4 November 2026
Why your “quick scroll” always turns into 20 minutes
You open an app for a “quick look”… and suddenly you’re deep into obscure woodworking, medieval cooking techniques, and raccoons stealing cat food.
So how does a recommendation algorithm figure you out so quickly? The answer lies in machine learning, behavioural data, and large-scale prediction systems.
Collaborative filtering: finding patterns in behaviour
Most recommendation systems start with a technique called collaborative filtering.
Instead of analysing what content is, the algorithm focuses on what people do with it. If users who watch Video A also tend to watch Video B, the system assumes the two are related.
This creates a huge user–item matrix across videos, posts, and songs, where machine learning models predict what you’re most likely to engage with next based on collective behaviour patterns.
Every click is a signal
Modern recommendation engines go far beyond simple viewing history. Every interaction becomes a data signal, including:
- Watch time
- Pauses and rewinds
- Likes, shares, and comments
- Skips or instant scroll-past behaviour
- Even scrolling speed
All of this feeds into ranking models, often powered by neural networks, which estimate the probability that you’ll engage with each piece of content. The system then ranks thousands of options and serves the highest-scoring results.
Embeddings: turning you into maths
To scale this process, platforms use embeddings – numerical vectors that represent both users and content.
Each video or post is converted into a vector based on features like topics, visuals, audio, and text. Your behaviour builds a corresponding vector that represents your interests.
When these vectors sit close together in mathematical space, the algorithm concludes: this is something you’ll probably enjoy.
Your feed is more revealing than you think
Your TikTok “For You Page” isn’t just entertainment – it’s a behavioural mirror. It reflects what you actually watch, not what you say you like.
These systems are so effective they can sometimes infer interests you haven’t explicitly expressed yet, purely from engagement patterns.
Which is why your feed can feel a little… uncomfortably accurate.
Prediction over intuition
Recommendation algorithms don’t “understand” content like humans do. Instead, they run an enormous statistical prediction engine to guess what you’ll do next — and they’re getting very good at it.
So next time you open an app for “just five minutes”, remember: you’re not just browsing content… you’re training an algorithm.
Want to know more? Kat, our Craig’n’Dave Lesson Hacker, explains all.
For more Lesson Hacker videos, check out the Craig’n’Dave YouTube playlist HERE.
Visit our website to explore more cutting-edge tech news in the computer science world!
