You’ll never believe how recommended stories are generated on otherwise serious news sites

January 9

In its top story Tuesday, Politico.com covered the Senate’s jockeying over a bill to extend unemployment benefits. Readers found out about the chances for passage of the measure, which will affect 1.3 million unemployed workers and cost $6.4 billion. And at the bottom of the story, in a section titled “Around the Web,” they also found links to some other stories:

●“Chuck Norris Listed His Dallas Home for Sale: Check Out Inside.”

●“How This Comedian Got ‘Accidentally’ Pregnant.”

● “In Case You Forgot How Beautiful Erin Andrews Is.”

The headlines appeared within Politico’s regular news reports. But Politico’s editors had no role in selecting the links or the thumbnail photos that illustrated them. A robot did.

Such links, which appear on hundreds of news sites, including CNN and The Washington Post, are the work of a “news discovery” company called Taboola. The company acts as a middleman between a Web site, such as Politico, and other sites that want to attract Politico’s readers. At regular intervals, Taboola’s computers feed new headlines and photos into the “Around the Web” sections from an inventory of articles, photo galleries and videos supplied by these third-party sites.

When a reader clicks on one of the “Around the Web” links, Taboola and Politico receive a few pennies from the originating Web site. The pennies add up: The payments from Taboola-fed links are “an effective source of additional online revenue,” said Sara Olson, Politico’s vice president of marketing, though she won’t say just how “effective.”

Politico is just one of the beneficiaries of this news-recommendation ecosystem, which has grown into a gigantic, if largely invisible, part of the digital news business. With millions of clicks each day, news recommendation engines drive swarms of readers to and from sites around the Internet. A handful of companies serve all those bottom-of-the-page come-ons — “Top 10 Sexiest Female Athletes of 2013” — to thousands of Web sites.

Taboola, founded in 2007, expects to generate $100 million in revenue this year, said its chief executive and founder, a former Israeli military intelligence officer named Adam Singolda. “Our vision is to change the way people around the world discover content they may like and never knew existed,” he said.

He seems to be well on his way to that goal. Singolda said that one-third of all American Internet users clicked on a Taboola-supplied link last month. Taboola’s main competitor, another Israeli start-up called Outbrain (both companies are now based in New York), served up links that were clicked 730 million times in December, said Rich Ullman, Outbrain’s vice president of marketing.

The recommendation engines do more than promote another site’s stories. Many of the links in those “Around the Web” and “Recommended for You” sections are for promotional content created by marketers, blurring the line between news and advertising.

“Which Country is Currently Leading the World in Innovation?” read one of the five “From Around the Web” links on washingtonpost.com this week. The link led not to a conventional news story, but to a video pitch for Merrill Lynch’s wealth management services.

The links on washingtonpost.com are “an advertising product,” said Emilio Garcia-Ruiz, The Post’s managing editor of digital news. “When you see big publishers in there, it’s because they paid Outbrain to promote their stories. Like with all advertising, if users complained about the appropriateness of an item, we would raise it with advertising.” But, he added, “I can’t remember it happening.”

As the example from Politico demonstrates, the more obvious problem might be the mismatch between a site’s usual reporting and the kinds of stories that a computer decides are “recommended.” Silly, scandalous or salacious links, the cheesy and the sleazy, often appear in places not typically known for such content.

Some of the “recommended” stories on CNN.com’s “From Around the Web” section on Thursday, for example, included “The HOT Pictures of Cheerleaders Will SHOCK YOU” and “These 30 Hot Female Celebrity Bodies WILL SHOCK YOU.” Among the links carried by the usually sober and serious NationalJournal.com was one to “20 More of the Raunchiest Latino Celeb Mag Covers,” illustrated with a photo of a young woman spilling out of her halter top.

CNN declined to comment. A National Journal spokeswoman, Emma Angerer, said the publication has been working with Outbrain to keep “potentially offensive” material off its site. She said no National Journal readers had complained about the listings.

Outbrain and Taboola say publishers can customize their offerings to screen out material they deem inappropriate. Many publishers, including The Post, already use these controls to prevent competitors from promoting their work on the publisher’s site. “You can say, ‘Never show Kim Kardashian on my site,’ ” and she’ll be banished, Singolda said.

The engines’ recommendations are based on algorithms shaped by a user’s Internet behavior and that of similar groups of people. Thanks to tracking software known as cookies, the companies’ computers can learn whether you like to read about sports or entertainment or prefer to watch videos instead of reading articles. They also do some educated guesswork based on broad categories. People in Washington, D.C., for example, might see more links to political stories than people in Washington state.

In theory, at least.

The algorithms are still a long way from knowing that you really don’t want to read the story “No Way These Celebrities Are 60-Plus!,” a Taboola recommendation on HuffingtonPost.com on Thursday.

In fact, digital media analyst Ken Doctor, author of the Newsonomics blog, said news-recommendation engines are “degenerating” in efficiency as they serve up more links from more publishers seeking readers and traffic. “Early on, the click-through rates [by readers] were as high as 6 to 8 percent, because the [recommended] stories were relevant to the stories on the [host] site,” he said. “But there’s been a junking up that does a disservice to the reader. There’s too much ‘click bait’ that has no relation to the actual readers of the site.”

Outbrain and Taboola responded that they constantly refine their systems to better match readers with stories. The ideal, Outbrain executive Lisa LaCour said, would be to offer 10 recommendations, all of which were “irresistible to a reader.” “But we’re not there yet,” she said.

In the meantime, there are all those pennies. Each time someone clicks on a recommended link, the host site and the recommendation companies share about 3 cents, a figure that can rise depending on a variety of factors.

“The idea is, your audience is going to leave [your site], anyway,” LaCour says. “This is a way for you to get paid when your audience leaves.”

Paul Farhi is The Washington Post's media reporter.
Comments
Show Comments
Most Read Lifestyle