By establishing initial guidelines and a regular process of evaluating the guidelines, changing them over time.
I get that there’s a core question of “what even is harmful?” that many will debate but “I know it when I see it” was the answer for the debate over obscenity and it broadly worked for most people. Sometimes a best effort is better than an endless debate over an final answer.
History? i.e. track record of prior harm by type of speech? If similar content has demonstrably caused harm before, then chances are high decent it's likely to cause harm again.
ASIDE:
It occurs to me that the kinds of discussions being had in the modern era, have their counterparts from earlier times. The arguments for and against the freedom to publish holocaust denialism has strong parallels to today's disagreements around freedom of speech, though they clearly predate social media networks. Interestingly, though global approaches to holocaust denialism (in the west) range from strong disapproval to outright bans, the strain of thought has continued to survive, albeit in far lower numbers. Would those numbers be vastly different in a world where those ideas were more freely permitted to flourish?
I get that there’s a core question of “what even is harmful?” that many will debate but “I know it when I see it” was the answer for the debate over obscenity and it broadly worked for most people. Sometimes a best effort is better than an endless debate over an final answer.