If I were an algorithm-profiling company I would try to anchor my profilings in the real world (what kinds of people I talk to, about what, what kinds of places I visit, what are the opinions of others on me, etc). LLM garbage would be just to draw voluntary participation in potential surveys.

It takes a particularly paranoid and stubborn individual to make the necessary efforts to consider what kinds of profiling could be done with such anchored data, and even more effort to probe it enough to acquire some knowledge on how it works.