I'm thinking that we can't reliably sus out what's human and what's AI from here on out, and I don't think there's any good way to enforce something like a human emdash. The way we're going you won't be able to verify unless users scan their iris or a 3 letter agency vouches for them first.
Talking to someone in real life.
Of course that's taking the piss, but realistically that is sort of the answer: Having access to a side-channel through which you can identify that a person is in fact a person, coupled with the trust that they won't try to sleight you by making you waste your attention. Honestly with how much LLM drivel is being generated and the sort-of renaissance of the personal website, blog, and RSS, I wouldn't be surprised if some kind of consensus-trust-based network for human authors were established.