Data scientist here! In addition to the data points others have mentioned, there is actually a lot more data available than you would think in the form of metadata. We call the process feature engineering - essentially building a set of inputs that help determine an output, or prediction. How long you spend in the app, how long you stay on a screen before changing, how long you view a TikTok before swiping, which of the default settings you change, into what, all of this is used in machine learning models to help build a more accurate advertiser profile for you. Even if you don’t volunteer data about yourself, your behavior in a way informs on you, even if you don’t realize it. Through inference, a machine learning model could accurately deduce your age based on your behavior, for example.