This from the person who got the bots to talk about it...

To the questions re: prompting - how NotebookLM works is you feed it research docs which the agents discuss.

In this case, they got a one pager about them being AI - what I found amazing was how they processed and reacted to it.