When Jonah Peretti cofounded the Huffington Post, search engines were the de facto traffic directors of the web. And his success was often tied to the rankings of HuffPo pages in search results.
“Partly it was the right keywords, partly it was being fast when the story would break, partly it was knowing which sort of nouns were important and how to write headlines and make an authoritative page,” he says. “And the thing that was unfortunate about all of it was that the person who decided whether you were successful or not was a robot.“
By the time he started BuzzFeed in 2008, times had changed. Social rivaled search as a way to discover content, and it proved a much more difficult puzzle. Instead of predicting the actions of methodical robots, Peretti now believes success lies in understanding what makes real human beings share a piece of content with their friends.
His new operating question is, “What makes something go viral?”
There is no black-and-white answer. But BuzzFeed–best-known for photo-heavy listicles such as “11 Totally Radical Birds On Skateboards”–has hired a team of four data scientists to find the gray ones. It’s built a proprietary data analytics platform that turns the site into one giant experiment in virality.
Most websites rely on third-party tools such as Chartbeat and Parse.ly to analyze their traffic. But for Peretti, data analytics are just as much part of a website as the content they monitor. As he sees it, Apple builds both the hardware and the software that runs on it; BuzzFeed makes both the content and the tools that analyze it. On Friday, it announced the acquisition of Facebook-data company Kingfish Labs, the maker of a dating app called Yoke, to help study what makes Facebook sponsored stories go viral–a move that highlights its desire to be both a content company and a data company.
In addition to its own site, BuzzFeed’s tools are also measuring traffic on sites such as Time magazine, the Huffington Post, and Fox News, which exchange their data for use of BuzzFeed’s data dashboard tools and alerts that tell them when one of their stories has viral potential. Together, the sites account for about 300 million unique views each month in addition to about 20 million unique views each month on BuzzFeed. That’s a lot of data.
“What [we][/we] do is take huge amounts of data and build models that tells us what it takes for content to go viral,” explains Ky Harlin, a data scientist at BuzzFeed. “We then use these models to analyze real-time data, generating “scores” which indicate the potential of an article to go viral.”
Editors use these scores when deciding how to promote each piece of content. The theory is that more “seed views” on content with viral potential will ultimately result in more traffic than content that gets clicked but not shared. All web publishers know this. Few have been able to quantify viral potential before it’s obvious.
Most still use clicks as their main metric. BuzzFeed is more interested in where the clicks come from. An overlay its editors can see on the homepage, for instance, assigns each article a “viral lift” score. This shows how many times one piece of content is shared (and clicked) per view it gets directly from the website or an ad (a “seed” view). While an image of a woman in a bikini might get a lot of clicks, it won’t get shared. Those radical birds on skateboards might ultimately get more attention on social networks. BuzzFeed tries to arrange its homepage so articles with the most viral power are most prominent. Birds before bikinis.
“It’s less about changing the content that is created and more about predicting early when something is taking off and being able to promote that more to people,” Peretti says.
But as many a failed YouTube marketing campaign suggests, predicting what will be viral is impossible. It’s not that BuzzFeed has discovered an exact recipe for cracking the viral code. Rather, it’s using machine-learning tactics to maintain a constantly evolving model. It’s acquiring talent like web video wiz Ze Frank to ask “what makes content viral” on more and more platforms.
“The way content went viral six months ago is not the same thing that makes it go viral today,” Harlin says.
“There’s art and science to it,” Peretti adds. “Which I think is a lot more fun business to be a part of, even if it’s less predictable…. The math helps you have better understanding and helps you have more creative ideas, but you can’t replace the creative ideas.”
[Image: Flickr user Giacomo Bucci]