Humans are really good at adapting their mode of interactions to different contexts. They do that all the times in different environment: work, family (nuclear and extended), friends, strangers…
They can also adapt and learn to interact with an ai using conversational interface provided they are given some cues about the context of this particular AI: they will learn to interact with a very narrow focused AI, or a more generalist one.
What doesn’t work is if the Ai pretends to be what it’s not, a human: uncanny valley.