Understanding Instagram’s Algorithm

We’ve all been there. You spend hours refining details, getting colors just right, and upload your latest work to instagram only for the post to receive little to no engagement and fade into obscurity. While creating great works is a reward in itself, for creatives, especially freelancers, visibility is key. In order to eat we need to work, and in order to work we need to be seen. Instagram is an important tool for getting new eyes on our work. We can connect with other creatives and clients from all over the world. Unfortunately, Instagram’s algorithm can also work against us, and hurt our ability to be seen by those we want to connect with. If your content doesn’t perform well initially, chances are it won’t be seen by anyone, even those who already follow you.

We’re confident in the quality of work. But if our content is great, then why don’t our posts perform well? There are many factors at play. When is the best time to post? What hashtags should I be using? Instagram tracks various factors like these and they all play a role in the performance of your posts. Those with business or creator accounts can access these insights and better understand these variables. But sometimes the success of our content can feel just downright random. Unless you have a full team working on generating content, it can be hard to post consistently enough to nail down these individual variables and gain any real insight into their bearing on the process as a whole. The inner workings of the algorithm are shrouded in mystery. Many claim to have cracked the code but the developers are notoriously tight-lipped about the matter.

What’s worse is when an update to the algorithm causes a sudden drop in your engagement. If you’re using Instagram, (if you’re reading this you probably should be) you’ve probably noticed fluctuations in likes and comments from time to time. Earlier this year, users reported an app-wide drop in engagement following an update. This left many speculating about the viability of Instagram as a platform for businesses. Beyond just connection or validation, if you depend on Instagram for business, these fluctuations can be scary. A sudden drop in visibility can mean a sudden drop in sales. If even your followers aren’t seeing your content, the situation can feel pretty dire. Instagram’s lack of transparency on the matter has definitely not helped to soften the blow.

Recently, however, a post made on Facebook’s AI blog offered some long-awaited insight into how Instagram recommends content. To be specific, the post details how user’s explore tabs are populated. You’ve likely noticed this before: you come across a photo of a turtle, and it’s funny, so you like it. Over the course of the following days, you start to notice more photos of turtles are popping up on your explore tab. It often seems like a single like can change the course of your feed.

More or less, as we all know, the algorithm tries to deliver more of what we like. Its goal is to keep us using the app, so naturally, the best way to do so is to deliver more content we will engage with. To do so, Instagram’s AI looks at content we interact with and uses a complex series of processes to find and deliver similar content. This list is affected most by interactions. The posts you like, save or comment on are all candidates for extrapolation. Each of these interactions will generate tens of thousands of similar candidates. This number will then be filtered for spam and policy violations to ensure the content is safe. From here 500 of the remaining candidates will undergo a 3-stage ranking process through various neural networks. The number is narrowed from 500 to 150, again from 150 to 50, and one final time, until only the 25 best candidates remain. These are the posts that make up your explore page.

The process isn’t perfect though, and the so-called “Explore” tab wouldn’t be much of an exploration if all of the content looks the same. To counter this, there are also measures in place to prevent posts from the same person from appearing multiple times within a single set of 25. Further measures are also taken to boost the diversity of the content.

If even after all that you still feel a bit left in the dark, you’re not alone. Perhaps the most useful bit of information to come from this blog post is that recommendations are based on similar accounts as a whole rather than individual posts. I suppose this means that having a consistent theme to your account will make it easier for the algorithm to correctly recommend you. If you’d like to read the blog post for yourself, you can find it here.