The key is to accuse everyone of being an LLM. Those who don't react are bots. Those that fight the charge no matter how much its levied are also bots, but with better programming. Those that complain at first but give up when too much effort is required are the real humans. Any bot able to feel frustration is cool.

Maybe a reasonable approach would be that people could flag posts with a "probably AI" button to eventually trigger a "bot test" for that account (currently, the "score 5 in this mini game" type seem pretty clanker proof). If they pass, their posts for the hour, week, whatever result in a "not AI" indicator when someone clicks the "probably AI" button.