this is especially significant around sensitive topics where being careless with words can cause harm.

Harm to ourselves or to others? Estimating expected utility is pretty complicated in cases where we have to weigh up upsetting others (according to the value we directly assign to others' utility in our utility functions), upsetting others (according to the harm we estimate to ourselves through creating enemies) and benefiting ourselves (by rapidly screening off others who we don't wish to interact with). I imagine typical individuals are generally doing all of these calculations at some level in typical interactions and it is not clear to me that we consistently err in any particular direction.