It's pretty trivial to finetune an LLM to output posts that are indistinguishable.