That would be a 26 billion dollar question, and one I would very much love to solve one day! :)
I believe that your example is fairly simplistic to capture the crux of the issue here. The metadata associated with posts often contain features that are quite revealing, but not necessarily the kind of data that the source would wish to be revealed. E.g. I have noticed multiple times that the number of cold emails I have received from recruiters is higher immediately after I update my LinkedIn profile, leading me to believe that the last-update timestamp is a feature that the LinkedIn search engine may be relying on to rank results. While my evidence is purely empirical, it isn't a stretch to imagine that it would be a reasonable thing to do, given that most users normally update their profiles when they are about to start searching for new opportunities.
On the one hand, this is a an excellent product that allows recruiters to reach targets that exhibit behavior associated with active job-seekers, resulting in better connections. It results in a win-win situation where the recruiter gets a pretty good return on his/her investment, and the target receives the attention/information they were looking for with their update. False positives in this situation result in a few unsolicited emails/unwanted attention.
On the flip side, this information can be repackaged to present a manager with a graph plotting the probability of an employee quitting. While this is a perfectly good product, no employee would every use a service that might reveal their future plans ahead of a time of their own choosing. Furthermore, false positives here can have a significant impact and LinkedIn may not remain in business for long if word of this product gets out.