what I do is instruct ai how to do end of thought in my prompts meaning if it seems like the user isnt done talking i tell my ai output "-" or "..." or "". So it doesnt say anything if it seems like the user isnt done talking. The issue with this is if the ai guess wrong though then it will only talk until the user speaks again and only why to counter that is using vapi idle message settings. So it best to use it in situations where you know the ai interrupts a lot(like when user is spelling out emails and etc).