How Does an App "Know" What You’ll Like Next?

It can feel eerie when a video app suggests exactly the kind of video someone was just thinking about, or a shopping site recommends something genuinely useful. This isn’t mind-reading — it’s a recommendation system, a type of AI that predicts what someone might like based on patterns in behavior, not by understanding personal thoughts at all.

One common approach is called collaborative filtering: the system looks at what large numbers of people with similar past behavior enjoyed, and assumes a new person with similar tastes so far will probably enjoy similar things next. If thousands of people who watched Video A also watched Video B, the system suggests Video B to anyone who just watched Video A — purely a numbers game, no understanding of what either video is actually about required.

A second approach looks at the content itself — comparing the features of things someone already liked (genre, keywords, style) to find similar new items, regardless of what other users did. Most real recommendation systems blend both approaches together, along with dozens of smaller signals, to make a final suggestion.

Recognizing this helps explain some odd behavior people notice: recommendations can feel spookily accurate because they’re built from huge amounts of collective behavior data, but they can also get stuck suggesting very similar things repeatedly — because the system is optimizing for "what worked with similar behavior before," not genuinely understanding when someone wants something completely different for a change.