Why rehearsing a conversation works
Most difficult conversations go wrong in the first 30 seconds or at the first pushback. Rehearsal helps with both: you find an opening that does not trigger defensiveness, and you have already heard the objection once, so you respond instead of react. Athletes, pilots and negotiators train this way for the same reason.
Simulation also reveals your own habits. Many people discover they over-apologize, bury the main point in long explanations, or back down at the first sign of emotion. Seeing this in a transcript is far easier than noticing it in the moment. Once you spot the pattern, you can practice one change at a time, such as stating your point in the first sentence or pausing after a defensive reply.
How to set up a realistic AI simulation
Give the AI three things: who the person is (role, relationship to you), how they communicate (direct, conflict-avoidant, quick to anger), and what they want or fear. Add one or two things they are likely to say. The more specific you are, the more realistic the simulation.
Ask the AI not to make it easy. A counterpart that agrees immediately teaches you nothing. Start at medium difficulty, then ask for the hardest realistic version, the one you secretly fear most.
A structure for hard conversations
A reliable structure: state the purpose in one sentence, share one specific observation, ask for their view and actually listen, then agree on one next step. Avoid opening with a verdict ("you have been careless") or stacking several complaints at once.
After each run, ask the character to step out of role and rate your clarity, tone and listening. Then rerun only the part that went badly. Two or three focused rounds are usually enough to feel much calmer in the real conversation.