Facebook may be used by billions of people, but the content posted isn’t always of the highest quality, and that remains true even when you discount the fake news the social network is trying to stamp out. But steps are being taken to improve content with the latest focus being “engagement bait.”

Facebook’s Henry Silverman, Operations Integrity Specialist and Lin Huang, an Engineer, have penned a blog post detailing how much users dislike spammy posts and how the social network intends to combat them.

By spammy posts, Facebook means anything that attempts to goad a user into an interaction including clicking the Like button, the Share button, or commenting. This is referred to as engagement bait, and the perfect example is when a post starts with a phrase such as “Like this if you’re…” and then uses some term that applies to a large percentage of the viewers, e.g. their star sign. The more people who interact with these posts, the higher the engagement is and therefore the post enjoys an artificial engagement boost on the network.

Of course, Facebook wants to tackle engagement bait through automation, so engineers have been working to identify the different types and teach a machine learning model how to spot them. It breaks down into five distinct categories: vote baiting, react baiting, share baiting, tag baiting, and comment baiting.

If any post is detected as falling into one of those categories, the chance of them appearing in a News Feed drops. If such posts appear repeatedly on Pages, Facebook will trigger “stricter demotions” and Page-level demotion to further discourage the practice. At the same time, Facebook promises that any posts genuinely asking for help will not suffer demotion.

Facebook’s focus when implementing this new system is authenticity. Any post that includes content deemed “spammy, sensational, or misleading” will suffer. Combine that with Facebook’s existing demotion rules for clickbait headlines, clickbait phrases, and low-quality web links, and engagement bait posting could result in content never being seen.