The Invisible Curator of Your Online Experience
Every major online platform — Facebook, Instagram, TikTok, YouTube, LinkedIn, Twitter/X, Google Search — uses algorithmic systems to decide which content from the available pool each user sees, in what order, and for how long. These decisions are made billions of times per day based on models trained to optimise specific outcomes (engagement time, ad clicks, retention) rather than outcomes the user would necessarily choose if they were deciding consciously.
Most users experience the result of these algorithmic decisions without any understanding of how they’re made — which makes it impossible to evaluate whether the experience is serving them or serving the platform’s interests. Understanding the basic logic of how recommendation algorithms work, what they’re optimising for, and how that optimisation relates to your interests as a user, produces a more informed relationship with the platforms that shape a significant portion of what you see and think about.
What Algorithms Are Actually Optimising For
Social media recommendation algorithms are optimised primarily for engagement — the actions users take (likes, shares, comments, watch time, click-throughs) that signal that content kept the user’s attention. Platforms sell advertising, and advertising revenue is proportional to the time users spend on the platform and the attention they give the ads interspersed with content. The algorithm’s incentive is to keep users on the platform as long as possible.
The critical distinction: high engagement and high value to the user are not the same thing. Content that produces strong emotional reactions — outrage, anxiety, envy, tribalistic satisfaction — generates high engagement while producing subjective wellbeing effects that researchers consistently document as negative. The algorithm that maximises engagement does not maximise the user’s experience of their time as well-spent; it maximises the platform’s advertising inventory.
How Personalisation Creates Filter Bubbles
Recommendation algorithms create personalisation by modelling what each user is most likely to engage with based on their past behaviour. As the model learns your engagement patterns, it increasingly shows content similar to what you’ve already engaged with. This produces a self-reinforcing loop: the content you see is increasingly tailored to your existing interests and beliefs, while content that challenges those interests and beliefs is progressively deprioritised because your engagement with challenging content is lower.
The filter bubble effect — the experience of seeing primarily content that confirms existing beliefs and interests rather than challenging them — is partly the result of this algorithmic personalisation and partly the result of users’ own selection behaviour (people share and engage with content that confirms their views more than content that challenges them). The algorithm amplifies the user’s existing selection tendencies rather than creating filter bubbles entirely independently of user choice.
What the Algorithm Can’t Override
Users have more influence over their algorithmic experience than they typically exercise. The recommendation algorithm responds to explicit signals: unfollowing accounts whose content you don’t value, actively selecting ‘not interested’ or ‘see less of this’ on content you don’t want more of, using platform features that allow following specific topics rather than accounts, and engaging actively (not just passively scrolling) with content you genuinely value. These signals recalibrate the model toward content you’ve indicated you want.
The most powerful recalibration signal is search and direct navigation rather than algorithmic discovery: deliberately searching for specific topics, subscribing to specific creators, or navigating directly to content you want to see rather than accepting what the algorithm surfaces gives you more direct control over the information environment than passive algorithmic consumption.
The Informed User’s Approach
The user who understands algorithmic curation approaches their online experience more as a design choice than as a passively received experience. The choice of which platforms to use and how much (the algorithm’s reach is proportional to time spent), which follows and subscriptions to maintain actively versus to prune, and whether to use algorithmic discovery or direct navigation for different types of content all shape the information environment in ways that passive use doesn’t.
The most practically useful single practice: once per quarter, review the accounts you follow on each platform and remove those whose content doesn’t consistently add value to your feed. This audit recalibrates the algorithm’s starting model for what you want to see and reduces the accumulated follows from years of adding accounts without corresponding removal. The resulting feed is more signal and less noise — and the algorithm learns from the improved signal.
