Machine learning algorithms prove that the best thing you can do for your website is to write well
Behaviour patterns showed text quality account for between 15 to 30% of time users spend on a website. Here is why.
You don’t have to work in a very big organization for very long, before the topic of which content elements perform best on a website comes up.
What makes users stay on a website longer?
Operations, ongoing projects, sales, marketing, and digital services all have their own opinions about what works and what should be prioritized. Not to mention the ability of management to discover the truth with a quick here-and-now answer to what it takes to get the business goals of the website in the bag.
But what is the truth really? At Canecto, we have analyzed data about user behavior in order to find the answer to what is most important for things like how long the users stay on a website.
What we did in our analysis was first to break down all the elements contained in a website: text, photos, links, videos, etc. Then we compared the elements to user behavior on the website.
Next, we ran a number of machine learning algorithms that can identify patterns in large data sets. This enabled us to see the most pronounced relationships between the content of a website and the way users utilize the site.
The better text your write, the longer users stay on your website
And the answer for the visually oriented is not positive, because it is quite obvious that the better text your write, the longer users stay on your website. And this conclusion applies to all types of websites – even for web shops with lots of photos and for text-heavy sites. Other factors are significant as well, but nothing comes close to the important of great text, if you want to keep users on your site.
Behavior patterns showed that the three elements in our analysis that concerned text each could account for between 15 to 30% of time users spend on a website. The three elements are:
- Information level – distribution of content words and empty words.
- Lix readability level – the complexity of the sentence structures on a site.
- Text length – number of words on a page.
It is, in fact, possible to calculate the information level of a text. Many probably wonder, how and why would you do that?
It’s important to relate to the information level of a text, because it is one indication of how precisely it is formulated. Thus, it makes a difference whether or not a text is written concisely.
As a rule, very informative sentences have a tendency to be more objective, whereas less informative texts are often more subjective.
By measuring the information level in a text, you can get an indication of whether it makes a different if a text is formulated objectively and precisely seen in relationship to subjective and less precisely formulated texts.
Lix readability level
This factor describes how easily accessible your sentence structure is. Here, the conclusion is somewhat surprising, but we’ll get to that.
The Lix level is generally very important for how much time users spend on a site, but it’s never quite at the level of the length of the text. The interesting thing is that the texts are generally written too simply on the sites we have analyzed, and that simple sentences make the users leave the site quickly.
None of the sites we looked at had sentence structures complex enough to scare users away from the site. Thus, the conclusion is quite clear: All the sites we looked at would do better to have more complex sentence structure.
One consideration is that users gravitate to texts with more substance, and that they don’t care for overly simplistic text, where they might feel they are being talked down to. This theory is actually supported by the first element: the information level.
Focusing on text length is nothing new. Everyone who works with SEO knows that there are standard lengths to aim for, so Google will give the site a high evaluation. But these types of rules of thumb say nothing about what your readers respond to best from a communications standpoint.
If it is important to you that the message on your pages is received by your readers, you need to know which text length provide the best effect for you and your readers.
Therefore, we calculated which text length performed best for the sites included in our analysis.
Text length matters
And the length of your texts does matter. This factor is consistently among the most significant for how long users stay on a site. But two interesting things jump out at you: The optimal number of words varies greatly from one site to the next, and there is a clear tendency of texts being too short on the sites we looked at.
Therefore, if you are looking at this through the lens of communication, SEO recommendations will do nothing for you. The SEO recommendations probably don’t match the context through which your readers access your website, and you most likely need to write more on your pages than you currently do.
How is it calculated?
The method for calculating the information level is a branch of the part of machine learning science that involves language. This is called Natural Language Processing. In the field, a method has been developed that identifies content words, which are seen in relation to all other words in a text.
The more content words a text has, the more informative it is. Content words are nouns, verbs, and adjectives. This method is implemented in Canecto’s solution, whereby we have the option of calculating the information level of any website we analyze.
Does it matter whether or not I write concisely?
The short answer is yes, it makes a big difference. This factor is consistently among three of the most important elements on websites we have analyzed. Typically, the information level means more than the use of photos, and often it means more than the use of video on a page as well.
We have been very surprised ourselves at the importance of text when it comes to user behavior on a website, and therefore we can only encourage taking this discipline seriously.
If you are unsure, this text has a lix level of 49.1, making it hard to read. The text on this page is 30% longer than the average text on this site, and it is among the top 10% for information level. Perhaps that’s why you made it to the end of the text on this page and that you have spent 2-3 minutes reading it.