Culture industries increasingly use our data to sell us their products. It’s time to use their data to study them. To that end, we created the Post45 Data Collective, an open access site that peer reviews and publishes literary and cultural data. This is a partnership between the Data Collective and Public Books, a series called Hacking the Culture Industries, brings you data-driven essays that change how we understand audiobooks, bestselling books, streaming music, video games, influential literary institutions such as the New York Times and the New Yorker, and more. Together, they show a new way of understanding how culture is made, and how we can make it better.
—Laura McGrath and Dan Sinykin
What does vibe look like as data? Spotify has an answer. Their vibes playlists encapsulate the cultural buzz about vibes as the key measure of online life (from “vibe shift” anxiety to thought pieces and histories). Drawn from sonic data, user likes, and maybe some old fashioned “payola,” vibes playlists, in the company’s words, “curate for culture rather than genres.” The whole process is exemplified in Spotify’s description of its POLLEN playlist: “Genre-less. Quality first always.”
For Spotify’s editors and project managers, genres like hip hop, country, salsa, and EDM seem old fashioned; they’re too closed and rigidly circumscribed, separating out different listeners, rather than bringing them together. Vibes, on the other hand, seem to promise connection through the algorithm. In Spotify’s PR campaign, vibes are a more open and fluid way to listen together.
As sweet as that sounds, it’s hard to trust a company that has been in the news this year for platforming Joe Rogan’s misinformation campaigns and hate speech to the tune of $200 million. If Spotify claims the shift from genres to vibes is a new way to organize a sense of community, maybe it’s worth questioning how their data organizes vibes. If we understand the algorithm better, maybe we’ll understand the potential consequences of becoming vibes listeners.
Journalist Liz Pelly, who has criticized the naked capitalism of Spotify’s playlist-to-podcast pipeline, was already making the case five years ago that the real “danger” of Spotify was in their “chill playlists.” These are the vehicles, she argued, for “what its algorithm manipulates best: mood and affect.”
Pelly was shouting from the virtual barricades what, implicitly, we might already know about our lives online: we are users who feed data to algorithms, which feed us new content on the basis of that data. Our history of clicks is the product of both our desires and what we’ve been told to like. Spotify’s chill playlists, in this mode, are like any other recommendation algorithms: they produce a mood of “emotional wallpaper,” which smooths over the real concerns of a burned-out userbase desperate to tune out.
More recently, the scholar Robin James has charged that it’s not just chill playlists, but Spotify’s entire set of vibes playlists, and the concept of vibe itself, that reveals how algorithms organize (and manipulate) users’ moods, affects, and tastes. “Vibes,” she writes, “are how we see ourselves the way algorithms see us.”
But vibes are more than just a reflection of machine-learning approaches to organizing our lives. Spotify’s attempts to self-consciously “curate for culture” intersect with the dynamic feedback loop of algorithmic biases and user upvoting. In other words, their playlists have become a company’s changing archive of musical and social desire. The ever-growing access to songs and playlists (for a fee) is what keeps people listening despite the misinformation, musician exploitation, and manipulation. And, as people keep listening, liking, and sharing, they’re doing something more: “Vibe,” James warns, “is one of the vectors through which we’re negotiating post-identity patriarchal racial capitalism.”
So, the politics of vibe, following Pelly and James, emerge through algorithms, like Spotify’s, that both reflect and manipulate affect. In that case, perhaps it’s worth using machine learning to examine the difference between genres and this new affective order. If we’re suspicious about the algorithm’s role, why not study the data that plays into it?
We set out to do just that, using a dataset of more than 10,000 songs (with artist names and release dates) encompassing 4,820 unique artists. In order to compare vibes with genre—and discover if the critics and Spotify are right about their differences—we drew our songs from a random sample of playlists from Spotify’s top 20 most “liked” vibes playlists (such as “Good Vibes,” “Chill Vibes,” etc.) and from four genres (country, hip hop, R&B, and rock).
Out of these 10,000 songs, our research team hand-labelled 2,310 tracks—selected at random, ensuring an even spread across categories—with demographic data on the 1,544 unique artists represented. This data included the artists’ race, gender, and language, as well as their city, state, and country of origin. We then used logistic regression classifiers, entropy tests, and simple counting to see what Spotify’s data might tell us about the emergence of vibes playlists, the potential shift from genres to vibes, and what a “post-identity” ideology sounds like to Spotify and its users.
Would the data reveal genres and vibes on Spotify defined not just by sound, but by social demographics? Do vibes hold up to the promise of representing a more open cultural space of intersecting identities? What kind of world does Spotify—through its algorithmic sorting of millions of users’ desires, through our aggregated listening—produce for us to hear? The following is what we learned.
Spotify’s algorithm—the object of so much social criticism and source of the company’s fortunes—incorporates a variety of audio features, which have been defined and discussed by data scientists and critics online. These include fairly straightforward ways to differentiate between songs—including “loudness,” “acousticness,” “tempo,” “speechiness,” and “instrumentalness” (whether or not a track has vocals)—but also harder-to-define characteristics, such as “danceability” (tempo and regularity), “energy” (based on difficult to quantify concepts like timbre), and “valence” (sentiment). In search of the sound of vibes playlists, in general, we built a classifier based on these audio features for each track on our vibe and genre playlists.
Our results revealed that the vibes playlists we studied are quieter, sadder, and less energetic than the playlists from the genres we’d chosen.1 Liz Pelly’s frustration with chill playlists would seem to describe a general condition of vibes on Spotify: they’re a soundtrack to subdue. These results give us a sense of what vibes sound like compared to the chosen genres. But what about the social dynamics linked to these sonic forms?
How is it that genres, in Spotify’s curation, have remained so racially segregated even as we hear all the time about crossover stars or boundary-breaking musicians or groups?
One obvious starting point for this question might be the metadata for POLLEN, that “genre-less” playlist that Spotify’s producers herald as the end of our old tribalism. POLLEN is updated regularly; our analysis is based on a version of the playlist from the spring of 2022. Does POLLEN break with genre and usher in a “post-identity” playlist?
At first glance, the racial demographics of the performers on the POLLEN playlist are striking. We found that traditional genre playlists on Spotify skew largely white or Black: country is 92.6 percent white, for instance, while rock is 88.6 percent white; hip hop is 79 percent Black, and R&B is 76.5 percent Black. But the “genre-less,” vibing POLLEN playlist features musicians who are 46.5 percent Black, 27.7 percent white, and 25.7 percent other people of color.2 In terms of racial demographics, POLLEN stands out even among the vibes playlists in general, which were 54 percent white, 34 percent Black, and 3.7 percent Latinx. The most racially diverse of the playlists we examined, POLLEN seems to have broken with the racial divisions that dominate traditional genres on Spotify, even though it continues to underrepresent non-Black people of color. So, perhaps vibes do enable greater diversity?
That’s only part of the story. If you examine how racial demographics have changed in music over time, as represented by Spotify, POLLEN appears less exceptional. Put otherwise, generation is correlated with racialized separation on the Spotify playlists we examined: as we move from music of the ’60s to the present, racial representation becomes increasingly diverse.
For instance, we found (using a classifier trained on race, the release date of tracks, and the musicians’ generational affiliation) that artists on Spotify are more likely to be Black or a non-Black person of color if they are in Generation Z. Boomer artists, meanwhile, are twice as likely not to be Black or a non-Black person of color. As we get closer to the present (in terms of birth year), artists on Spotify playlists are more racially diverse. And, relatedly, more Black artists and people of color are present on playlists that draw on music made in the past few years.
While these changes correlate with changing generational demographics in the United States, we can’t underline enough that these observations aren’t a raw statement about the music world overall. Instead, this vision is an effect of the portrait Spotify presents to its listeners, albeit one the company’s listeners have upvoted to become the most popular. The playlists we analyzed, after all, were a random sample taken from the most popular playlists on the platform; users, as voters, are at least partially responsible for that popularity. Whatever bias exists in the algorithm, users fill in the dots with their support for one playlist over another.
As it happens, this generational trend also distinguishes vibes playlists from genre playlists. Given the generational identity of an artist (Gen Z, millennial, Gen X, or boomer), our linear-regression model can reliably predict, with statistical significance, if the musician will appear on a genre or vibes playlist. Is the singer or band millennial, or Gen Z? Then you’ll probably find them on vibes.
In other words, vibes, when taken as a group, trend ultracontemporary. Indeed, 96 percent of songs on vibes playlists were released in the past five years. (Rock, at 30 percent, and country, at 48 percent, are the oldest playlists; but even hip hop, at 75 percent, and R&B, at 63 percent, don’t come close to the presentism of vibes.) And while release dates don’t necessarily match musician age—Bob Dylan and Dr. Dre (a cusp boomer!) still release new tracks—our data shows Spotify’s vibes playlists are dominated by Gen Z and millennial musicians. Rather than a black box of impenetrable mystery, vibes are simply younger musicians putting out recent music.
How is it that genres, in Spotify’s curation, have remained so racially segregated even as we hear all the time about crossover stars like Taylor Swift, or boundary-breaking musicians or groups like Odd Future or Lil Nas X? Are these genre-defying figures visible in the data, recognizable against the genres they’re mashing up or fleeing? Which musicians help to define their genres, and do some musicians define vibes better than others?
With a logistic-regression classifier using musicians’ names as our data, we found that names like Billie Eilish, Olivia Rodrigo, and Justin Bieber, as well as Jeremy Zucker and Ruel (the latter an Australian singer) are the key features or figures that define vibes against and across individual genres. Meanwhile, rock is associated with My Chemical Romance, David Bowie, U2, and Guns N’ Roses; hip hop with Drake and Nas, but also more capacious tags like “dogg” and “big”; R&B with Mary J. Blige, Jazmine Sullivan, and Ari Lennox; country with Toby Keith, Kenny Chesney, and Alan Jackson. Names, in other words, are the key tags that allow the machine to classify between one genre and another, as well as between genre and vibe. (There are some differences at the level of genre vs. genre classification: 2Pac, for instance, becomes important to distinguish hip hop from R&B, and Drake drops out, although R&B retains its familiar names.)
If vibes have created a newly diverse and inclusive world of music, differentiated from that created by genres, is it because their playlists are saturated with genre-defying artists? It doesn’t look like it.
One way to model and test this question is to examine the entropy, or randomness, associated with a given musician and a given genre: a low score shows the musician is easily classifiable; a relatively higher score means they don’t fit so neatly into one genre or another. Notably, not one of the names strongly associated with vibes appears on either end of the spectrum. Certain names—Rascal Flatts, SZA, Eric Church, and Muni Long—are exceptionally stable in their genres, but Bieber, Eilish, Rodrigo, and others closely associated with vibes still fit within traditional genres. If we can call Rascal Flatts a country artist, it’s unclear who might be a vibes artist. So, the vibes playlist vibe—in other words, the features that would distinguish vibes from genres—must be elsewhere.
Vibes playlists, it seems, are composed of songs by young, relatively racially diverse musicians associated with mainly one genre who have released tracks recently. Oh, and most of them are from the Southern United States. If ’90s hip hop was dominated by battles (lyrical, musical, and physical) between the coasts, and rock relied on LA, San Francisco, and New York to generate the most recognizable of its new sounds, the country’s musical center now seems to have shifted southward.
It’s not just vibes: the preponderance of musicians in all categories we analyzed—with the notable exception of rock—come from the South. Country remains the most homogenous (with 79 percent of its songs made by artists from the South), but in hip hop (33 percent) and R&B (41 percent), the South still outpace other regions. Rock (at 25 percent) is only more dependent on the West (38 percent). Vibes playlists, however, remain the most regionally diverse (with 32 percent of the songs we analyzed made by artists from the South; 15 percent from the Midwest; 30 percent from the West; 24 percent from the Northeast). And, after rock (with 40 percent of songs made by foreign-born artists), vibes playlists emerge as the most international (at 35 percent) as well.
Spotify (like Apple Music, Pandora, and other online music-streaming platforms) might impress listeners by recommending new songs that immediately sound like new favorites. This process can feel like magic. But what’s remarkable, we found, was how well Spotify’s algorithms work with the metadata without having to take into account any actual sounds. It’s more like name dropping at a party than talking with friends or strangers about the drive of a specific beat, the oddness of a chord, or the twist of a lyric.
Of course, we can hear variance between and even within individual tracks on vibes playlists, and we can see some of that variance in the sonic data of Spotify’s audio features. However, when we listen with the machine—when we participate in aggregated listening—Spotify’s vibes cohere, sonically, into music that’s sadder, quieter, and less energetic than the genres Spotify’s tastemakers claim the vibes supersede.
It’s worth recognizing the difference between aggregated listening and listening together. Spotify’s aggregated listening bases recommendations, as the platform tell users, on its “editors,” its algorithm, and, “in some cases, commercial considerations.” The platform’s ambition to “curate for culture” is meant to sell things, to produce a portrait of the music world that enough users want to pay for. Other platforms, like Bandcamp, take a different approach, one skeptical of “algorithmic cultural determinism.”
In the compromises of platform life, we’re caught in that negotiation Robin James attaches to vibe. To participate in that negotiation—rather than just buy what they’re selling—you need a community of listeners, restless and skeptical, eager to hear something different from someone new, aware of the roles we play in making patterns for ourselves, with others.
- Those broad differences are also reinforced by identifying which tracks on a vibes playlist are the least vibes oriented, that is, the least recognizable, according to our classifier, as a vibes track. (Whereas the average vibes “loudness” score (on a scale of -60db to 0) is -17.61, outlier tracks average -5.272. Likewise, the average vibes valence score (0 to 1) is 0.351, while the outliers average 0.709. And so on.) ↩
- Since the data is so sparse for artists on this playlist who are non-Black people of color, we combined these musicians into one group for the purposes of our analysis. ↩