Everyone had to see this. It was early 2007 when Sadia Harper called her YouTube co-workers to her desk to watch. On her screen, a preteen with a buzz cut and an oversize dress shirt was belting out an Alicia Keys song. “This kid is amazing,” Harper said. The singer’s mother had been badgering her with emails to feature her son, Justin Bieber, on YouTube’s homepage.
Harper was one of YouTube’s “coolhunters,” a team once tasked with curating videos on YouTube.com. Today, YouTube is known for its powerful recommendation algorithm: a system criticized for driving people to radical beliefs, conspiracies, and online echo chambers. (YouTube and its parent company, Google, declined to comment for this story.) But in the site’s early days, the service took a different approach to recommending footage.
When YouTube was founded in 2005, people often arrived at its videos from a link sent by a friend or found in a Google search. A fair number also came in through its homepage, which company engineers populated by slotting in popular or amusing videos between coding sessions. But wider exposure and new business deals demanded a better approach. By the summer of 2006, YouTube had already become a mass-media must-see, with more than 100 million daily views. Verizon Wireless, which agreed to put a limited version of YouTube on its mobile phones later that year, wanted a more finely curated selection of clips. Apple, which was preparing to debut a new gadget, the iPhone, was interested in something similar, according to multiple people at YouTube involved in discussions.
So YouTube hired an editor. Mia Quagliarello, who had been a manager at iTunes, joined the company to program YouTube’s homepage and packages for partners such as Verizon. Flush with Google cash—the internet giant had just acquired the start-up for $1.65 billion—Quagliarello recruited a small team of staffers to keep close tabs on YouTube’s sprawling culture. She started with Joseph Smith, a graveyard-shift screener at YouTube whom everyone called “Big Joe,” who was remarkably adept at spotting budding viral hits before they exploded in popularity. Harper, a high-school friend of one of YouTube’s founders, Steve Chen, joined shortly after. In the months that followed, Quagliarello brought in a journalist, a radio DJ, and actual YouTubers as editors to sift through content on sports, comedy, politics, and other fields, finding gems that viewers might love. Officially, these team members were called “community managers.” But a colleague devised a more resonant title for the team: the coolhunters.
YouTube wasn’t yet a commercial hit. The company started sharing advertising sales with popular YouTubers in 2007, but the funds were sparse. Influencers as we know them now didn’t exist. No one used the term creators. Instead, YouTube was overflowing with aspiring comics, filmmakers, musicians, hobbyists, and enthusiasts in every niche imaginable, searching for an audience or just tinkering with a new canvas. (Briefly, in 2006, one of the most popular YouTubers was Peter Oakley, a well-dressed British retiree who went by geriatric1927 and would reminisce about his life.) Quagliarello encouraged her team to make videos introducing themselves. Harper shot hers in her bedroom and also posted clips of DIY crafting, another emerging YouTube subculture. She asked viewers to send videos to her email address, which is what Bieber’s mother kept doing. Harper had to politely tell her that YouTube preferred to feature original songs, not covers. Still, even when the coolhunters passed on a technicality, YouTube minted stars—a year later, a record executive would find Bieber’s videos on YouTube and make him a pop sensation.
The coolhunters each had their own way of finding hits. Every morning, Harper scoured a list she’d assembled of blogs and internet arcana, searching for interesting videos. When she found ones worth putting on the homepage, she would add them to YouTube’s “Featured Videos” banner, which stacked small frames of videos in a column of 10. Her team swapped these slots every four hours, giving YouTubers behind the videos they selected a guaranteed cascade of views. Before Donald Glover was a celebrity, Harper promoted a comedy sketch in which he mocks an old hip-hop pose, the “B-Boy Stance.” Some picks showed up in wider pop culture. Harper discovered a music video with a catchy, whistling hook from a band called Peter Bjorn and John. A week later, Drew Barrymore wore the band’s T-shirt on Saturday Night Live.
YouTube’s coolhunters were tastemakers, and they were among the first at the company to actually reach out to the delightful weirdos populating its site. Michele Flannery, a former local-radio director who curated YouTube’s music videos, told me she liked to prowl for unconventional artists, such as quirky ukulele players and indie rockers. “Make it really personal and intimate,” she advised musicians posting on YouTube who wanted to be featured on the homepage, “like you’re sitting in your bedroom.” The coolhunters latched onto this aesthetic and experimented themselves. They invited the filmmaker Rob Zombie, and later Wes Craven, to guest edit for Halloween. Steve Grove, the news and politics manager, arranged to have YouTubers submit video questions that would run during televised presidential debates. In 2007, a bizarre, poetic homemade music video, “Chocolate Rain,” blew up on YouTube, and dozens of people uploaded covers; today the video has more than 133 million views. As a lark, the community managers planned their first “takeover,” filling the entire YouTube homepage with tributes to the song. An engineer rushed over in panic, assuming that YouTube had been hacked. The coolhunters would repeat the gimmick and Rickroll everyone visiting YouTube.
During YouTube’s first couple of years at Google, the parent company largely left it alone. Google lawyers helped YouTube fight legal battles, and Google’s balance sheet funded YouTube’s blitzkrieg expansion into new countries around the world. But YouTube was mostly independent.
Eventually, though, Google’s sensibilities started to trickle in. “The Google way of solving problems is to throw machines at them, not people,” recalls Andy Stack, a former YouTube manager. Sadia Harper encountered this first with cars. A few years into coolhunting, she began curating automotive videos as the coolhunters added new categories to cover the site’s expanse. She liked cars. People liked watching them on YouTube—car races, Humvees climbing walls, detailed tutorials on engines. Periodically, Harper would slot interesting footage onto YouTube’s homepage. One day, she told me, a programmer approached her desk and explained that engineers had developed an algorithm for picking homepage videos designed to get optimal clicks. They wanted to test it on a trial category. They picked cars.
The coder loaded a sample page of videos that the algorithm had selected. Enter. Refresh. The reloaded page filled with “revving” videos—footage shot inside luxury vehicles where cameras lingered on the foot or lower half of the driver, usually a woman in heels, pumping the accelerator. Often, leather was involved. Harper had seen those sorts of videos and intentionally ignored them. “That’s a fetish,” she protested. “That’s not what we’re about.”
YouTube wasn’t new to algorithmic sorting, but its first iterations were fairly primitive. When someone clicked on a video, the page’s right flank—its “related videos” section—filled with clips that other viewers who clicked on that same video had watched. The algorithm was accounting for “co-visitation”: People who like this also like that. Erik Klein, an early YouTube engineer, recalls the limits of this approach with huge viral hits that everyone co-visited; viewers, staff joked, were always two videos away from seeing Justin Bieber. Algorithm experiments could go awry, sometimes showing too much of the internet’s dark mirror. Before Google’s acquisition, YouTube programmers once tweaked their system and saw video clicks shoot up, only to discover that “every three or four videos, you would end up with a cat video or someone in a bikini,” remembers Jasson Schrock, a former YouTube designer.
Programmers went back to the drawing board, adding more filters for decency into the code. Google brought more computing horsepower and coding proficiency, letting YouTube measure granular signals such as how long people lingered on videos, what time of day they watched, and from where. YouTube’s algorithms improved. At first, they couldn’t detect a butt from a peach and left that to human moderators, but eventually, YouTube developed skin-detection software to remove obscene stuff automatically. Related videos started clocking more clicks. The formula looked ready for prime time on the homepage.
By 2009, YouTube’s business was preparing for prime time too. Google began tightening its belt after the financial crisis, and YouTube, although a cultural phenomenon, was a perpetual money pit. New managers arrived to whip YouTube into profitable shape. The coolhunters looked less and less relevant to YouTube’s commercial future and to Google’s culture. When Lady Gaga’s hit “Telephone” debuted in March of 2010, Google salespeople wanted the music video, a raunchy, slick featurette set in a women’s prison, to premiere as a paid promotion on YouTube. Flannery remembers the coolhunters protesting that similarly raunchy videos from amateur YouTubers would be “age-gated” and prohibited from the homepage. The salespeople and Lady Gaga won. On another occasion, Harper recalled, a YouTube sales leader asked her to feature an advertiser’s video. When Harper declined, citing the material’s so-so quality, the salesman pointed to the assorted clips that coolhunters had curated and asked her, “Is that stuff any better?”
[Read: How YouTube’s algorithm really works]
According to more than 10 interviews with YouTube staffers familiar with the dynamics, some of the company’s higher-ups had growing misgivings about the coolhunter team. Viacom had sued YouTube over copyright infringement, arguing that YouTube knowingly let pirated copy run rampant. To some, an operation that sifted through videos to feature didn’t help the legal defense that YouTube was a hands-off platform (Google and Viacom eventually settled out of court). And as YouTube launched in more and more countries, replicating a curatorial team for each country felt too costly and time-consuming. “We couldn’t run that fast enough,” recalls Chen, YouTube’s co-founder and early technical chief. Besides, software was cheaper. Some signs indicated that the curated homepage wasn’t driving growth; people went there to search, not to linger or click on clips. A few people at Google thought the coolhunters operated as hidden kingmakers, picking, almost like Hollywood producers or agents, which YouTubers would become stars. And Facebook, then surging in popularity, was attracting users and advertisers based on a social feed of content tailored just so for every individual.
But the most damning case against the coolhunters was that they lacked a way to measure themselves. At Google, everything was measured. Harper, who had studied mathematics, put together a data analysis trying to quantify the impact of her team’s virtual town square. It was not enough. Around early 2010, a new YouTube product manager began meeting with the editorial team to discuss ways that the homepage could be more “relevant” to viewers, primarily with more personalized algorithms like Facebook had. Mark Day, YouTube’s comedy editor, had an epiphany during one of these gatherings. Oh, wait a minute, Day thought. Your job is to eliminate my job.
Shortly after, in 2010, the coolhunters were disbanded. Most members were reassigned to work in marketing roles helping brand-name companies sell on the site. YouTube’s machines would now pick the videos.
YouTube would likely never have become a staple of pop culture without the small, early team that cemented its coolness and nurtured its first viral hits. But at the same time, YouTube may not have become the behemoth it is today if it had kept relying on human curators. In the years since, YouTube has sometimes toyed with handpicking videos, featuring select creators or trends that the company wants to highlight. And YouTube has resorted to curation to manage messes from its scale—putting content from health agencies and news outlets in prominent places during the pandemic and other events prone to conspiracy mongering. Often, these efforts feel drowned out by the platform’s sheer enormity. It’s like “putting a thimble in a gushing geyser,” says Claire Stapleton, a manager hired in 2014 to curate marketing content.
Either way, viewers may not notice these attempts: They usually just watch the video that shows up next in their YouTube feed.
This article has been adapted from Mark Bergen’s forthcoming book, Like, Comment, Subscribe: Inside YouTube’s Chaotic Rise to World Domination.
