Realize that the state of the art in chatbots is still abysmal. The biggest mistake you can make is to try to do something bizarre, in which case both a human and an AI might reasonably respond in a bizarre manner. Instead, just engage in a normal conversation and resist any significant deviations proposed by your correspondent. That should easily let you identify anything operating at the current published state of the art.
If you encounter something well beyond state of the art, I think the next threshold would be a directed inquiry into a aesthetic decision: "What's your favorite [movie|book|artist|meal]? Why? So would you say you never like...? Are there aspects of that you dislike...?" etc.