How the Algorithm Knows You
Think about this for a second
You opened your phone. You didn't search for anything. You didn't say anything out loud to anyone. And the first three videos are about exactly what's been on your mind.
The usual explanation is: "my phone is listening to me."
It isn't. The real explanation is more unsettling: it doesn't need to. What you do says more than anything you'd say out loud.
Why nobody explained this
Because in this system you aren't the customer — you're the product. There's no class called "understand the system designed to keep you scrolling." But you spend a few hours a day inside it.
And it isn't even a secret: TikTok, Instagram and YouTube have each published, in their own words, roughly how their recommendation systems work. Almost nobody read those posts. You're about to.
How it actually works
1. Signals: the questions nobody asks you
The algorithm never asks "what do you like?" If it did, you'd lie — everybody does. Instead it measures what you do:
| Signal | What it tells the system |
|---|---|
| Watch time | Did you finish it, or bail after 2 seconds |
| Rewatch | Did you come back to it (a very strong signal) |
| Hesitation | Did you linger half a second longer on one image while scrolling |
| Like / comment / save | Explicit intent — but weighted less than you'd think |
| Share | The most expensive signal: you put your reputation on it |
| What you skip | Skipping fast is data too |
Notice how few of these are conscious decisions. You choose to hit like. You don't choose to pause half a second longer on something. That's exactly why the system is hard to fool: it's reading your reflexes, not your opinions.
2. You aren't a list — you're a coordinate
Picture it this way: every video and every user gets placed as a point on a huge multi-dimensional map. You're a coordinate on that map. The system asks:
"What did other people sitting near this coordinate watch all the way through?"
This is called collaborative filtering, and it's powerful. It doesn't need to know you. It needs to know the 40,000 people who resemble you. That crowd predicts your next click before you've noticed the impulse yourself.
That's why the algorithm sometimes feels like it knows something about you before you do. What it actually knows is the average of people like you.
3. Explore and exploit
The system doesn't only serve you what it already knows. Every so often it tests you: something you've never shown interest in lands in your feed. That's not a glitch — it's a deliberate exploration move. Your reaction gets measured:
- You skipped → that door closes
- You watched → that door opens all the way
Which means one moment of curiosity can reshape your feed for weeks. A video you watched out of curiosity at 2 a.m. tells the system "this topic works." The system reads it as a statement of identity.
4. The thing being optimized isn't your happiness
This is the part that matters most. These systems are not optimized for "the user feels good." They're optimized for "the user stays."
Those two are frequently not the same thing. Content that makes you angry, anxious, or envious gets watched more. Because it gets watched more, it gets distributed more. Nobody wrote down a goal saying "let's make teenagers feel worse" — but the metric they chose produces that outcome anyway.
In Mozilla's 2021 study with 37,000 participants, the overwhelming majority of videos people described as regrettable weren't searched for. They were recommended.
What it does to you
The rabbit hole. Show interest in a topic and the system offers you denser, more extreme versions of it. Fitness → extreme dieting. Studying → sleep-hostile hustle culture. Feeling low → darker content. Amnesty International's 2023 research found that accounts engaging with low-mood content saw their feeds measurably darken within hours.
A narrowing world. Your reality and your friend's reality are no longer the same. You open the same app and see completely different worlds. Something feeling like "everyone's talking about this" doesn't mean everyone is.
Drifting calibration. If everything you see is the most extreme, most polished, most successful example, you lose track of what normal looks like. This isn't a willpower problem — your brain builds an average from the examples it's given. Feed it a broken sample and it produces a broken average.
What you can do about it
You don't have to delete the app. The moves below turn the system's own mechanics back toward you.
1. Use "Not interested" aggressively. Silently scrolling past is a weak signal. Long-pressing and choosing "Not interested" is far stronger. Do it ten times in a week and your feed visibly shifts.
2. Watch your hesitations. Since watch time is the strongest signal, watching something that upsets you just to see what it says is a vote for it. Skip the things that make you feel bad, fast.
3. Bring search back. Using search instead of the feed puts you back in the driver's seat. If you want to learn something, go look for it — don't wait for the feed to deliver it.
4. Separate accounts, separate apps. If the account you learn on and the account you goof off on are different, neither one contaminates the other.
5. Audit your feed on a schedule. Once a month, spend five minutes: look at your last 20 pieces of content and ask — "did I choose these?" If the answer is no, go back to step 1.
6. Know what you're feeding. The simplest rule there is: watch time is a vote. Whatever you give time to, you're voting for. Including the things that make you angry.
Try it: The 20-Click Experiment
Do this on one account this week. It takes 20 minutes and you'll see the result within a couple of days.
- Pick an app. Jot down the first 10 things in your feed right now (topic names are enough).
- Deliberately watch 20 videos all the way through on a topic you have zero interest in. (Rock climbing, antique restoration, whatever — the more unrelated the better.)
- Close the app. Wait 24 hours.
- Open it again and write down the first 10 things.
How many changed? Most people find more than half their feed has drifted within 24 hours.
The point isn't to wreck your feed. The point is to see with your own eyes that this system isn't modelling your personality — it's modelling your last 20 actions. Once you've watched that happen, the feed telling you "this is who you are" never lands the same way again.
Summary card
- Your phone isn't listening. It's measuring behaviour — which is more accurate anyway.
- The strongest signal isn't likes, it's watch time. And it isn't conscious.
- The system doesn't know you. It knows thousands of people who resemble you.
- The optimized metric isn't your wellbeing — it's your time on the platform.
- One moment of curiosity can reshape your feed for weeks.
- Watch time is a vote. Even for the things that make you angry.
Sources
- TikTok Newsroom — How TikTok recommends videos #ForYou, 2020
- Adam Mosseri / Instagram — Shedding More Light on How Instagram Works, 2023
- Mozilla Foundation — YouTube Regrets: RegretsReporter Report, 2021
- Amnesty International — Driven into the Darkness: How TikTok's For You Feed Encourages Self-Harm and Suicidal Ideation, 2023
- Wall Street Journal — The Facebook Files, 2021
Related lessons: Your Digital Footprint · Verify a News Story in 3 Steps · The Social Media Comparison Trap Related tool: Algorithm Simulator — watch your feed narrow in 10 clicks