Matt our marketing intern was looking for a meaty project. So we gave him an algorithm to solve. We asked him to predict the performance of any piece of content that ran on our network.
His algorithm achieved a 78% coefficient of correlation when predicting page views – it was highly accurate. This exercise yielded a lot of insights about how engineers decide what to read.
Our yardstick for this exercise was the number of pages viewed. In a later post we’ll measure audience engagement such as time on page, comments and social votes.
Matt generated a list of variables that represent the information available to a reader when he is deciding whether or not to read an article. And (drumroll please) here are the results, in declining order of importance:
- Brightness of the image. No kidding. This was the biggest deal. Maybe because literally catching someone’s eye is harder than we think.
- The author. Known authors pull more than unfamiliar names on the byline.
- Technical image. More important than the topic or the title was whether an image was technical. By contrast, fun images (which we love) or dull images actually were net harmful to the open / read rates.
- Title creativity. Boring titles fare far worse than clever titles, such as plays on words
- Length of title (longer is worse)
- Popularity of topic – yes, the actual topic was number 6 on this list.
So, we plan to take more care selecting image(s) and titling articles. How about you?