Publishers have always had to weigh the risks and rewards of quick news bulletins and deeply-reported longreads. Write too much and you might lose readers who are just looking for the facts. Write too little and you might cut the sections that turn those casual visitors into loyal readers.
While striking this balance is the responsibility of a good editor, one of our responsibilities is to uncover data and insights that make those decisions easier. It’s in that spirit that our Data Science team recently investigated the relationship between word count and engagement.
We sought to find out how Average Engaged Time changes as word count increases and how editorial teams can use this data to optimize their content. More on those findings below.
Average Engaged Time increases with word count, up to a point
In the undertaking of this study, Chartbeat’s data scientists analyzed millions of articles of 10,000 words or fewer that were published between January 2019 and April 2022. With a global network of publishers, languages, and grammar rules in the same dataset, this was not as simple as scraping HTML and calculating engaged time, but our team was up to the challenge and delivered a trove of data illuminating the relationship between article length and engagement.
When we plot Average Engaged Time by word count, two clear patterns emerge:
1. Between 0 and 2,000 words, Average Engaged Time increases as word count increases.
2. Once word count grows beyond 4,000, the variability in engaged time also grows, and the return on additional length is less certain.
In other words, we can confidently say that for articles of less than 4,000 words, the longer the article, the more engaging it will be. Beyond 4,000 words, however, the interval of expected engaged time varies much more widely, and we can no longer conclude that engaged time will increase with word count.
While this doesn’t mean that there is no more engagement to be had beyond this point, it does mean that performance will depend more heavily on how well optimized a page is for engaged time after publication. This data also allows you to benchmark your articles of various lengths against our global averages.
Decoding the binned scatterplot
Since we frequently use bar and line graphs in our research presentations, it’s worth pausing here to explain the graphic above. Rather than plotting millions of articles on one graph, the binned scatterplot uses one dot to represent the Average Engaged Time of all the articles published at a given word count. For dots that are intersected by a line, this denotes the expected range of engaged time for an article of that length. From 0 to 2,000 words, the variation is only hundredths of a second and the dot is essentially also the line. Around 10,000 words, the range visibly stretches to almost 9 seconds.
If we look at the 6,000 word mark as an actual example on the graph, the dot tells us that the Average Engaged Time for all articles of 6,000 words is 80 seconds. The line through the dot then shows us that the least engaging articles of 6,000 words receive about 77 seconds of engaged time, and the most engaging receive about 83 seconds.
You must be logged in to post a comment.