It depends on the definition of “most personal”. I guess it's relative to what each person considers most personal.
To give you some examples of things I think most people would consider highly personal: religious views, sexual orientation, even the fact that the wider datasets were being used to infer attributes in people related to their psychological disposition, which they may or may not have known or wanted to be inferred. If you are inferring, for example, some sort of psychological neurosis in a person, that may feel very sensitive and very personal, even if that data wasn't necessarily collected as such but rather inferred from the other data that was collected.
The issue is not just what direct observations were acquired in datasets but also how less intimate observations can be transformed into inferred information about things that would be quite personal.