Nieman Foundation at Harvard
HOME
          
LATEST STORY
A big week for tech blowback: Regulation, broken promises, and Facebook victimhood
ABOUT                    SUBSCRIBE
Sept. 15, 2017, 10:14 a.m.
Audience & Social

BuzzFeed’s strategy for getting content to do well on all platforms? Adaptation and a lot of A/B testing

Multiple versions of articles — with different headlines but also of different lengths and using different thumbnail art — are shown to BuzzFeed.com visitors until a winning combination emerges after a couple of hours.

OMG: BuzzFeed gets so many of its posts — from Facebook videos to quizzes to listicles (though apparently it no longer refers internally to articles in numbered format as ‘listicles’) — to go viral through tailoring subject matters to the platforms with the most receptive audiences, plus constant A/B testing.

BuzzFeed data infrastructure engineer Walter Menendez shared an overview of the publisher’s strategy at a talk at MIT on Thursday night. There’s no one secret sauce, and many other digitally savvy publishers employ related tactics (though there were internal benchmarks and metrics BuzzFeed uses that Menendez declined to share during the talk).

“The core secret, I guess, is that we’re focused on people. When we’re thinking about ways that we make content, we focus ultimately on the end user engagement and the emotional state they’ll have after reading our content,” Menendez said. “We want to focus on making sure we’re not just optimizing for eyeballs — we care about that, but also care about a much more substantial interaction beyond you saw the page, liked it, and then left. Ultimately we’re asking, what job does this piece of content perform?”

It’s a strategy that works at scale. BuzzFeed gets more than 50 percent of its traffic from distributed platforms. It uses an internal formula that measures how much traffic every post gets from Facebook, Twitter, and so forth versus from the BuzzFeed homepage, and weights traffic from those other platforms higher than BuzzFeed’s traffic, according to Menendez: “We want to make sure our traffic gets to the farthest reach of people as possible.”

Starting at the idea stage, what are the topics that will hit specific segments of an online audience, who are guaranteed to react strongly — whether negatively or positively — to content about those topics? Animals, music, sex, fandoms, celebrities, and quizzes, according to Menendez. Another successful format is capturing and aggregating the things that are trending on various social media platforms for other platforms, which exposes Facebook-only audiences to, say, entertaining Tumblr posts they might never have seen, because they aren’t on Tumblr (e.g., “What Colors Are This Dress?” which took a Tumblr user’s image and drummed up some lighthearted controversy among readers on BuzzFeed’s own site):

We want to make sure our audience will interact with a piece of content in the way that we think they will. We’re basically taking a gamble: We’re declaring that this is how we think our audience is going to react, if you write a piece of content this way. That’s what we mean by controlling sentiment. We think that that’s the effect that a piece of content is going to have, and as a result, the audience is going to actively share it — is going to say, someone else should see this, should see how I feel.

It collects mounds of data from both its own site and from the distributed platforms, such as when was a post accessed, from where, and on what device. That helps BuzzFeed decide when to push certain types of content to certain types of platform. Quizzes, for instance, do better on weekends.

BuzzFeed data scientists can run their own more complex calculations on this data. Analytics people can look at real-time traffic broken down across different sources (a surge in traffic from a new platform suggest to BuzzFeed that it might be worth getting more of its content onto that platform). Based on internal thresholds, BuzzFeed can add an additional “Trending” badge to stories performing better than expected.

That data also feeds various Slackbots that send alerts when a post reaches a certain threshold or meets a certain goal — information that helps BuzzFeed determine, for instance, when a post might be a good candidate for translation. A published post, of course, is never just one post: Multiple versions of it — with different headlines, but also of different lengths and using different thumbnail art — are shown to BuzzFeed.com visitors until the winning combination emerges after a couple of hours (an editor can still choose not to use that version).

“We A/B test pretty much everything,” Menendez said. “Not just the headline, but also the number in the headline. The thumbnail that’s rendered on Facebook or Twitter: Sometimes people don’t even see the headline, they just see an image and say, ‘Wait. What is in that image?”

POSTED     Sept. 15, 2017, 10:14 a.m.
SEE MORE ON Audience & Social
SHARE THIS STORY
   
Show comments  
Show tags
 
Join the 45,000 who get the freshest future-of-journalism news in our daily email.
A big week for tech blowback: Regulation, broken promises, and Facebook victimhood
Among many weeks of bad press for the big tech companies, this week stands out.
The Honest Ads Act would force Internet companies to change their disclosure practices by January 2018
Plus: A former Russian troll speaks out; a definition of disinformation; Wikitribune’s preferred news sources.
From Nieman Reports: The powers and perils of news personalization
News personalization could help publishers attract and retain audiences — in the process making political polarization even worse.