Your preferred video-streaming service depends largely on what you want to watch—but what you want to watch won’t always stay in view. Series and movies hop from platform to platform as rights lapse or get awarded to higher bidders; even streamers’ homegrown offerings can disappear to free up bandwidth or save on licensing fees. By comparison, music-streaming services are bastions of stability—and because music streamers generally offer the same catalog, barring the occasional protest, they have to differentiate themselves in other ways. Tidal and Qobuz trumpet their hi-fi offerings; YouTube Music infuses its catalog with the output of the world’s largest video site; Apple Music offers a karaoke mode. And Spotify has its playlists.
In Mood Machine: The Rise of Spotify and the Costs of the Perfect Playlist, a new book from the journalist Liz Pelly, the playlist is the locus for many of Spotify’s troubling practices. Pelly began her project after a music-industry contact suggested that she investigate how the company’s official playlists were shaped by the major labels. The book’s title suggests a critical corporate biography, but it’s more like a siege campaign, with Pelly pounding away at nearly every aspect of Spotify’s business: its rosy origin story, its entanglements with the big recording companies, the power dynamic of its relationship with independent artists. Her biggest swings are aimed at Spotify’s recommendation framework: the back-end machinations that silently power the playlists available to its 600 million users. You may not agree with Mood Machine that Spotify’s mixes are an existential threat to the way people discover music, but you may marvel at how much effort goes into recommending a song that sounds like a different song you liked three months ago.
The company brands itself as the ultimate curator, which is amusing considering that, for years after its 2006 founding, Spotify had little interest in playlists at all. Initially, Pelly says, the company was content to let its subscribers (and a galaxy of third-party music-discovery services) churn out genre guides and road-trip soundtracks. By 2013, Spotify users had created 1 billion playlists. That year, Spotify acquired the playlist company Tunigo, registering a shift in its marketing strategy. On top of access to its library—which, again, was practically the same as its competitors’—Spotify was now offering expertise. Millions of people started to follow professionally curated playlists such as Viva Latino and RapCaviar, which were constantly updated with the buzziest tracks. In 2015, Spotify rolled out Discover Weekly, its personalized recommendation playlist; the next year, users began receiving up to six personalized playlists a day.
All of this may have felt like a bounty to anyone who came of age when discovering music was much more difficult. Before the internet, the methods of finding new sounds were decentralized and not equally accessible: terrestrial radio, cable television, music magazines, cool older sisters. Maybe your town could support an alt-weekly with a robust arts section; maybe it had at least one independent record shop. But any approach would cost money and time. Often, you would hear a cult artist or album praised in the most compelling language—and because you couldn’t actually find the music in the real world, it would remain an object of speculation for years.
Spotify wasn’t created for this type of obsessive. Initially, it wasn’t created for any music fans. As Pelly notes, the business started like many other 21st-century tech companies: as an ad service looking for a delivery mechanism. (“Should it be product search? Should it be movies, or audiobooks?” Spotify’s co-founder Martin Lorentzon, who made his fortune in affiliate marketing, is quoted as recalling.) Other streaming services such as Imeem, PressPlay, and Spinner had tried countering the pay-to-own model of Apple’s iTunes Store. The major labels negotiated tentative treaties with these early players, capping monthly streams and making only some of their content available.
But by the mid-2000s, it was clear that these were merely half measures. Spotify settled on its model at just the right time: Reeling from plummeting CD revenue and the rise of file sharing, the music industry was suddenly open to software offering access to tens of millions of songs, engineered to play tracks instantly and without limits. The major labels signed over their libraries in exchange for massive concessions: dedicated advertising space, guaranteed minimum revenues, shares in Spotify’s business. The biggest labels—Sony, Universal, and Warner—were betting that Spotify could make a product more compelling than piracy. The bet paid off. As Spotify became the most successful music streamer on Earth, the American music industry stanched the bleeding, and revenues rebounded to pre-internet levels.
Once Spotify built its audience, it wanted to keep users on its app as long as possible. According to Mood Machine, the company’s data indicated that a huge portion of its streams came from “passive listening,” an increasing percentage of which involved functional music: songs designed to enhance everyday activities—meditation, answering emails, even sleeping—without sticking out too much. Where other services had experimented with exclusives from pop A-listers, Spotify’s playlist editors began churning out “chill” mixes populated with songs from mostly anonymous ambient- and instrumental-hip-hop producers. These artists found that placing the right song on the right functional playlist could pay their rent, though a “hit” rarely implied an active fandom outside the platform. Playlists enabled musicians to earn a living without cutting in record labels or PR pros, one former editor tells Pelly, “but that didn’t help them fill a one-hundred-fifty-cap venue or sell merch.”
The great chill-out boom is just one stop on Mood Machine’s fascinating history of Spotify’s official playlists. At first, they were made by professionals hired for their taste and judgment. In 2017, Spotify introduced “algotorial” playlists, a slightly knotty process through which the company’s algorithms generated a pool of songs based on their particular emotional and sonic qualities, editors whittled these songs into a manageable playlist, and the algorithms then sorted these editor-selected songs into an ideal order. Noting the popularity of TikTok’s “For You” feed, Spotify leaned even harder into algorithmic recommendations. Click on a mood-focused or genre-specific mix, and you’ll usually get a version that was “picked just for you.” In practice, this means tracks that share enough of those characteristics with tracks you’ve already played.
In a data-drunk era, this is what passes for discovery. Like so many other products influenced by machine learning, Spotify’s playlists can’t generate something new—say, a wholly fresh and unheard sound—for its users. They instead offer the flash of recognition, rather than the mind-scrambling revelation that comes only when you hear something you’d never expected. Because they’re all drawing from the same massive catalogs, the music streamers are conduits—between artist and listener, or listener and song. In a library north of 100 million tracks, you’d think it would be easy work finding candidates for playlists such as “lofi autumn beats” and “Bossa Nova Dinner.”
But around 2016, as Mood Machine details, Spotify developed the Perfect Fit Content initiative, partnering with licensing companies that paid studio pros to bang out easy-listening ditties for the streamer—in essence, replicating popular playlist sounds on the cheap. (Spotify has not disputed the existence of the PFC program.) In recent years, Spotify developed its Discovery Mode program, in which labels exchange a portion of a song’s royalties for priority status on an algorithmically generated playlist—with no disclosure to the listener. Pelly argues convincingly that this is the streaming version of payola: the illegal practice of promoting songs on the radio without disclosing the payment. (Payola laws apply only to terrestrial radio stations.) But in a crowded marketplace with fewer revenue streams, enough artists enrolled that, according to Pelly’s reporting, Spotify’s internal Slack channels were lit up with glee.
Mood Machine is at its most compelling when peeking into Spotify’s internal strategies and its employees’ real-time reckoning with their implications. This is also where Pelly overplays her hand. In the conclusion, she analyzes two alternatives to the Spotify model: streaming services run by public libraries, and cooperatives of independent musicians. Spotify’s playlists—atomized, contextless—clearly run counter to her ethos, which is rooted in community building and intentional listening. “At a certain point, a streaming listener may very well come to believe that what the machine suggests is indeed what they like, not because it’s true, but because they can see or feel no other option,” she writes. But less clear is whether the company’s playlists are truly changing how the median listener approaches music discovery. Users are still generating hundreds of millions of their own playlists, mined from one of the largest collections of songs available in history. Mood Machine persuasively demonstrates how Spotify guides its users down certain roads—but it’s not impossible to choose a detour.
Today, Spotify boasts that one-third of its users’ discoveries come “via personalized recommendations in algorithmic contexts.” Like video streamers’ triumphant press releases about how many hours were spent watching their “hit” movies, the statement strains credulity. Is a discovery a song the user favorites—or just doesn’t skip? And how could Spotify possibly know if someone had never heard a given song before? In any event, the company’s cited percentage is eerily similar to one given in its 2018 IPO filing, which Pelly summarizes: “Spotify-owned playlists, both editorial and algorithmic, then accounted for over 31 percent of users’ listening”—less than half of all listening done through playlists.
In raw minutes, that’s a staggering sum. But even by Spotify’s accounting, it’s just one part of the streaming user’s diet—often supplemented, still, with radio and physical media. At one point, Pelly gives a potted history of Muzak, drawing sharp parallels between Spotify and the bygone mood-music purveyor: their self-serving audience research, their tendency to generate palatable versions of the now sound. Anonymous label owners share concerns about Spotify promoting “emotional wallpaper” and “the watered down pop sound”; Pelly accuses the company of merely “filling the air to drown out the office worker’s inner thoughts.” The possibility that this office worker might click on a playlist of more interesting sounds—say, amapiano or noise-pop—is not the book’s concern. Neither, for that matter, are some of Spotify’s more controversial programming decisions that go beyond music—such as its nine-figure deal with the wildly popular podcaster Joe Rogan. (However objectionable prefab piano jazz may be, Rogan’s critics would argue that his “just asking questions” conspiracism has done far more to hurt society than cheap music has.)
Still, it would be disingenuous to claim that cloud-based listening hasn’t altered music discovery. For the obsessives, the streamers are a sort of last-mile service: a way to dig into something you encountered off-platform. Yet those encounters are becoming rarer. Music-video budgets have tightened, radio playlists have shortened, alt-weeklies and newspapers have closed, and no one’s publishing the kinds of comprehensive reference books that used to come from AllMusic, Rolling Stone, and Spin. Some streamers offer openly human curation—Apple’s radio stations, Tidal’s blog—but that’s just one aspect of their role as warehouse, marketer, broadcaster, and guide. There’s never been this much music available to this many listeners; the ecosystem for discovery shouldn’t feel this fragile or be this centralized.
Mood Machine casts Spotify as the apex predator of this new ecosystem, which, according to Pelly, has the same old vulnerabilities. “The problems faced by musicians,” she writes in the concluding chapter, “aren’t technological problems: they’re problems of power and labor.” She’s correct that Spotify is not a town square but a walled garden. Although your average music fan will always tend toward legible rather than challenging sounds, a functional cultural scene depends on well-maintained, materially rewarding, and diverse avenues for finding new music. Barring a global data-center failure—admittedly not a remote possibility, with the widespread implementation of AI sucking up more and more energy—the streaming model may be with us for the foreseeable future. We are surrounded at all times by an ocean of sound, but Spotify is content to leave us stranded on our islands.
The post The Paradox of Music Discovery, the Spotify Way appeared first on The Atlantic.
