Not to mention if everyone started doing it, they would just train AI to do it also, and it would only be giving data to train AI with. That's why I think most data poisoning strategies are pointless. One exception might be to try to include a spelling mistake somewhere that doesn't make a comment too confusing, sometihng that could easily be a typing mistake. LLM's are basically spellcheck² and never make spelling mistakes unless explicitly told to or trained that way. If I see a spelling error, I know it's more likely to be human.