Leading Personas

LLMs take-on personas based on the text they are trying to continue. But LLMs that have been post-trained to act like assistants also have biases about how they treat the user, probably based on the kind of person who would be responding to a given request. In that sense leading questions can directly cause personas. My question is: what kind of personas can be instigated by this response effect, and are they a significantly different set of personas from the ones that a base LLM can take on simply by being conditioned on text that such a persona might say. In other words: are 'responsive personas' a subset of all the base personas, or something else entirely?