Sometimes they are not speaking to you at all, but rather to your listeners. And the larger your pool of listeners the more likely you'll encounter arguments in bad faith because it turns out that if the goal is persuasion then good-faith arguments don't scale.
A tangent here is that on private-public platforms like twitter "your listeners" could be an entirely separate set from the people you actually have contact with. Algorithms that signal-boost opinions out of their actual social circle become essentially propaganda posters for a host of varying in and out groups.
My personal opinion is that this is a misaligned incentive that social platforms should correct structurally, rather than via governance or policies. And pushing the corrective actions back down to the individual is a cop-out.