Social media platforms don’t share all the information about their algorithms. This makes it very difficult to fully understand what does well and what doesn’t on each platform. I’ve tried my best by researching through engineering blogs and company interviews to collect these insights:
1/ First-Stage (Content-based filtering): This approach analyzes post content such as hashtags, descriptions, and keywords to determine the content a user likes and serve that user more of it.
2/ Social graph analysis: This approach leans on a user’s current network to share more of their posts and the posts they enjoy with the user. It assumes users are more apt to like the things their network likes.
3/ Algorithms don’t care about people, they care about profits. The platforms do that by focusing on the intent of keeping users on the platform as long as possible, like TikTok makes 17% of kids scroll the platform ‘almost constantly’.
4/ Algorithms always interpret what the content is, and where it belongs. They aren’t the audience, algorithms aren’t providing you feedback on your quality of content. That’s the job of the audience.
Marketers need to rethink strategy when the audience doesn’t respond like no comments or DMs originated from content. Not when a post fails to perform.
5/ LinkedIn: While the majority of factors remain similar to Instagram with the platform using Author activity, content and behavior. What stuck out for me was them hinting at optimizing their system to factor creator-side optimization, their plan to reward consistent creators sharing quality content on the platform.
6/ Pinterest: Pinterest factors Old pins to show you while you are offline, using data about your last active session. When you are online, Pinterest tracks your ongoing activities and if a new pin is uploaded that might interest you, they show you that pin to keep you engaged.
That approach from Pinterest completely explains why having a good landing page, caption or visual identity on the platform matters. Pinterest knows they aren’t getting flooded with content, so they look for more evergreen content and pins that cater to a changing user behavior.
7/ Algorithms prioritise content diversity: Because there is much more activity from Authors of the content, IG makes sure that the content from one single creator doesn’t get bundled together on your feed.
8/ Algorithms make predictions about content based on recent data about your content and content around you as mentioned in the ranking explanation by Meta.
We usually talk about how our new post will perform, but your current post’s performance also relies on how your past 3-5 posts did. Because algorithms make predictions about content distribution and follower relevancy.
Every Interaction counts, that means the change in follower behavior counts when they predict how much engagement your post needs today to get pushed further also happens.
IG’s CEO also confirmed this insight that most people don’t reach most followers when they post because of the user activity on that day. That’s why content that appeals to a broad audience is always better than content only focused on your followers.
9/ TikTok has always had a very targeted algorithm, often creators with different interests disappear. But TikTok has been making changes like promoting DMs, Stories and Reposts. Because the changes of Interests on TikTok are fast, you should always have effective CTAs and advise your audience to always leave a follow. That will help you reach them again.
10/ ML algorithms take a lot more data into consideration when making recommendations. Seemingly insignificant things such as time spent reading comments may affect the algorithm.
That’s why many marketers ask businesses to engage in comments
11/ Formats matter, each time a platform comes up with a new organic or Paid Ad format. They intend to give it a push for better adoption by brands and marketers.
Like IG did with reels, jumping on a new format is always beneficial for better growth.
I partnered with ML expert at Google to write about social media algorithms, these were short points I could share on reddit. If you would like to read the full breakdown with sources, let me know.