You probably think of yourself as a complex, irreducible person. You have moods, opinions that shift, contradictions you can't fully explain. You surprise yourself sometimes. You defy easy categorisation. Most of us hold this self image without really examining it—and why would we? It's both accurate and comforting.
And all of that is true. But from the perspective of a recommendation system, you are also remarkably, almost boringly predictable. The gap between how we experience ourselves and how we appear to machines is one of the defining tensions of life in a digital society. And bridging that gap—understanding what the algorithm sees when it looks at you—is one of the most practically useful things you can do.
You are unpredictable to yourself because you experience your inner life. Algorithms don't need your inner life. They just need your behaviour—and behaviour, it turns out, is highly predictable.
What the Algorithm Sees
Every click, pause, swipe, search, and share generates a signal. Over time—and it doesn't take long—these signals accumulate into a behavioural profile of extraordinary resolution. The algorithm doesn't know what you're thinking. It doesn't need to. It can predict what you'll do next based on what millions of people who behaved like you in the past went on to do.
Netflix famously estimated that over 80% of what people watch on its platform comes from algorithmic recommendation—not from search, not from friend recommendations, not from editorial picks. People believe they choose what to watch. In practice, the algorithm surfaces what they watch, and they choose from that narrow selection. The autonomy is real but operates within a funnel built entirely by the machine.
Spotify's Discover Weekly playlist is another example worth examining. It was introduced in 2015 and immediately became one of the platform's most popular features—not because Spotify's curators were exceptional, but because the algorithm had been trained on the listening behaviour of hundreds of millions of users and could identify patterns that no human curator could see at that scale. Users describe it as feeling uncannily personal. What it actually is, is uncannily accurate pattern matching. The intimacy is a product of scale, not understanding.
Why This Is Both Useful and Dangerous
The same mechanism that powers a genuinely useful Netflix recommendation also powers micro targeted political advertising. The same model that introduces you to a song you didn't know you'd love can identify with clinical precision which anxieties make you most susceptible to manipulation.
The most disturbing case study in recent memory is Cambridge Analytica. In 2018, it emerged that the political consultancy had harvested the Facebook data of up to 87 million users without their consent and built psychographic profiles that were then used to target political messaging. The methodology wasn't science fiction—it was straightforward behavioural modelling, the same kind platforms use every day for commercial advertising. What made it alarming was the application: using algorithmic understanding of individual psychological vulnerability to influence democratic elections. The algorithm's power to know you was turned against your autonomy as a citizen.
It's not that algorithms are malicious. It's that they're indifferent. They'll optimise for engagement just as efficiently whether that engagement is joy or outrage, curiosity or fear.
Algorithmic recommendation doesn't care about the quality of the outcome for you. It cares about whether you continue engaging. Those two things are sometimes aligned—and sometimes dramatically opposed. Outrage is extremely engaging. Anxiety is extremely engaging. The content that most effectively exploits those emotions will always outcompete more measured, more accurate, more nuanced content in an engagement-optimised system.
The Mirror You Didn't Ask For
The most unsettling thing about being known by an algorithm is what it reveals about your own self knowledge. You might not realise how much time you spend consuming certain kinds of content. You might not see the emotional patterns in your own usage—the correlation between stress and certain browsing habits, between loneliness and the types of videos you watch late at night.
I keep thinking about a detail from Netflix's early personalisation research: the company found that the thumbnail image that motivated someone to click on a film was a reliable indicator of their psychological state at that moment. Someone feeling vulnerable might click on a thumbnail showing a character looking distressed. Someone feeling optimistic might click on one showing connection and warmth. The same film. Different emotional entry point. The algorithm knows which version of you showed up that evening—often before you do.
Understanding this isn't about paranoia, and it's not about trying to deceive the system. It's about cultivating a clearer picture of your own behaviour—because the clearer that picture becomes, the better positioned you are to decide which patterns you want to keep, and which ones the algorithm is simply feeding back to you because they keep you on the platform. The algorithm is a mirror. What you do with what you see in it is still entirely your choice.
Using the Knowledge to Your Advantage
Here's the genuinely useful flip side: if you understand how recommendation systems work, you can shape what they show you. Actively seeking out content that represents where you want to go—not just where your past behaviour points—trains the algorithm in a different direction. Think of it as deliberate cultural investment. Pause on the long form journalism. Finish the documentary. Save the thoughtful analysis rather than just reacting to the outrage. These are inputs into a system that responds to them. You have more agency than you think—if you use it consciously.
