How does the LinkedIn publication algorithm work?

Published by Guillaume Portalier on

How does the LinkedIn algorithm define what content to display to you? At what moment? Behind the “black box”, there is a fairly basic logic. I’ll explain it to you.

Show you the right LinkedIn post

LinkedIn is a social network. And like the majority of social networks today, their business is based on advertising.

The business model of advertising is to sell the attention of users to brands looking to promote their services, products or content.

We can therefore imagine that if they sell attention, the more they have, the greater their potential gains. This is the whole model of social networks like LinkedIn: getting more of your attention.

A social network must therefore make you spend the most time on its site or application.

To do this, they set up different mechanisms:

  • External triggers: notifications and emails to bring you back to the network
  • The use of cognitive biases (such as “Fear Of Missing Out” or “Fear of missing something” that keep us coming back, also considered an “internal trigger”)
  • Random reward: you never know what you will find by scrolling a news feed. From time to time a particularly relevant post gives us a surge in dopamine (the hormone of pleasure). Like any good primate, we crave more of this dopamine by scrolling on other content.

I do not go into detail with these psychological mechanisms. I invite you to read “Hooked” by Nir Eyal, a best seller on the subject.

What particularly interests us is the 3rd point and how the content display algorithm tries to optimize it by showing us relevant posts.

Emotions and engagement rate on a post

The vast majority of social media audiences are silent. It is also one of the major pivots of Twitter: they initially thought that everyone was going to Twitter. By observing their data, they realized that the majority of people were looking to follow content without creating or interacting with it.

By observing viral posts on social networks we notice two things:

  • They often have a strong emotional impact (this is why videos of kittens are always so popular on the web)
  • They have a high engagement rate (ratio of comments + likes / views)

These two observations are directly linked: content with a strong emotional impact has a greater chance of making us react (whether this impact is negative or positive in the process).

However, as we said before, LinkedIn (and other social networks in general) aims to make you feel intense emotions, which will release hormones in your body.

But the LinkedIn algorithm being (still today) incapable of feeling these emotions and determining the quality of content, it will be based on the emotions felt by people before you.

And is it measured? Via the engagement rate of course!

Comments, likes and views and the LinkedIn algorithm

The higher the proportion of comments and likes / views, the more the content will be shown to a large audience.

The process works in the form of a spiral:

  • When you publish your content, it’s displayed to a small part of your network, which acts as a sample. The sample size varies depending on the optimization of your post
  • The algorithm observes what is the engagement rate on this sample, ie the rate of people who like and comment. This initial rate is essential because it largely influences the final reach of the post. Or rather the reverse: the algorithm considers that if there is no strong initial commitment, the content is not relevant. It is said to be measured between the first hour and the first three hours of content life.
  • Based on this initial engagement rate, the algorithm widens the audience by prioritizing the people in your network, but especially the people in the network of those who engaged on your post, considered as “similar profiles”, who could therefore also be interested in your content. (Yes we tend to like the same things as people around us, culture, network, so to feel similar emotions on similar content).
  • If the engagement rate continues to be similar, the post continues to grow in the number of views. This is what allows content with a high engagement rate to go viral.

There is an important criterion to take into account:

  • As you can imagine, posting a comment is an act that requires much more involvement than a like. They do not have the same “emotional weight” at all, and therefore not the same weight for the algorithm. I would say that the ratio is between 10 and 20 (1 comment = 10 to 20 likes).

Other factors to take into account

We can say that the engagement rate far exceeds all the other criteria for distributing a publication.

However, certain other criteria will be able to influence (especially downward), the scope of a post:

  • The number of hashtags used. The hashtag is like a “category” of the post for LinkedIn. If there is none, the algorithm cannot categorize it. If there are too many, it will consider that it is excessive and that you try to make it appear in all the categories.
  • Network size: although it seems that this factor influences very little, it would seem that the size of your network can play on the initial sample of the post. (A study is planned to invalidate or confirm this hypothesis).
  • Readability: a view on LinkedIn is just a person who passes on your post, without necessarily stopping. If your post is not readable, the likelihood of engagement is very low.
  • Views on your latest publications. If your last posts have had a lot of views, LinkedIn will tend to enlarge the initial sample, considering that the quality of the post is more likely to be good.
  • Outbound links. As we said, LinkedIn wants to keep users on the network to monetize them. Including links out of LinkedIn in your posts increases the chances of the user leaving LinkedIn. Publications containing an outgoing link are therefore devalued.

As the algorithm is a black box, these explanations are based on observations made by the LinkedIn community over time. The algorithm is likely to evolve and the influence of certain criteria remains unknown.