Most people have a draft version of a difficult conversation sitting somewhere in the back of their mind. A team member who keeps missing deadlines. A colleague whose communication style is creating friction. A direct report who's been producing work that isn't good enough, and who probably senses it but hasn't been told directly. The conversation that needs to happen keeps getting rescheduled — mentally, if not on the calendar.
This isn't laziness or cowardice, usually. It's that giving hard feedback well is genuinely difficult. You want to be honest without being harsh. Specific without being clinical. Clear about the problem without making the person feel attacked. Most of us haven't been trained to do this, and the stakes feel high enough that the easiest short-term move is to wait for a better moment that never quite arrives.
AI won't write the conversation for you — and if you try to use it that way, the result will be obvious and counterproductive. But it can do something more useful: help you prepare for the conversation in ways that make you more likely to actually have it, and have it well.
The Preparation Problem
Most difficult feedback conversations go sideways not because the person giving feedback is unkind, but because they haven't fully worked out what they want to say before they start saying it. They know something is wrong. They know it needs to be addressed. But the specifics — what exactly the problem is, what a better outcome would look like, how to say it clearly without triggering defensiveness — are still fuzzy when the conversation begins. And when you're fuzzy going in, you tend to either over-soften the message until it doesn't land, or overcorrect and say something blunter than you intended.
The preparation work is where AI is genuinely useful. Not to generate a script, but to help you think more clearly before you open your mouth.
Step One: Get the Problem Out of Your Head and Onto the Page
Start by telling Claude or ChatGPT the situation in plain, unfiltered terms. Not polished, not diplomatic — just what's actually going on as you see it. Something like:
"I need to give feedback to someone on my team. They've been missing deadlines consistently for the last two months — not by huge amounts, usually a day or two — but it's affecting the rest of the team because other people have to pick up the slack. I've hinted at it a couple of times but never said it directly. I'm not sure if it's a workload problem, a time management problem, or something personal. I need to have a real conversation about it but I keep putting it off."
Then ask the AI to help you identify what you actually need to communicate. A prompt that works well: "Help me figure out what the core message is here and what a good outcome from this conversation would look like."
What comes back will usually help you see the situation more clearly — partly because the AI reflects your own thinking back at you with some structure, and partly because the act of writing it out at all tends to clarify what you actually think.
Step Two: Separate the Observable Facts from Your Interpretations
One of the most common ways difficult feedback backfires is when the person receiving it hears a judgment rather than a description. "You're not reliable" is a judgment. "The last three project deadlines were missed by one to three days each" is a description. The first one invites defensiveness. The second one invites a conversation.
Ask Claude or ChatGPT to help you pull apart what you've observed from what you've concluded. A simple prompt: "Looking at what I described, what are the specific observable things I could point to — things the person couldn't reasonably dispute — versus the conclusions I'm drawing from them?"
The AI will typically surface a useful distinction. It might point out that "they don't seem to care" is an interpretation, while "they haven't responded to two follow-up messages about the timeline" is a fact. Starting with the fact gives the other person room to explain — which sometimes reveals information you didn't have. Starting with the interpretation tends to close the conversation down before it opens up.
Step Three: Anticipate How It Might Land
This is where the preparation work gets particularly valuable. Once you have a clearer sense of what you want to say, ask the AI to play the role of the person receiving the feedback — and to respond realistically, not in the most charitable possible way.
"I'm planning to open by saying: 'I want to talk about some patterns I've been noticing around deadlines. Over the last couple of months, deliverables have been coming in a day or two late, and it's putting pressure on the rest of the team.' Can you respond as the person I'm giving this feedback to, including some of the defensive or deflecting responses I might realistically get?"
What comes back is often surprising. The AI will surface responses you hadn't considered — "I wasn't aware this was affecting the team," or "I've been dealing with unclear briefs from the other side," or simply a hurt silence followed by "I didn't realise you felt that way." Sitting with those responses before the actual conversation gives you a chance to think about how you'd handle each one, rather than improvising under pressure in the moment.
This isn't about scripting every response. It's about doing enough thinking in advance that you're not thrown by the first thing that doesn't go according to plan.
Step Four: Check Your Own Framing
There's one more thing worth doing before a hard conversation, and it's the most honest one. Ask the AI to reflect back whether the way you're framing things puts all the weight on the other person.
Something like: "Reading what I've described, is there anything I might be doing — or not doing — that's contributing to this situation? Am I missing anything from the other person's perspective?"
The AI can't know the full picture. But it can sometimes surface things worth considering — whether expectations were communicated clearly, whether the workload is genuinely manageable, whether you've given the person the tools they need. Coming into a feedback conversation having already considered your own role is different from coming in convinced the problem is entirely with someone else. It changes the tone, and the other person usually senses that difference.
What AI Can't Do Here
It's worth being clear about the limits. AI can help you think through what you want to say. It cannot make you say it with warmth and genuine regard for the other person — that comes from you. It can help you identify the observable facts in a situation. It cannot give you the judgment to know when a situation calls for directness and when it calls for curiosity.
There's also a real risk of over-preparing in ways that make you sound rehearsed. If you've spent an hour generating and memorising AI responses to every possible objection, the conversation can start to feel like a cross-examination. The preparation is meant to settle your nerves and clarify your thinking — not produce a script. Once you're in the conversation, set the preparation aside and actually listen.
The other thing AI can't do is make the conversation less uncomfortable. Giving someone feedback they need but haven't asked for is uncomfortable by design. That discomfort is part of what makes it worth something when you do it well. No amount of preparation removes it, and trying to eliminate it entirely often leads to feedback that's so softened it doesn't communicate anything useful.
A Note on Written Feedback
Some feedback genuinely belongs in writing — a formal performance review, a documented concern that's been raised multiple times, something that will form part of an HR record. In those cases, AI can help you draft more directly.
The same principle applies: give Claude or ChatGPT the situation with specifics, ask for a draft, and then edit it heavily to make sure it sounds like you and reflects what you actually want to say. AI-written performance feedback that hasn't been meaningfully edited tends to have a slightly institutional quality — precise, but flat. The person on the receiving end can usually tell.
Review anything AI drafts for you before it goes anywhere official. Not just for tone, but for accuracy. AI tools will sometimes add plausible-sounding specifics that aren't quite right, or smooth over a nuance that actually matters. Read it as if you're reading someone else's draft, not your own.
Starting With One Conversation
If you have a feedback conversation you've been putting off, the most practical thing this article can offer is a small homework assignment. Before you do anything else, open ChatGPT or Claude and write out the situation — honestly, without editing yourself — the way you'd describe it to a trusted friend. Then ask for help identifying the core message and the specific, observable facts.
That twenty minutes of preparation won't make the conversation easy. But it will make it more likely to happen, and more likely to be useful when it does. The conversation you've been postponing probably needs to happen. AI can help you walk into it a little more ready than you would be otherwise.