How Recommendation Algorithms Decide What You See

Open a video app and the first screen is already filled with things you might like. Behind it is a recommendation system that predicts what will keep you watching.

The signals it uses

• What you watch, like, skip or search for.
• How long you stay on each item.
• What similar users enjoyed.
• Time of day, device and location.
• Details about the content itself, such as topic and length.

Two main methods

Collaborative filtering says people like you enjoyed this, so you may too. Content-based filtering says you liked this item, so here are similar ones. Most services combine both, plus machine learning to rank thousands of candidates in a fraction of a second.

What they optimize for

Systems are often tuned for engagement, meaning time spent or clicks. That can reward sensational content and keep you in a narrow bubble of similar material.

Taking back control

• Use the not interested and hide options.
• Clear or pause your watch and search history.
• Follow a variety of sources on purpose.
• Look for a chronological or following-only feed.
• Check ad and personalization settings.

Recommendations can be genuinely helpful, but it is wise to remember that they are designed to hold your attention. Choose some of what you consume on your own.