Every platform you use is running a real-time model of who you are, what you fear, and what keeps you scrolling. Here's what that actually looks like and why it matters.
Every platform you use is running a real-time model of who you are, what you fear, and what keeps you scrolling. Here's what that actually looks like and why it matters.
When people talk about algorithms, they usually frame it as a convenience story. The algorithm learns what you like and shows you more of it. Music you'll enjoy. Articles you'll find interesting. Videos that match your taste. Sounds helpful.
The problem is that "what keeps you scrolling" and "what's good for you" are two entirely different things, and the algorithm is only ever optimising for one of them.
Every major platform — Instagram, TikTok, YouTube, X, Facebook — runs on an engagement model. The core metric is time on platform. More specifically: how long do you stay, how often do you come back, and how much do you interact?
The algorithm's job is to maximise that number. And it is very, very good at it.
What the algorithm learned — not from theory, but from billions of data points — is that certain emotional states keep people on the platform far more effectively than others. Curiosity works well. Outrage works better. Fear is excellent. Anxiety about social status is exceptional. These aren't editorial choices made by a person. They're emergent patterns that the system discovered because they work.
The result is a content feed that is, quite literally, calibrated to find your psychological pressure points and lean on them.
The model isn't just tracking what you watch and what you skip. It's tracking how you watch it.
On a platform like TikTok, the model notes how long you paused on a video before swiping, whether you rewatched any part of it, whether you scrolled away quickly or slowly, how your behaviour in the next few minutes changed after viewing it. It notices if you searched for something similar afterward. It runs this analysis continuously, updating its model of you in real time.
This model is extraordinarily granular. It knows things about your current mood that you might not consciously register. It knows which topics you slow down for even when you don't interact. It knows which kinds of content you consume when you're tired versus when you're bored versus when you're anxious. And it uses all of that to decide what to show you next.
The goal isn't to make you happy. The goal is to keep you there.
The filter bubble — the idea that personalised feeds narrow the information you're exposed to — has been widely discussed. The concern is that if you only see views that match your own, you lose the ability to understand people who think differently.
That's real. But it understates the problem.
The deeper issue isn't just narrowing — it's amplification. The algorithm doesn't just show you content that matches your existing beliefs; it shows you increasingly extreme versions of those beliefs, because extreme content generates more engagement. A person who shows mild interest in health content gets served increasingly intense health anxiety content. A person who engages once with a political opinion gets served more heated, polarised versions of that opinion over time.
The model pulls you toward edges because edges generate reactions, and reactions are the data the model feeds on. You don't choose to radicalise. The feed does it incrementally, one recommended video at a time.
At this point in the article, you might be thinking: fine, I'm aware of this now, so I'm safe.
You're not. Knowing that a system is manipulating your attention doesn't make you immune to it. It might reduce the effect at the margins, but the fundamental mechanism still operates. Your curiosity still triggers when you see a hook. Outrage still rises when the content is designed to provoke it. The emotional responses the algorithm exploits aren't conscious choices — they're automatic, physiological reactions.
The former president of Facebook's investor relations, Roger McNamee, once described what the platform built as "a system that exploits the brain's dopamine reward circuits in the same way that slot machines do." He helped build it. He was describing it from the inside.
You cannot willpower your way out of a slot machine by understanding how it works. You have to put it down.
You cannot opt out of personalisation while using these platforms. That's not an option they offer. But you can change the inputs the model receives, which changes what the model returns to you.
The most effective intervention is behavioural: stop pausing on content that makes you feel bad. Don't linger. Scroll past quickly. The model will learn. This sounds trivial but it's the actual lever — the algorithm's model of you is built from your behaviour, and behaviour is the one thing you can control.
Secondly, use search and direct navigation rather than the feed. When you go directly to accounts you chose, you're curating. When you let the feed curate for you, the algorithm is curating toward engagement, not wellbeing.
Finally, build in time where the platform simply isn't open. Not because willpower is a reliable strategy, but because the model can't learn from you when you're not there.
These systems are not neutral. They were designed, by very talented engineers, to capture as much of your attention as possible for as long as possible. That's not a conspiracy — it's a business model, openly stated in every investor earnings call.
You don't have to stop using the internet. But treating these platforms as neutral tools — like a search engine or a map — is a mistake. They have goals for your behaviour. It helps to have goals of your own.
When people talk about algorithms, they usually frame it as a convenience story. The algorithm learns what you like and shows you more of it. Music you'll enjoy. Articles you'll find interesting. Videos that match your taste. Sounds helpful.
The problem is that "what keeps you scrolling" and "what's good for you" are two entirely different things, and the algorithm is only ever optimising for one of them.
Every major platform — Instagram, TikTok, YouTube, X, Facebook — runs on an engagement model. The core metric is time on platform. More specifically: how long do you stay, how often do you come back, and how much do you interact?
The algorithm's job is to maximise that number. And it is very, very good at it.
What the algorithm learned — not from theory, but from billions of data points — is that certain emotional states keep people on the platform far more effectively than others. Curiosity works well. Outrage works better. Fear is excellent. Anxiety about social status is exceptional. These aren't editorial choices made by a person. They're emergent patterns that the system discovered because they work.
The result is a content feed that is, quite literally, calibrated to find your psychological pressure points and lean on them.
The model isn't just tracking what you watch and what you skip. It's tracking how you watch it.
On a platform like TikTok, the model notes how long you paused on a video before swiping, whether you rewatched any part of it, whether you scrolled away quickly or slowly, how your behaviour in the next few minutes changed after viewing it. It notices if you searched for something similar afterward. It runs this analysis continuously, updating its model of you in real time.
This model is extraordinarily granular. It knows things about your current mood that you might not consciously register. It knows which topics you slow down for even when you don't interact. It knows which kinds of content you consume when you're tired versus when you're bored versus when you're anxious. And it uses all of that to decide what to show you next.
The goal isn't to make you happy. The goal is to keep you there.
The filter bubble — the idea that personalised feeds narrow the information you're exposed to — has been widely discussed. The concern is that if you only see views that match your own, you lose the ability to understand people who think differently.
That's real. But it understates the problem.
The deeper issue isn't just narrowing — it's amplification. The algorithm doesn't just show you content that matches your existing beliefs; it shows you increasingly extreme versions of those beliefs, because extreme content generates more engagement. A person who shows mild interest in health content gets served increasingly intense health anxiety content. A person who engages once with a political opinion gets served more heated, polarised versions of that opinion over time.
The model pulls you toward edges because edges generate reactions, and reactions are the data the model feeds on. You don't choose to radicalise. The feed does it incrementally, one recommended video at a time.
At this point in the article, you might be thinking: fine, I'm aware of this now, so I'm safe.
You're not. Knowing that a system is manipulating your attention doesn't make you immune to it. It might reduce the effect at the margins, but the fundamental mechanism still operates. Your curiosity still triggers when you see a hook. Outrage still rises when the content is designed to provoke it. The emotional responses the algorithm exploits aren't conscious choices — they're automatic, physiological reactions.
The former president of Facebook's investor relations, Roger McNamee, once described what the platform built as "a system that exploits the brain's dopamine reward circuits in the same way that slot machines do." He helped build it. He was describing it from the inside.
You cannot willpower your way out of a slot machine by understanding how it works. You have to put it down.
You cannot opt out of personalisation while using these platforms. That's not an option they offer. But you can change the inputs the model receives, which changes what the model returns to you.
The most effective intervention is behavioural: stop pausing on content that makes you feel bad. Don't linger. Scroll past quickly. The model will learn. This sounds trivial but it's the actual lever — the algorithm's model of you is built from your behaviour, and behaviour is the one thing you can control.
Secondly, use search and direct navigation rather than the feed. When you go directly to accounts you chose, you're curating. When you let the feed curate for you, the algorithm is curating toward engagement, not wellbeing.
Finally, build in time where the platform simply isn't open. Not because willpower is a reliable strategy, but because the model can't learn from you when you're not there.
These systems are not neutral. They were designed, by very talented engineers, to capture as much of your attention as possible for as long as possible. That's not a conspiracy — it's a business model, openly stated in every investor earnings call.
You don't have to stop using the internet. But treating these platforms as neutral tools — like a search engine or a map — is a mistake. They have goals for your behaviour. It helps to have goals of your own.
