How to Use AI to Prepare for Difficult Conversations

There's a conversation most of us have been putting off. Maybe it's telling a direct report that their performance isn't where it needs to be. Maybe it's asking a business partner to change something fundamental about how you work together. Maybe it's a family conversation about money, or a health situation, or an old tension that never got resolved. The kind of conversation where you've already rehearsed it in your head a dozen times, and it still doesn't come out right.

One of the more quietly useful things you can do with AI tools is use them to prepare for exactly these conversations — not to script them word for word, but to think through them more clearly than you can manage alone in your own head. I've been doing this for about a year, and it has changed how I approach conversations I used to dread.

Why Your Own Head Is a Lousy Rehearsal Space

When you rehearse a difficult conversation by yourself, you have a problem: you're playing both parts, but you have a vested interest in one side. You imagine how the other person will respond, but you're imagining a version filtered through your own anxiety, your assumptions about them, and your preferred outcome. The conversation you rehearse in the shower is not a neutral simulation. It's a projection.

This is partly why the conversation keeps looping without resolution — your mental rehearsal isn't introducing any genuinely new information. It's just cycling through the same worried territory.

AI tools like ChatGPT and Claude are useful here precisely because they're not you. They don't share your assumptions about the other person, they don't have an emotional stake in the outcome, and they can offer a different angle — including angles you hadn't considered — without the social cost of asking a friend to help you rehearse something personal.

The Basic Setup: Thinking Through What You Actually Want to Say

The simplest version of this is just using AI as a thinking partner before the conversation happens. You describe the situation, explain what you want to achieve, and ask the AI to help you figure out what to say and how.

Here's what that looks like in practice. Say you need to tell a colleague that they've been interrupting you in meetings and it's becoming a problem. You might open ChatGPT and type something like this:

What comes back will usually include a few different ways to open the conversation, some things to consider about tone and timing, and possibly some language that's softer or more direct than what you'd naturally reach for. You don't have to use any of it verbatim — the point is to get outside your own head and see the situation from a slightly different angle.

Claude tends to be particularly good at this kind of nuanced interpersonal guidance. It's thoughtful about social dynamics and usually avoids generating language that sounds wooden or corporate. ChatGPT works well too — it's worth trying both and seeing which output feels more natural to you.

The More Useful Version: Roleplay the Conversation

Once you have a sense of what you want to say, the next step is to actually rehearse the other person's responses. This is where AI becomes genuinely valuable in a way that nothing else quite replicates.

You ask the AI to play the role of the other person and respond as they might — including pushback, defensiveness, questions you haven't anticipated. This is not about being pessimistic. It's about getting yourself ready for the real thing, so you're not thrown off when the conversation doesn't go exactly as you planned.

From here you conduct the conversation as if it's real. You say what you plan to say. The AI responds in character. You respond to that. It keeps going until you've worked through the main beats you're worried about.

The value is in the surprises. The AI will sometimes generate a response you hadn't anticipated — something defensive, or something unexpectedly emotional, or a deflection you didn't have a good answer to. Those moments are gold, because they tell you exactly where your preparation has gaps. Better to find them in a practice run than in the actual conversation.

After the Roleplay: The Debrief Question

Once you've run through the conversation once or twice, there's one more question worth asking the AI before you step away:

This tends to produce genuinely useful feedback. Because the AI was playing the other person, it has a vantage point on the gaps in your argument, the moments where your phrasing was ambiguous, or the questions you didn't have good answers to. Treating it like a debrief after a practice run gets you more out of the exercise than just running through the conversation and closing the tab.

What This Works Well For

This approach is particularly useful for conversations that are high-stakes but not crisis-level. Performance conversations with people you manage. Renegotiating terms with a client or vendor. Bringing up a recurring issue with a family member that nobody wants to name directly. Asking for something at work — a raise, a change in role, a different project — where you want to make a strong case.

It's also useful for conversations where the other person's likely response is hard to predict. If you know someone is going to be defensive, or tends to change the subject, or has a habit of turning things around and making themselves the victim, running through those scenarios in advance means you're less rattled when they happen in real life.

What it's less useful for: conversations that are primarily emotional rather than practical. If someone close to you is grieving, or you're having a real relationship rupture, or the conversation requires genuine human empathy in the room — AI rehearsal might give you words but it won't prepare you for the emotional weight of it. Know the difference.

A Few Things to Keep in Mind

The AI doesn't know the actual person you're going to talk to. It's working with what you've told it, which means its simulation of "how they might respond" is a useful exercise but not a prediction. The real conversation will go differently. That's fine — the goal isn't to predict it exactly, it's to reduce the number of moments where you're caught completely off guard.

Also: don't share more personal information than you need to. You don't need to name the colleague, describe their personal situation in detail, or give identifying details about your workplace. A general description of the dynamic is enough for the AI to be useful, and it's good practice to think about what you're putting into these tools.

Finally, the conversation you rehearse with AI and the conversation you actually have are two different things. Don't walk in with a memorised script — walk in with a clearer head. The preparation shifts you from "anxious and scattered" to "clear about what matters and ready to adapt." That's a real shift, and it makes a real difference.

The conversations you've been putting off don't get easier by waiting. But they do get more manageable when you've already thought them through once — even with an AI standing in for the other person. Sometimes that one rehearsal run is the difference between dreading the conversation indefinitely and just having it.