This article first appeared in DCN.

What makes a great headline, and why? How can headlines make the casual skimmer stop and read? While much has changed in media’s shift from print to digital, these fundamental questions haven’t. Editors and writers correctly describe headline writing as an art — but with all the technology out there, there is a scientific way to put evidence behind that art, and help publishers grow their engaged readership as a result.

Changing behaviors in headline reading

Reading, scanning, skipping, sharing – our reading behaviors have changed dramatically in recent years. Chartbeat data shows that on average, only around half (55%) of readers who click through to content actually read what they land on.

The context of headline writing has changed as well. Media objectivity, which involves writing factually true and balanced content, is sometimes at odds with the goals of social media marketing, which values metrics like shares, likes and clicks. Increasing readers’ engagement with content can align objectivity and editorial integrity with the need to grow audience in a world where more than half of traffic to publisher sites is driven by platforms like Facebook.

However, publishers can support those better reading behaviors. Recent Chartbeat research shows that despite our changing reading and writing habits, there are scientific ways to improve the likelihood that something will get read.

The role of language and technology in headlines

In an analysis of around 100,000 headline tests and 250,000 individual headlines, we examined linguistic traits of successful and unsuccessful headlines and found that language really does matter.

What we see is that words like “what” and “where,” as well as numbers, quotations and superlatives (like best and worst) lead to more readership, whereas using question marks or time references can actually hurt. Interestingly, short headlines actually have a negative effect on readership of content as well, whereas notably longer headlines have no effect.

In a separate study, we also looked at the impact of headline testing technology and its ability to improve the number of visitors who read for more than 15 seconds. What we found surprised us. In a comprehensive evaluation of headline tests that use Chartbeat’s multi-armed bandit testing model, we discovered that alternative headlines – ones that, without testing, would never have seen the light of day – outperform the original roughly two-thirds, or 62% of the time. That means most headline writers only get headlines right the first time for 38% of stories.  But technology can vastly improve these results.

A headline should not only entice readers to click and see more; it should drive consumption of a story. Of those 62% of stories, the alternate headline saw on average a 78% lift in traffic. It also led to a 71% lift in readership, measured by quality clicks: visitors who spend more than 15 seconds or more of engaged time with an article.