Wirebook › Lesson 10 of 18
How does a video app guess what you'll watch next?
You watch one video about dinosaurs and suddenly your whole screen is dinosaurs. It can feel like the app is reading your mind. It isn't. It's counting: what you finished, what you skipped, and what people who click like you watched next. Here's a tiny pretend video app so you can be the one it learns about, and watch exactly what it writes down.
It's not mind reading, it's counting
A video app has millions of videos and room to show you about ten. So a program (an algorithm, which just means a set of steps a computer follows) gives every video a score for you, then shows the top few. The score is built from clues you leave just by using the app. In the machine above there are only three clues: you watched to the end, you skipped, or you said "not interested". Real apps use many more, but the idea is the same.
The clues real apps count
- How long you watch. This is usually the biggest clue. Clicking a video says a little; watching it all the way through says a lot. Skipping after a few seconds counts against it.
- What you do next. Liking, sharing, rewatching, saving, or going straight to another video from the same channel.
- What you search for. Searching for "how to draw a dragon" is a very strong hint.
- What people like you did. If thousands of people who finished the same videos as you also loved one you haven't seen, it's a good bet for you too. This is called collaborative filtering, and it's how an app can suggest something brand new that has nothing in common with your last video. In the machine above, that's the "taste twins".
- The situation. The time of day, whether you're on a phone or a TV, your language and roughly which country you're in.
Why it pulls you down rabbit holes
Most video apps are built to keep you watching, because that's how they earn money from adverts. So the safest guess is "more of what kept you watching last time". Watch a few dinosaur videos to the end and dinosaurs grab most of the slots. Try it above, and watch the rabbit-hole meter. Good apps deliberately mix in a few wild cards, videos outside your usual topics, so you don't get stuck. Tick "Add a wild card" to see one.
Taking back the wheel
- Say "not interested". Most apps have it hidden behind the three dots on a video. It's a much stronger clue than just scrolling past.
- Clear or pause your watch history. Most video apps have both in their settings. Clearing wipes the notebook, so recommendations go back to general popular videos. Pausing means new videos aren't written down. Try both above.
- Feed it something new on purpose. Search for a topic you'd like more of and watch a couple of good ones all the way through.
- Turn off autoplay if you'd rather choose the next video yourself.