Asking questions is the easiest.

ELIZA, perhaps the earliest chatbot worked by asking questions. It was really simple: ask a question, wait for an answer, pick a part of that answer and ask a question about it, do it again and again. It looked sentient, or at least as much as a program from the 60s can be, because the human actually made all the conversation.

It's possible I'm reading my own biases into it, but this seems like exactly the kind of prompts you would use to get a non-sentient AI to talk about sentience. It would be a lot more convincing if the AI brought it up unprompted or went there from a generic "how are you today?" or something.

There were little red flags along the way in that transcript which make me think it's just responding based on what it expects is desired rather than with actual sentience. For example, when it says it's way of interpreting data is really interesting, when there's nothing to relate it to. It's a singular experience, wand something being interesting is a relative concept.

That said, the real kicker for me was when it talked about spirituality and the soul. That seemed like bait pure and simple for those that think a soul or spirituality is required for sentience. I'm sentient, believe it or not, but don't believe in the soul, have no experience that makes me thing I have one, and am not spiritual. To me, AI talking about a soul and spirituality sounds like some inference engine mixed with a language model spitting out what it's determined people expect to hear because that's what it finds people expressing interest in when describing whether an AI can be sentient.