You may enjoy learning more about Differential Privacy. https://machinelearning.apple.com/2017/12/06/learning-with-p...
And WRT differential privacy, if enough data is captured and associated with a single identifier, then it's sufficient to get a pretty good idea what the user actually does. That's the point of differential privacy in the first place, since it indicates that a statistically meaningful amount of the data is valid.
For example, if my user id has five visits out of 1000 DP recorded visits to google.com, it's unlikely that I actually visited google.com. However, if there are 200 recorded visits, it can be safely said that yeah, I intentionally visit google.com.