to music.
Liz Pelly: We’ve seen a lot of these headlines like, “How Spotify plans to get into your head and tell you what music you like before you even know you like it yourself.” And I think it’s really important to sort of push back on those types of narratives and continue reiterating that that’s not possible. They can’t do that, they can’t that. But what they can do is look at everything that you’ve listened to in the past and make predictions of tracks that you might either be willing to hit play on or that you won’t hit skip on.
Morgan Sung: Spotify gives users access to a seemingly endless library of music, but it’s really become known for being a music discovery platform. For years, its editorial playlists and personalized features introduced hundreds of millions of users to new artists and genres. But now…
Liz Pelly: Something that I commonly hear from people who use Spotify is that, you know, the recommendations used to introduce them to lots of new music, but over time, the recommendations have gotten a lot more boring. And now all they ever hear is stuff that they’ve already listened to over and over again.
Morgan Sung: This TikTok creator named Anthony Sistilli summed it up in a video complaining about a feature called “song radio.”
Anthony Sistilli: Because if I started a radio about a song, it meant I wanted to hear more songs similar to that one that I may or may not know. But now when you do that, it just gives you a bunch of songs that you’ve already listened to, you already have saved to your library, and it’s just insane. It’s so hard to discover music on it now.
Morgan Sung: So what’s behind Spotify’s shift from a music discovery engine to a platform for songs you already know? Has personalization gone too far? And will any of this explain why I couldn’t escape espresso?
This is Close All Tabs. I’m Morgan Sung, tech journalist, and your chronically online friend, here to open as many browser tabs as it takes to help you understand how the digital world affects our real lives. Let’s get into it.
So did playlists ruin music? Before we get into all of that, let’s start at the beginning. And that means we open a new tab.
The birth of the Spotify playlist.
When Spotify launched in Sweden in 2008, it looked a lot different. When it first launched, the product was really more like a search bar. It was a sparse white page with a tagline that said, “A world of music, instant, simple, and free.” There was a search function and an option to create “play lists” — two words. They were essentially digital mixtapes.
Liz Pelly: You would have to know what you were looking for when you went to the platform, whether it be an artist or an album. And when it came to playlists, it was a lot more dependent on playlists made by other users. There was this sprawling landscape of playlist influencers that were sort of dictating things in terms of what types of playlists became popular.
Morgan Sung: Originally, the company’s CEO, Daniel Ek, wanted the platform to be the “Google of music.” He was pretty resistant to the idea of Spotify ever becoming a curated service. When it launched in the United States in 2011, it had to compete with iTunes, so Spotify tried to stand out by being a social media platform and then a marketplace for music apps. Basically, the country marketed itself as a library for third-party apps to make playlists. And while that didn’t really work out, it turned out that playlists were super popular. In 2012, the company conducted a research study on its user base.
Liz Pelly: What they learned through this research is that while their early adopters might have been coming looking for that type of experience, there was increasingly a type of user that was looking for an experience more like that of Pandora. People were looking for specific lean back or passive listening playlists that they could just hit play on to soundtrack a moment in their day.
Morgan Sung: In this research, they discovered something that they called the “lean back listener.” So what can you describe this list- this kind of listener?
Liz Pelly: It’s this idea of this listener who isn’t particularly concerned with what artist or album they’re seeing out, but is more content to just pick a mood or a vibe or a playlist category and hit play and have a feed of music play in the background.
Morgan Sung: After that, Spotify rebranded from a world of music to music for every moment.
Liz Pelly: The homepage was reimagined with more of a focus on these sort of moody images of people in their everyday lives, like sun flares reflecting off of the window of a car while driving, people on vacation together at the beach, you know, like selling people on the idea of Spotify being something that would accompany them in these different moods and moments in their lives.
Morgan Sung: To accomplish this, in 2013, Spotify acquired a company called Tunigo, which made playlists for different vibes. Tunigo’s editorial staff became the first in-house playlist editors at Spotify. They launched playlists for every genre, like Rap Caviar and Hot Country, but also invested in playlists for “functional music.” That’s a term for music for different moods or activities. Think of how workout mixes always feature upbeat, and 10 songs to get you pumped! Or how cafes put on acoustic, mellow songs for ambiance. It’s less about the music itself and more about facilitating a mood. And it goes hand in hand with lean back listening.
So instead of putting on a specific album to unwind after work, a lean back listener might just pick from Spotify’s array of chill playlists. There are lots of categories of functional music playlists like Party or Focus or Romance. And playlists update every few days, so there’s always something new. Leanback listening and functional music weren’t invented by streaming. They’ve been around for decades.
Liz Pelly: In some ways, the history of commercial radio could also be looked at as a type of history of lean back listening as well. The story of streaming is as much about what’s changed as it is about what stayed the same.
Morgan Sung: Before the days of Spotify’s playlist empire, a lot of leanback listeners turned on the radio. Instead of the Top Hits playlist, they listened to the Top 40s station. And instead of the Mellow Morning playlist, maybe they put on the Easy Listening station. But the way those songs are chosen has changed.
Liz Pelly: The way in which listening happens today in this lean back environment also involves, I’d say a lot more data collection and user surveillance than perhaps lean back listening of generations past did, which in some ways I think allows the services to in an even more invasive way, really optimize for this type of listening experience.
Morgan Sung: Liz makes the argument that Spotify’s real value isn’t its library of music. It’s the data that it’s collected from hundreds of millions of users for over a decade. These listening habits have informed the way that Spotify categorizes and recommends music. So what happens to the culture of music when songs are reduced to data? That’s a new tab.
Streaming optimized music.
I still think handpicking songs for someone else to listen to is one of the most romantic things you can do.
John Cusack: I’ve started to make a tape in my head for Laura. Full of stuff she’d like, full of stuff that would make her happy.
Morgan Sung: Back in 2013, Spotify marketed playlists by invoking that same sense of intimacy. But today, Spotify’s in-house playlist creation doesn’t really work like that. It’s all about optimization. They analyze user behavior, music trends, and other kinds of data to put together a playlist that’s likely to get a ton of streaming time. And by 2016, these playlists were immensely influential. A lot of artists got their big break because their songs ended up on an official Spotify playlist.
Liz Pelly: Some of the earliest narratives that streaming services and Spotify in particular were pitching to independent musicians were, yeah, this idea that Spotify was gonna level the playing field and that now the power of gatekeepers like commercial radio DJs and major labels would be diminished.
Morgan Sung: So up-and-coming artists could submit their tracks. And if a playlist editor liked it, they would add it to what was known as a “feeder playlist” — usually a smaller and niche playlist that wasn’t as popular. If the track had a low skip rate, it might move up to a bigger playlist with more followers.
Editors also looked at data from user-generated playlists. If a song had a certain number of streams or saves, then editors would consider adding it to the flagship playlists like Morning Commute and Warehouse Party. But data doesn’t always paint a full picture. Liz pointed out that some songs had a lower skip rate because they were just inoffensive enough to listen to in the background.
Liz Pelly: So some musicians would notice that even though they had these whole varied catalogs, it was only music that was playlist friendly, background music friendly, maybe kind of like more straightforward, softer on the edges that would do well on streaming because it was music that was more amenable to playlist curation.
Morgan Sung: And this is exactly what happened to lo-fi beats. You might have heard of this YouTube channel called Lo-Fi Hip-Hop Radio, Beats to Relax / Study To. It’s a 24-7 stream that features this anime girl wearing headphones, who’s now known as Lo-Fi Girl. She’s usually writing or reading at her desk, while her cat lounges on a windowsill in the background. The view through this window shows this idyllic cityscape. Sometimes it’s sunset, sometimes it’s raining. Meanwhile, there’s a continuous rotation of calming, overwhelmingly cozy instrumental music.
Lo-Fi Girl is wildly popular. She’s inspired countless spinoff channels from jazzier lo-fi beats set at a busy cafe to lo-fied versions of movie and game soundtracks. Producing the genre is a whole cottage industry now. And Spotify has launched a bunch of playlists to capitalize on the trend. There’s the Lo-Fi Beats playlist, one called Lo-Fi Covers, one called Chill Hits, Late Night Beats, you get the picture. And they all kind of sound the same.
But today’s version of the genre sounds very different from how it did when it first emerged in the early 2010s. Back then, DIY musicians, who were inspired by hip hop producers like J Dilla and Madlib, shared their beats with each other on SoundCloud and internet forums. It wasn’t called lo-fi because it was chill music. It was literally low fidelity. It sounded a little unpolished and nostalgic.
Liz Pelly: It was less about making background music to study to, and more about the art of beat making, sample flipping.
Morgan Sung: Early YouTube streamers like Lo-Fi Girl acted as curators, handpicking a constant rotation of lo-fi mixes from these independent beatmakers. But when Spotify jumped in, the genre started to change.
Liz Pelly: It kind of, I think is a good example of something that has happened like repeatedly across other genres and sounds in the streaming era, which is you may have like a playlist that becomes the main way in which people are coming to experience the certain type of music. So the fandom becomes a lot more about the playlists than it does necessarily about the different artists that are on the playlist. And streaming users come to expect a specific sound and a specific idea. So then there’s incentive for musicians and producers to sort of like make the same things over and over again.
Morgan Sung: This strategy could be lucrative for musicians, but because people were often passively listening to ready-made, data-optimized playlists, the artists themselves remained largely unknown.
Liz Pelly: It is in some ways kind of also on the listeners to take the initiative to learn about the musicians that they’re listening to if this is a type of music that you are interested in, but I also think that the interface doesn’t really in any way prioritize that. You know, lo-fi hip-hop beats on Spotify are one of the genres that has been most impacted by the emergence of what I talk about as ghost artists.
Morgan Sung: Okay wait what are ghost artists? We’ll get into that after this break.
Okay, we’re back. So Liz was just telling us about how lo-fi beats is one of the genres most impacted by “ghost artists.” Let’s open a new tab.
Ghost artists in the machine.
By ghost artists, Liz is referring to the artists who make up the majority of Spotify’s lean back functional music playlists. The ones made for listening to in the background, like lo-fi beats or deep focus. For a while, people have been accusing these playlists of being stocked with “fake artists.”
Liz Pelly: For me, I think that at the very beginning, I thought that maybe these were just kind of DIY hustlers trying to game the system or maybe like teenagers in their bedrooms trying to like juice Spotify royalties or something. And it didn’t strike me as like super concerning at first.
Morgan Sung: But Liz found that it actually went way deeper than that. These playlists rack up millions of streams a day, even if people are only half-listening to them. By 2017, the company figured out a way that it could skimp on paying royalties. That’s when Spotify quietly rolled out what they called Perfect Fit Content, or PFC.
Liz Pelly: I was able to pull the veil back a little bit and came to learn that, you know, these weren’t individual DIY hustlers trying to game the system. This was an actual effort from within Spotify.
Morgan Sung: Essentially, Spotify made deals with production companies, which were already commissioning musicians to write and produce low-budget stock music. Think of the kind of background music that goes into commercials, documentaries, YouTube videos. Even this show, when my colleague Chris doesn’t have time to compose original music for each episode. Except these were commissioned for specific playlists like lo-fi beats or ambient piano. And the thing is, these ghost artists were still marketed as if they were real individuals, complete with fake names and bios.
Liz Pelly: It’s not even that it’s encouraging a relationship with the playlist instead of the artists. There’s actually in this arrangement, there’s no opportunity for a relationship with the artists when the artists don’t actually exist.
Morgan Sung: Spotify paid lower royalties for PFCs, and in exchange, these tracks were given priority on Spotify’s functional music playlists.
Liz Pelly: It was clear that there was special privilege being given to these types of tracks and there were certainly internal conversations where senior executives or senior music programming staff was drawing attention to these tracks and encouraging playlist editors to add them.
Morgan Sung: Liz says that by 2023, there were hundreds of playlists that were over 90% PFC tracks. Spotify got to cut costs and the stock music production companies profited off the streams. But ghost artists, who are real musicians doing real labor, got the short end of the stick. They were making anonymous stock music. So even if they got paid upfront, they rarely got a cut of royalties. And even if their tracks went super viral, they didn’t get any credit.
Liz Pelly: Most of the people I spoke with said things along the lines of, “I just make music, submit it, I get paid, and I don’t know what happens after that.” They didn’t know really anything about the broader arrangement with the streaming services, what was guaranteed or not guaranteed, or even like what that relationship looked like.
Morgan Sung: And Liz found that most playlist editors were not thrilled about this setup either. Even if it involved data, Spotify’s editorial playlist curation was something of an art form that required a deep understanding of music and culture, which editors took a lot of pride in.
So, Spotify eventually hired an entirely new team of editors to handle PFCs, who were charged with prioritizing metrics above anything else. They were less focused on taste-making, and more on analyzing user behavior and sorting data. And that practice has become the norm for Spotify’s playlist editors. Spotify started to focus more on personalized playlists around the time they launched the PFC program. The editorial playlists aren’t nearly as influential as they used to be, because now it’s all about the algorithm.
Liz Pelly: There’s one person who talked to me who said, as someone working in an independent record label, they’d basically given up on the idea that their music would ever be considered for editorial playlist placement because everything had shifted to algorithmic recommendation and personalization products. And even former employees I spoke with, there was a sense that algorithmic personalized recommendations had sort of taken over to some extent or kind of won out.
Morgan Sung: Which brings us back again to…
Sabrina Carpenter: Say you can’t sleep, baby, I know / That’s that me espresso
Morgan Sung: Okay, new tab.
Algorithm fatigue.
Liz says Spotify has always been driven by data.
Liz Pelly: This is curation that’s really based on streaming metrics and less so on the willingness or openness to take risks. Personalized recommendation doesn’t just exist to offer the user a satisfying user experience, but also because they want to retain you as a subscriber. And one of the ways in which they can retain you is by simply, you know, showing you the music that they think you’re the most likely to stream.
Morgan Sung: With that in mind, the company started collecting listener data way before they even knew what to do with it.
Liz Pelly: In my research on Spotify, something that kept coming up as far back as like 2013, was this aspiration to create essentially like a button that someone could play when they opened the app and get the perfect recommendation at the perfect time.
Morgan Sung: In 2014, the company bought a music data startup called Echo Nest, which had a database of 30 million songs and their defining characteristics. A year later, Spotify used all that data, combined with their own, to launch Discover Weekly. This has been one of Spotify’s defining features. 30 songs every Monday tailored to each individual user’s music taste. This was the start of Spotify pivot to hyper-personalized, algorithmically recommended music.
Liz Pelly: This idea of self-driving music, that someone would just be able to open the app and based on all this information that the app had on the user, the playlist recommendations would just guide themselves. A former employee mentioned to me that the whole goal of the playlisting ecosystem was to reduce the cognitive work that people had to do when they opened the app.
Morgan Sung: After that, editorial playlists took a backseat. There are the big ones, Rap Caviar, Today’s Top Hits, and the EDM playlist, Mint. There are a few others that Spotify still curates and invests in, and says that they still have human editors working on. But personalized recommendations were taking over. The job of a Spotify playlist editor began changing. Spotify calls it algo-torial. Playlist editors focus on creating large lists of songs. But an algorithm picks the ones shown to each individual listener. And all of it is optimized to keep you from turning it off.
Liz Pelly: This idea of like, “Oh, we want to reduce like the thinking that people have to do when they open the app,” I find troubling for all sorts of ways, also, especially as a music critic. But I, I think that it’s really interesting thinking about that as context for a couple of recommendation products that they did roll out the past few years, which are Daylist and AI DJ.
Morgan Sung: These are totally algorithm-controlled playlists. Daylist is like Discover Weekly on steroids. It’s a playlist of 50 songs that’s updated three to five times a day, every day. It’s not just based on your taste, but also on the genres of music that you tend to listen to throughout the day.
Liz Pelly: It’s providing you the perfect soundtrack tailored just you and your unique way of listening to music because we have all this data on you and we know all of these different niches you’re into.
Morgan Sung: AI DJ also serves users with algorithmically recommended songs, but with interjections from an AI-generated voice, who acts kind of like a personal radio host.
AI DJ: Hey Max, what’s going on? I’m X and from this moment on I’m gonna be your own personal AI DJ on Spotify. Let’s go!
Liz Pelly: They are both different manifestations of that idea that, you know, there’s just this one place that you go on the app and you just hit play and it shows you the perfect recommendation at the perfect time.
Morgan Sung: And then there’s Smart Shuffle, which auto plays music and is probably responsible for my Espresso hell. And like the others, it’s also hyper personalized algorithmically recommended songs based on each user’s listening history. But Liz points out that you don’t have to give into the algorithm.
Liz Pelly: The idea of like curating your own queue as a sort of like antidote to the feed. If you maintain your own queue, it can kind of like help in those moments to have a different way of listening to music.
Morgan Sung: Still, Liz says she’s concerned about the convenience and widespread adoption of Spotify’s recommendation system. Thanks to these fine-tuned algorithms, people have started to expect recommendations to come to them. And the practice of discovering and curating music for ourselves is withering away.
Liz Pelly: I think a lot of people rely on personalized recommendations because they don’t know what to listen to or because they open the app and it’s just the first thing they see. And they might even be like a curious listener, someone who cares about music, but they open it, they see Daily Mix One, they say, “Oh, I like some of these artists, like, sure, I’ll just hit play on this.” It’s like a slippery slope when your daily listening habits come to sort of revolve around these algorithmic personalized recommendation products that are so tightly controlled by streaming platforms.
Morgan Sung: I made my Spotify account back in 2014. And since then, every decision I’ve made on Spotify, consciously or unconsciously, has been tracked and used for the platform’s various algorithms. The amount of time I’ve listened to certain songs, the music I’ve added to my own playlists, the songs I’ve skipped, and even the mixes I’ve put on as background music. All of this somehow contributed to me getting bombarded with Espresso last year.
I think we’ve heard enough of that. I’ll admit it, it’s so convenient to just turn on my Daylist. Sometimes I’m in the car, or about to get on the treadmill, and I just don’t want to put in the mental work of looking for something to listen to. Here’s the other thing about Daylists that make them tempting to pick. They’re just so hyper specific.
Xorje Olivares: My Daylist playlist is called Supergroup Ripped Jeans Tuesday Afternoon.
Chris Egusa: Indie Chill Hipster Evening.
Olivia Cueva: Clavé, Barbershop, Friday Morning.
Morgan Sung: For the record, as I write this narration, mine is “Ethereal Alien Club Classics Afternoon,” with genres like Brat Summer, Indie Sleeze, and Recession Pop? Somehow, Ethereal Alien Club Classics includes Frank Ocean, Midski, and of course, Charlie XCX. But how did Spotify come up with these niche micro genres? How do artists feel about getting lumped into them? And how does this affect the way that we discover music? That’s a new tab, but we’ll have to save that for next week. So for now, we’re actually leaving this last tab open. We can close the rest though.
Close All Tabs is a production of KQED Studios and is reported and hosted by me, Morgan Sung. Our Producer is Maya Cueva. Chris Egusa is our Senior Editor. Jen Chien is KQED’s Director of Podcasts and helps edit the show. Sound design by Chris Egusa and Brendan Willard. Original music by Chris Egusa. Additional music by APM. Mixing and mastering by Brendan Willard.
Audience engagement support from Maha Sanad and Alana Walker. Katie Sprenger is our Podcast Operations Manager, and Holly Kernan is our Chief Content Officer. Support for this program comes from Birong Hu and supporters of the KQED Studios Fund. Some members of the KQED podcast team are represented by the Screen Actors Guild, American Federation of Television and Radio Artists, San Francisco, Northern California Local.
Keyboard sounds were recorded on my purple and pink Dustsilver K-84 wired mechanical keyboard with Gateron red switches.
If you have feedback or a topic you think we should cover, hit us up at [email protected], follow us on Instagram @CloseAllTabsPod, or drop it on Discord. We’re in the Close All Tabs channel at Discord.gg/KQED. And if you’re enjoying the show, give us a rating on Apple Podcasts or whatever platform you use.
Thanks for listening.
