iStock/aydinynr
We all know what it’s like: You click on CNN’s website for the latest election news. After reading it, you get pointed to a story about Beyoncé’s Super Bowl performance and then to one about Kanye West’s latest meltdown. Before long, you’re reading an interview with El Chapo’s “beauty queen wife.”
Some of these recommendations might be dismissed as annoying “clickbait.” But in a media landscape where millions of additional articles surface every day, recommendations are hard to do without. Readers need help navigating through the ocean of content. Publishers need to keep readers engaged, and they often rely on revenue that comes with hosting recommendations.
The result has been the rise of content recommendation services, which direct people from articles they are reading or video they are watching to other content that is likely to interest them. It is a difficult challenge because the recommendations have to be generated in real time and provide real value.
If people feel they’ve been tricked by a piece of genuine clickbait — say, a headline like “Hollywood Starlet Arrested Again!” on what turns out to be a 5-year-old story about Lindsay Lohan — they are likely to stop clicking on these recommendations. That can cost a publisher its readers in the long run. As in so many other businesses, success in online publishing is about building relationships.
Now, a team of researchers from Stanford and Columbia universities is proposing a smarter strategy. The key, they write in a new paper, is to go beyond an article’s “clickability” and examine its “engageability.” Instead of looking only at whether readers are likely to click on a recommendation, the idea is to look at what readers are likely to do after they click on a suggestion. How likely are they to pursue a follow-on suggestion at the end of the second article?
The team consisted of Yonatan Gur at Stanford Graduate School of Business and Omar Besbes and Assaf Zeevi at Columbia University. Gur and his colleagues don’t dispute the relevance of clickability, but they argue that it’s equally important to focus on keeping the reader engaged. The objective isn’t simply to get a person to click a single suggestion. It is to create a whole chain of clicks — the longer, the better.
Gur and his colleagues analyzed billions of recommendations produced by Outbrain, a worldwide leading provider of such content recommendations. They came up with a system for measuring engageability and then showed that it is indeed a critical “click driver.”
In a pilot project involving visitors to a major media website, the team tinkered with Outbrain’s algorithms and compared between recommendations that accounted for engageability with recommendations that only looked at clickability. The result: a significant increase in the number of clicks when the recommendations accounted for engageability.
Gur and his colleagues outlined a new way to evaluate content based on different combinations of clickability and engageability.
Tricia Seibold
The “good articles,” as Gur calls them, are those that have high ratings on both measures. These would be recommendations that attract a lot of readers and also prompt many to click on follow-up suggestions. The “bad articles” have low ratings on both measures and aren’t likely to be recommended in the first place.
A third major category is what they call “traffic traps” — with high click-through rates but low engageability — and they can be a hidden source of trouble. Think here of “Hollywood Starlet Arrested Again!” It may well attract clicks, but readers are turned off by what they find and leave the site.
A big problem for both recommendation providers and media websites is that traffic traps seem like good recommendations by the traditional criteria, and thus stay in the system a long time. That can cause long-lasting damage to readership. By measuring engageability, though, it’s relatively easy to spot traffic traps and separate them from the good articles.
The fourth category is “niche opportunities,” which are the mirror image of traffic traps. These recommendations attract a fairly small audience but are highly engaging to those particular readers. Think here, for example, of a high-quality article on architecture or chess or an important health issue. It piques the interest of only a small target audience, but those readers will be eager to read more.
In contrast to the traffic traps, niche opportunities are a source of hidden value. They are typically ignored by conventional recommendation algorithms because of their low click-through rates, but they can be very valuable to media websites.
Does high engageability guarantee high quality? It’s a complicated issue because it is hard to objectively define quality. Which has a “better” quality — a bikini picture from Sports Illustrated’s swimsuit issue, or a meaty article from the New York Times? It depends on whom you ask.
What their research does show, the scholars say, is that measuring engageability in real time, together with clickability, offers a practical yet much richer way to evaluate content, keep readers involved, and get the right content to those who want to see it.
For media inquiries, visit the Newsroom.
