There is a specific kind of dread that comes with knowing you have a hard conversation ahead of you. A performance issue you need to raise with someone on your team. A request to your manager that you're not sure will land well. A disagreement with a colleague that has been building for weeks. A boundary you need to set with a client. The conversation sits in the back of your mind, and every time you think about it, you run through versions of it — what you'll say, how they'll respond, how it might go wrong.
Most people walk into those conversations having done this kind of mental rehearsal, but alone. The problem with rehearsing alone is that you tend to rehearse your own perspective very thoroughly and the other person's perspective much less so. You know exactly what you want to say. You're less prepared for what they might say back.
AI is genuinely useful here — not to script the conversation for you, and not as a replacement for thinking it through yourself, but as a structured way to pressure-test your thinking before you walk in. Used well, it's a bit like running the conversation by a thoughtful colleague who has no stake in the outcome and will give you an honest read.
Start by Getting Clear on What You Actually Want
The most useful thing AI can do before you open any conversation is help you articulate your goal clearly — which turns out to be harder than it sounds. Most people going into a difficult conversation have a muddle of objectives: they want to be heard, they want something to change, they want to preserve the relationship, they want to express how they've been feeling. These don't always point in the same direction, and arriving unclear about what you're actually trying to achieve is one of the most common reasons hard conversations go badly.
Try opening ChatGPT or Claude and describing the situation plainly. Don't craft it — just write it out the way you'd explain it to a friend. Then ask directly: "What do you think I'm actually trying to achieve here, and are those goals compatible?" The response won't always be revelatory, but the act of reading your own situation described back to you from a slight distance often surfaces contradictions you hadn't noticed. In my own work, I've found that this step alone — five minutes of typed thinking before a difficult meeting — changes how clearly I walk in.
Use AI to Argue the Other Side
This is where AI preparation becomes most practically useful. Once you've described the situation, ask the AI to represent the other person's perspective as charitably as possible. Not to tell you you're wrong — but to lay out the reasonable version of the other side.
You might frame it like this, in ChatGPT or Claude:
"I need to talk to a colleague about missing deadlines on a shared project. I've been frustrated because it's affecting my work. Can you give me the most reasonable version of why they might be struggling or why they might not see the problem the same way I do?"
What comes back will often include things you already knew — but laid out explicitly, in order, with language that isn't charged by your own frustration. That's useful. It means you walk into the conversation having already considered the version of events where you're not entirely right, which tends to make you more flexible and better at listening when you get there.
The other version of this exercise is to ask the AI to play devil's advocate with your planned approach. Describe how you intend to open the conversation and ask: "What could go wrong with this framing? What might the other person hear that I'm not intending to say?" AI is reasonably good at this because it has no investment in your approach being the right one.
Draft Your Opening, Then Stress-Test It
The opening of a difficult conversation matters more than most people realize. How you introduce the topic sets the other person's defensive posture for everything that follows. An opening that feels accusatory — even unintentionally — can make the rest of the conversation an uphill climb.
Ask the AI to help you draft a neutral, specific opening. The key word is specific: vague openings ("I wanted to talk about how things have been going") signal that something is wrong without giving the other person any way to orient themselves, which tends to spike anxiety and defensiveness. Specific openings ("I wanted to talk about the timeline on the Henderson project — specifically the last three milestones") tell the person exactly what the conversation is about, which is usually less alarming than the unknown.
Once you have a draft opening, ask the AI to read it as the other person would. Try:
"Here's how I plan to open this conversation: [your opening]. How might this land for someone who doesn't know it's coming? Does anything in it sound more accusatory than I intend?"
This is a simple check, and it catches things. Phrases that feel neutral to you — "I've noticed that," "it seems like," "I've been concerned about" — can carry more weight than you realize when you're the one receiving them. Having a second read, even from an AI, is better than none.
Prepare for the Responses You're Dreading
Every difficult conversation has a moment you're most afraid of. Usually it's a specific response — the defensiveness, the denial, the counter-accusation, the tears, the dismissiveness. Most people rehearse up to that moment and then stop, because thinking about it is uncomfortable. That means they arrive having prepared for the first half of the conversation and not the second.
Ask the AI to simulate the responses you're most worried about, and then help you think through how you'd handle each one. You don't need a scripted reply — scripts tend to fall apart in real conversations anyway. What you need is to have thought it through enough that you're not caught completely off guard.
For example:
"If the person responds defensively and says the delays weren't their fault — that they were waiting on me for information — what's a calm, non-escalating way to respond that keeps the conversation moving forward?"
Claude tends to be particularly good at this kind of nuanced interpersonal scenario — its responses in these cases are usually thoughtful and calibrated rather than generic. ChatGPT works well too. Neither will give you a perfect answer, but both will give you something to react to, which is more useful than staring at the ceiling the night before.
What AI Can't Do Here — and Why That Matters
It's worth being clear about the limits, because they're real. AI doesn't know the person you're talking to. It doesn't know the history of your working relationship, the dynamics of your team, the thing that happened six months ago that still affects how this person hears feedback. It will give you generically reasonable responses based on how these situations tend to go — which is useful as far as it goes, but it's not the same as insight from someone who actually knows the players.
AI also can't replace the emotional preparation the conversation requires. Knowing what you want to say and being genuinely ready to say it calmly are different things. If you're still very angry, or very anxious, the best-drafted opening in the world won't carry you through. The preparation AI can help with is cognitive — clarifying your goals, checking your framing, thinking through responses. The emotional piece is still yours to work through.
And there's one more thing: over-preparing can make you rigid. The goal of all this preparation isn't to arrive with a script — it's to arrive with enough clarity and flexibility that you can actually listen, adjust, and respond to what the other person actually says, rather than what you imagined they'd say. The preparation should make you more adaptable, not less.
A Simple Pre-Conversation Routine
If you want a repeatable process, here's one that takes about fifteen minutes and covers the main bases:
Open Claude or ChatGPT. Describe the situation in plain language — who's involved, what's happened, what you need to address. Ask the AI what your core goal seems to be, and whether there are any tensions in what you're trying to achieve. Then ask it to give you the most reasonable version of the other person's perspective. Draft your opening line and ask the AI to flag anything that might land harder than you intend. Finally, name the two or three responses you're most dreading and ask for help thinking through each one.
That's it. You don't need to spend an hour. The point is to arrive having thought about the conversation from more than one angle — which is harder to do alone than it sounds, and easier with a thinking partner that has no stake in how it goes.
The conversation is still yours to have. AI doesn't change that. What it can do is make sure that when you sit down across from that person, you've thought it through well enough that you're not just reacting — you're actually present.