The Spotify conspiracy theories about “Espresso,” explained

Steven Phillips-Horst had a problem — er, “problem,” depending on who you ask — that plagues many a modern Spotify listener. No matter what song the podcaster and comedian played, whether by Taylor Swift or Celine Dion, Morgan Wallen or the band How to Dress Well, the platform would immediately queue up Sabrina Carpenter’s “Espresso” next.

“How is it that no matter what direction my unique personality takes me in, it leads me back to this same song that I only heard three days ago that I don’t even really like that much?” he asks over the phone, not knowing that the same thing had been happening to me, too.

The disco-inflected soft pop hit, performed by a Polly Pocket-sized former Disney Channel star, was deemed the song of summer nearly as soon as it was released in mid-April. But Phillips-Horst wondered, was that because the song was good? Or, as he proposed on X, was it because of the tantalizing (if unfounded) theory that Carpenter’s label was paying Spotify to force-feed “Espresso” down the throats of listeners the world over?

“It’s not an ‘algorithm,’” he wrote. “The label is paying for radio play in an old school, industry plant kinda way and it’s honestly refreshing. We are so back etc!” Responses flooded in, with hundreds of others saying they too couldn’t escape “Espresso” or other hits from new-ish entries into the pop game, namely Chappell Roan’s “Good Luck, Babe!” and Tommy Richman’s “Million Dollar Baby.”

The theory seems to be an increasingly popular one: that the biggest music streaming platform in the world has arrangements with record labels to boost streams for particular songs and artists. It’s not exactly a far-fetched conclusion. The history of pop music is rife with payola scandals, in which powerful executives (and sometimes the literal mafia) manufactured hits by paying for radio play, and, after the practice was first outlawed in 1960, finding loopholes to get around each new restriction. “When publicity is valuable, people will inevitably find ways to pay for it,” says Gabriel Rossman, a sociologist at UCLA who studies bribery in the music industry.

There’s no hard evidence to suggest that labels are directly paying Spotify to boost certain artists and songs, according to the music industry scholars and insiders I interviewed. Yet it’s almost impossible to prove that it’s not happening: Deals between record labels and streaming platforms are confidential and algorithms are intentionally designed as black boxes. (Spotify would not agree to an interview, but said in a statement that “The remarkable growth of fan bases for artists like Sabrina Carpenter, Chappell Roan, and Tommy Richman on Spotify is a testament to their hard work and artistry.”)

While the most likely explanation is the most boring — that certain songs take off on music discovery algorithms because listeners seem to like them — the question of why users keep getting served Sabrina Carpenter is far more complex than that. No wonder so many of us think something shady is going on, even if we can’t quite put our fingers on what.

Users tend to refer to a singular, all-powerful Spotify algorithm, but there is no single one that determines what songs are served to its 600 million listeners. Rather each of the platform’s features work differently and have specific (albeit similar) goals. “Smart Shuffle,” for example, queues new songs to a playlist you’ve already made, while the personalized “Discover Weekly” uses an algorithm meant to design a playlist of new songs that are similar to the music you’ve recently listened to. Newer features like daylists and the AI-powered DJ X are also based on users’ “taste profiles.” Search for a specific song and Spotify will typically autoplay music that’s sonically similar to the one you searched for, while Spotify Radio creates personalized playlists out of any song, album, or artist you choose.

Many of these features are the subject of mistrust and theorizing among the listening public. The website PopFiltr found that Carpenter’s new single “Please Please Please” took the No. 2 spot on dozens of artists’ Radio playlists. This would make sense for modern pop acts like Harry Styles or Lana Del Rey but less so for, say, Frank Sinatra, Creedence Clearwater Revival, or one playlist based on the artist “Fart Fest,” which publishes literal fart noises. (I wasn’t able to recreate these findings, as they’re algorithmically tailored to each user, but “Please Please Please” was second in Radio playlists for Clairo, Boygenius, Charli XCX, and Dua Lipa, and was almost always followed by a Chappell Roan song and “Espresso.”)

You’d think all this means that somewhere in the depths of Spotify’s servers, there’s a giant lever that pushes Carpenter’s music out to millions of people at once. But, as Glenn McDonald, the former “data alchemist” at Spotify and founder of the music microgenre catalog Every Noise at Once, explains, “None of it is as centralized as you’re imagining — it’s not like there’s a master sticker on the song that causes everybody to promote it.”

McDonald, who understands how Spotify works better than perhaps anyone, says that the question shouldn’t be about payola or some other seedy arrangement between Spotify and music labels, but about algorithms: “A lot of things on Spotify are trying to keep you in your comfort zone and play you the right things.”

Noah Askin, a computational social scientist who studies the music industry, says that algorithms like Spotify’s typically use collaborative filtering, a concept used by recommendation engines to predict what users want to be served next. The gist is that if User A and User B have overlapping tastes, User A will probably enjoy the other music that User B is listening to.

He also points to the Matthew Effect, or the idea that the already successful will have an easier time becoming more successful. Music recommendation algorithms can make big songs even bigger at the expense of others. “Once you have a bunch of attention, it’s much easier to get more attention,” Askin says. “When a Beyoncé song comes out, it’s more likely to get played because she’s already more popular … and so the people who don’t listen to her but are into pop music, guess who they’re going to get fed?”

He compares this to the Google and Netflix algorithms, which take an impossibly large slew of content and present it to users in a way that will make them spend as much time as possible on their platforms. The platforms do this by pushing out what’s already proven catchy with people with similar tastes.

Neither Roan nor Carpenter is anywhere near as famous as Beyoncé, but Askin suspects part of the reason they keep showing up on pop fans’ Spotify algorithms is due to a sociological concept called “optimal distinctiveness” that explains many listeners’ tastes. “What we found was that the songs that are pretty similar to what's popular but a little bit distinctive are the ones that do best,” he says. “If it all just sounds the same, it gets boring. On the other end of the spectrum, anything that’s too weird, it’s like ... I’m not interested in really pushing my listening horizons that much.”

One 2009 study on optimal distinctiveness found that young people preferred and identified more with music styles at “intermediate” popularity levels, as opposed to either massively popular or esoteric ones. Spotify’s algorithms likely reflect and strengthen that bias, to its own benefit. “What Spotify cares about [is that] you continue to listen on their platform, that you stay on their platform for as long as possible. And what’s going to do that is providing people with songs that sound like what they like but have some degree of novelty to them,” says Askin.

McDonald’s assessment of the Spotify algorithm echoes Askin’s: “Especially if [the artist] is slightly less popular and a bit more attuned to what you usually like, then that makes them a better match than someone who’s very popular and crosses genre or audience boundaries.” A Taylor Swift listener, then, would be more likely to get fed “Espresso” or “Good Luck, Babe!” than a Carpenter or Roan listener would get served Swift.

There are also less tangible reasons why certain songs appear to be overnight successes when they’re anything but. “Espresso” may sound like a lot of disco-inspired pop these days, but Carpenter’s opening stint on Taylor Swift’s Eras Tour, her buzzy romance with actor Barry Keoghan, and a fun persona as “a proudly flippant ditz who is so much savvier than she seems,” as my colleague described it, all undoubtedly played a part in making the song the sensation it became.

Chappell Roan, meanwhile, has been publicly trying to “make it” in music for years; she was dropped by her label after her 2020 single flopped — a single that just entered the Billboard Hot 100 for the first time. After scoring a spot on Olivia Rodrigo’s tour earlier this year and an NPR Tiny Desk concert, more fans were exposed to her high-energy, drag-inspired stage performances, drawing enormous crowds at music festivals this spring. Despite the accusations Roan and Carpenter have faced of being “industry plants” (allegations often marred in sexism), there are clearly other explanations for their recent success beyond literally paying Spotify for it.

If record labels were paying Spotify for streams, it still wouldn’t technically be payola. US laws against payola, or the practice of paying a radio station to play a specific song without disclosing the payment, only apply to public airwaves. Because there is a finite amount of bandwidth for radio, they have to be managed in the public interest, Rossman explains. “The [Federal Communications Commission] gets to tell broadcasters as a condition of their license that they have to follow certain rules.”

The internet, on the other hand, is capable of disseminating virtually limitless data, and therefore the payola laws don’t apply. In theory, anyone can build a Spotify competitor and queue up whatever songs they desire (fun dinner party question: you’re the CEO of Spotify, what’s your “Espresso”?).

Spotify actually does offer a version of pay-for-play that’s theoretically available to any artist who opts in. In 2020, it debuted a feature that allowed recording artists to forfeit 30 percent of streaming profits in order to increase the likelihood that a song will be played during a listening session. The program, called Discovery Mode, was immediately likened to payola by the Recording Academy when it launched.

While music managers said they loved the program — some said it increased a song’s streams by 200 to 300 percent — three members of Congress slammed Discovery Mode for penalizing up-and-coming artists in an effort to beef up Spotify's bottom line, according to a letter obtained by Variety. The program “may set in motion a ‘race to the bottom’ in which artists and labels feel compelled to accept lower royalties as a necessary way to break through an extremely crowded and competitive music environment,” they wrote to Spotify co-founder and CEO Daniel Ek.

The artists using Discovery Mode are likely not the Sabrina Carpenters of the world. In its marketing materials, Spotify highlights the success stories of a handful of smaller acts signed to independent record labels who have used the service. But because of the stigma associated with pay-for-play, it’s understandable that few would be transparent about it. It’s impossible to say for sure how many or which artists use Discovery Mode, however, because Spotify doesn’t label its songs as such.

Major label artists may not need to use Discovery Mode in part because two of the three biggest labels in the world, Universal and Sony, own between 6 and 7 percent of Spotify stock, according to one 2020 estimate. “Spotify is reliant on the major labels in particular for access to the catalogs of music that are driving its growth. It needs to play to their tune,” one music business scholar told NME. The major labels “have these tier one relationships with people that work at Spotify … and opportunities can come from that.” explains George Goodrich, founder and CEO of Playlist Push, an agency that allows artists to pay for consideration on top Spotify playlists and in influencers’ TikTok videos. “It’s all just about connections.”

Indeed, “playlisting” has been another important strategy for marketers since the 2010s. Artists and their teams enlist agencies to submit songs to independent playlist curators ($285 for consideration in 48 playlists, $3,000 for 500), who can either choose to include them or not; services like Playlist Push are clear in noting that paying for editorial consideration is no guarantee that a song will end up in a popular playlist. Playlist Push says they have worked with more than 25,000 artists, labels, and managers since 2017. A prime spot in a big playlist can launch an artist’s career: One record exec compared the Spotify editorial playlist RapCaviar to “Hot 97 in the early ’90s,” helping to blow up songs like Cardi B’s 2017 breakout “Bodak Yellow” and Lil Uzi Vert’s “XO TOUR Llif3.”

One illicit way music marketers have paid for artificial boosts: stream farming. In 2021, Rolling Stone investigated the pay-for-play “black market” where third-party companies promise to net hundreds of millions of streams per month, typically via bot accounts, for a certain network of artists. “There are a few third-party companies out there running this for a lot of the major companies,” one anonymous record label employee told the magazine. “We use them too for some of our artists.” It’s risky business, though. Spotify penalizes artists whose songs have been found to be targets of artificial streams, even in cases where the artist or label wasn’t involved in the scheme. “It’s really terrible for your data profile on Spotify,” says Goodrich.

Without resorting to stream-farming or some other hidden algorithmic boost, music marketers have gotten far savvier at predicting what will hit with listeners. What seems like sudden ubiquity for a song or artist is often extremely canny strategizing by well-connected industry insiders who are adept at responding to online chatter and listening habits. “People will tease songs for weeks, if not months, where the audio is already on short-form platforms but not commercially available on streaming services yet. People also test singles that way to see if their feelings on songs are right,” says Jenny Kaufman, the head of global streaming promotion at Crush Music. Post Malone and Morgan Wallen teased their song “I Had Some Help” on social media for months before it was released in May; it was recently the No. 1 song on the Billboard charts (though now second to “Please Please Please.”) “We’re seeing more and more artists drop full albums, we’re seeing more and more artists do less lead-in singles. That in part is due to the interest in seeing which songs fans will react to.”

On TikTok, which wields enormous influence on the music industry, labels will often partner with talent agencies to pay influencers to use their music in videos. Goodrich points to the success of “Million Dollar Baby,” which Richman teased on TikTok in April and which blew up on the platform immediately. He guesses it capitalized on the savviness and connections of manager Ty Baisden, and artist and agency founder Brent Faiyaz. “Usually, these songs will have an entire team behind it that’s literally social engineering the whole thing,” says Goodrich. “They have extremely creative people who are thinking about how they can make the song take off, and they’re usually pretty good at it. And they’ve figured out some type of formula that no one else has.”

Considering all the unseen, mysterious ways that algorithms function and music marketers do their work, it’s almost easier to think that everything we’re hearing on streaming platforms amounts to straightforward payola. Perhaps it’s not a coincidence that Phillips-Horst’s tweet expresses a bit of ironic nostalgia for the days when payola was a reasonable explanation for a song’s popularity.

True, other kinds of pay-for-play transactions could very well be happening — again, those contracts are confidential, and none of the record labels I reached out to responded to my request for comment — but it wouldn’t fall under the FCC’s payola law. Though experts on algorithms disagree, Rossman says he could still envision labels paying the platform to promote certain songs. “It seems plausible,” Rossman says. “And I would expect it to happen.”

Rossman suggests that such agreements should fall under the Federal Trade Commission’s laws that require influencers to disclose advertisements. If Spotify “wanted to be safe, all they’d have to do is include something in the metatext saying, ‘This artist opted into the Discovery Mode program,’” he says. “Or, ‘You got this certain song because it has these attributes and we provide this weight to those attributes.’ They’re doing that internally somewhere on the Spotify servers, they’re just not sending it to the user.” (An FTC spokesperson told me it does not comment on specific companies unless it has publicly announced an action against them.)

The only reason we’re having this conversation at all is because Spotify and other streaming platforms intentionally obscure how they work. “As a listener, I wish I could tell on any given song, ‘Why is this here? Why should I trust you?’” says McDonald. But crucially, he says, “To me, this is not a Spotify-specific problem, not a music-specific problem. It’s everything about how technology mediates human interaction, if it goes through these black boxes of machine learning or AI. You’re just left [asking] ‘Here's the answer, how do I trust it?’ … I think that's a bad state, that we as people who work on technology have to improve.”

In 2023, Spotify laid off around 2,200 workers, including McDonald. Its efforts of late have centered around launching new AI features, including one in which users can prompt an AI tool to create a playlist based on a few words. Pricing for the service rose again this June, with individual plans going up $1 per month, up to $11.99. The result will no doubt be more anxiety and confusion among its listeners, who will continue to wonder why its algorithms have so much power over the soundtrack to their lives, or whether it’s even an algorithm at all.

“People should demand transparency and control over their own experience,” McDonald adds. “Algorithms are tools, but many times these tools are only operated by the people who run the streaming platform, the search engine, or the shopping site. To the extent that those tools can be given to you, the person, to empower yourself and to amplify your curiosity ... that’s more interesting to me than things that just ask you, ‘Hold still, and we’ll discover something for you.’”

For now, the summer of Spotify conspiracy theories rages on, while the internet debates which artists are industry plants and which labels are doing payola, even if — and in fact, because — we can’t quite grasp who or how. “And the grammy for the best autoplay song of the year goes to,” wrote one person on X, “our payola queen Sabrina Carpenter.”