A practical walkthrough for using ChatGPT or Claude to think through what you'll say, how someone might react, and where you're likely to freeze, before you ever sit down to have the actual conversation.
Sixty-nine percent of managers say they’re often uncomfortable communicating with employees, and manager engagement has fallen from 31% to 22% globally since 2022, according to Gallup, largely because the job has gotten harder while the support hasn’t kept pace. Using AI won’t make that discomfort go away, I want to be upfront about that. What it does is change how ready you are once you’re walking in. Use it before the conversation to nail down the actual issue instead of just the feeling behind it, draft an opening line or two, and role-play the pushback you’re already dreading. Don’t ask it to write you a script you read word for word, and don’t paste identifying details about a real employee into a public AI tool, that part isn’t optional. The conversation itself is still entirely yours to have.
In almost every corporate workshop I run, I ask managers to name the conversation they’re currently avoiding, and the room goes quiet for a second before someone admits it out loud. It’s usually one of three things: the missed deadlines that keep happening despite three gentle reminders, a role that’s about to be eliminated, or the tension between two people on the team that everyone can feel in every meeting. Most of them have rehearsed an opening line in the shower and lost their nerve by the time they sat down at their desk.
That discomfort isn’t a personal failing, and I say that to every cohort I train. Harvard Business Review reported on a survey by Interact and Harris Poll that found 69% of managers say they’re often uncomfortable communicating with employees, and 37% specifically dread giving direct feedback when they expect the employee to respond badly. Atana’s research on why managers procrastinate these talks found something even more specific: 74% agreed it’s harder to raise an issue when they’ve done something similar themselves, and 63% said feeling nervous makes it harder to even start.
Meanwhile the job itself has gotten less forgiving. Gallup’s State of the Global Workplace found that global manager engagement dropped from 31% in 2022 to just 22% in 2025, and pinned lower manager engagement as the single biggest driver behind the overall workplace slump. The managers I coach are stretched thinner than they were even two or three years ago, expected to run more of these conversations with less support for how to prepare.
Let’s be honest: no amount of prep makes telling someone their role is being cut, or that a colleague complained about them by name, feel comfortable. I’ve never once seen a manager walk out of that kind of conversation feeling good, and I don’t think they should. What prep actually buys you is walking in clear-headed instead of winging it on nerves, and that’s the exact gap AI is useful for closing.
Treat AI as a thinking partner for the hour before the conversation, not a participant in it. Used well, it earns its keep in three places. It helps you separate the facts from your own frustration. It drafts language that’s direct without tipping into harsh. And it plays out how the other person might respond, which is the piece that actually keeps their reaction from blindsiding you.
Some tools are now built specifically for that third piece. LinkedIn Learning’s Role Play with AI-powered Coaching lets you talk or type through a realistic scenario with a customizable AI personality that responds dynamically, then gives you feedback on your communication style afterward. If you don’t have access to a dedicated tool, a general assistant like ChatGPT or Claude does the same job with the right prompting, which is what most of this guide walks through.
What AI can’t do matters just as much, and I’d rather say that plainly than oversell this. It doesn’t know your employee’s history, their tone on a normal week, or what’s actually going on in their life right now. It can’t read the room once you’re in it, and it definitely can’t soften a layoff into good news no matter how cleverly you phrase the prompt. If you’re building out a broader system for how you use AI as a manager day to day, How to Use AI to Be a Better Manager covers where it fits across the rest of the job, not just the hard talks.
A useful rule of thumb: if a prep step involves you thinking out loud before the conversation, AI can help. If it involves the employee, a real document with their name in it, or anything happening during the actual talk, it’s yours alone.
Most difficult conversations go sideways in the first thirty seconds, and in the workshops I run, it’s almost always for the same reason: the manager hasn’t separated the actual issue from how it feels to deal with it. “He’s just not a team player” is a feeling. “He’s missed the last three sprint deadlines and hasn’t flagged it in advance either time” is an issue you can actually raise.
Before you touch AI at all, write down, in plain language, what specifically happened, when, and what outcome you need from the conversation. Then hand that to a chat tool and ask it to pressure-test your framing. A prompt like this works well: “Here’s a situation with an employee: [describe what happened, factually, with dates if you have them]. Help me separate the objective facts from my own interpretation or frustration, and flag anywhere I’m assuming intent I can’t actually prove.”
This step catches things you won’t catch on your own, mostly because you’re too close to it. I’ve had managers do this exercise live in a workshop and visibly deflate a little once they see it laid out, five sentences of interpretation and one actual fact. AI is genuinely decent at spotting the difference between “missed the deadline” and “doesn’t care about deadlines,” and forcing that distinction before you’re in the room keeps you from opening with an accusation you can’t back up.
If you can’t state the issue in one factual sentence with no adjectives in it, you’re not ready to have the conversation yet, no matter how good your talking points sound.
This is the part that actually changes how the conversation goes, and it’s also the part most managers skip because typing back and forth with a chatbot pretending to be your employee feels a little silly. I’ll be honest, the first time I had a manager I was coaching try this, it didn’t land the way either of us expected, he thought the AI’s “defensive” employee was almost comically over the top and nearly wrote off the whole exercise. We tightened the prompt to match how his actual employee talks and reacts, and the second attempt was genuinely useful. So do it anyway, but go in expecting to adjust the prompt once before it clicks. Ask AI to play the employee, based on how you’d describe their likely reaction, and have the conversation with it before you have it with the real person.
A prompt that works: “Play the role of an employee I need to give tough feedback to. Here’s the situation: [context]. They tend to get defensive and change the subject when confronted directly. Respond the way you think they realistically would, including pushing back or getting emotional if that fits, and don’t make this easy on me. I’ll type what I’d actually say to you.”
Go back and forth for a few exchanges. You’ll notice fast whether your planned opening line lands as intended or comes across harsher, or softer, than you meant. If the AI-as-employee gets defensive, you get to practice staying calm and redirecting to specifics instead of freezing, which is exactly what I see happen to most first-time managers the moment a real employee pushes back.
LinkedIn Learning’s version of this does something similar with voice, letting you actually speak the conversation out loud and get feedback on tone afterward, which some people find more useful than typing. Either way works. The point is rehearsal, not the specific tool.
Run the role-play at least twice: once assuming the employee stays calm and cooperative, and once assuming they get defensive or upset. You need a plan for both, because you genuinely don’t know which one you’ll get.
Managers tend to over-prepare their opening line and under-prepare for everything that happens after it, and I see this same imbalance in basically every cohort I run. The first thirty seconds are honestly the easy part. It’s the follow-up questions, the silence, or the “why didn’t anyone tell me sooner” that actually derails people.
Ask AI to generate the hardest questions or objections the employee is likely to raise, given the situation, and draft calm responses to each one in advance. Try: “Given this situation, list the five toughest questions or pushback points this employee might raise, including ones that might catch me off guard. For each, suggest a short, honest response that doesn’t overpromise or get defensive.”
Write your three hardest anticipated questions on a sticky note, not a full script. You want prompts to glance at, not a document you’re reading from while someone watches you read.
The five-step approach above holds for almost any difficult conversation, but the specifics shift depending on what you’re actually walking into. Here’s how the prep adjusts for four situations managers ask me about most.
This is the one where tone matters most and there’s the least room for error. Use AI to help you get the facts and next steps airtight (severance details, timeline, what happens to their email and access) before the conversation, not during it. Practice saying the actual words “your position is being eliminated” out loud in the role-play, because vague language here reads as evasive and makes it worse, not kinder. If you’re also managing who takes on the responsibilities left behind, How to Use AI for Succession Planning covers that separate, following conversation.
Get specific with dates before you open a chat tool. “You’ve been late a lot” invites a debate about what “a lot” means. “You missed the March 3rd, March 17th, and April 2nd deadlines” doesn’t. Ask AI to help you frame the conversation around a forward-looking plan, not just a list of failures, since the goal is a changed pattern, not a confession. If the real issue turns out to be an overloaded plate rather than a motivation problem, How to Use AI to Delegate Work Without Losing Control is worth reading before this conversation, not after it.
This one has three sides in the room even when only two people are talking: theirs, theirs, and yours. Role-play both directions separately, since each person will likely respond differently to the same framing. Ask AI to help you draft a version of the conversation that stays focused on specific, observable behavior between them rather than either person’s character, which is the fastest way this type of talk turns into a shouting match.
Missed promotion, a project getting cancelled, budget cuts to their team. These get underprepared because they don’t feel as high-stakes as a termination, but they’re often where trust actually gets built or broken. Use AI to help you draft an honest answer to the question that’s coming regardless of the specifics: “what does this mean for me going forward.” Vague reassurance here reads as spin, and most employees can tell.
A few patterns show up consistently when managers start using AI to prep for hard conversations, and most of them come from treating the prep tool like it should do more of the work than it should. I’ve seen a version of this play out often enough across cohorts that I can usually predict which mistake someone’s about to make just from how they describe their prep.
The test I actually use with the managers I coach: if the AI conversation left you feeling more ready to sit down with the real person today, it worked. If it just felt like a productive-feeling way to put that off, it didn’t do its job.
Yes, for the preparation itself, and I tell every manager I coach through a layoff to actually do this rather than winging it. Using AI to organize the facts, draft clear language, and role-play likely reactions before a termination conversation is a reasonable use of the tool. Keep any real employee details generic or anonymized, confirm every logistical detail (severance, timeline, benefits) with HR and legal since those facts have to be exactly right, and never let AI-generated language substitute for your own judgment about tone once you’re actually in the room.
Avoid full names, specific performance review content, medical or personal details, basically anything that would let someone identify the employee if that chat history were ever seen by another person. Describe the situation the way you’d describe it to a friend outside the company: enough detail to get useful help, nothing that ties it back to one identifiable person. I say this especially to managers on smaller teams, where even a job title plus one detail is enough to narrow it down to exactly one person.
No, and I’ll be straight with managers about that upfront. AI role-play gives you a plausible reaction based on how you describe the person, which is genuinely useful for practicing your own composure and language under a bit of pressure. It isn’t a forecast, though. Real people say things a role-play won’t anticipate, so treat the rehearsal as training for flexibility, not a script for exactly what’s going to happen in the room.
Twenty to thirty minutes is usually enough, and that’s roughly what I tell managers to budget: a few minutes getting clear on the factual issue, one or two role-play exchanges, and a quick list of the toughest questions you’re expecting. Go much longer than that and you start tipping into over-rehearsing, which tends to make people sound scripted instead of present once the real conversation actually starts.
Preparing means you walk in with a clear sense of the facts, your key point, and a rough idea of how you’ll respond if things get tense. Scripting is different, and it’s the trap I warn managers about most: you’re reciting fixed lines no matter what the other person actually says. I’ve watched this go wrong in real time, a manager stiffens up the second the employee says something that isn’t in their notes, and the whole conversation stalls while they scramble to find their place again. Prep keeps you present. A script just gives you further to fall when the conversation doesn’t go how you planned it, and most employees can tell within a sentence or two which one they’re getting.
I checked every statistic in this piece against its original source before using it: Harvard Business Review’s coverage of the Interact/Harris Poll survey, Atana’s published research on why managers avoid hard conversations, Gallup’s State of the Global Workplace report (I pulled the manager engagement numbers directly off Gallup’s own page, not a secondhand summary), and SHRM’s 2026 State of the Workplace research. I also weighed each stat against what I’ve actually seen training managers through this exact scenario, in corporate workshops and one-on-one coaching, because a number that doesn’t match what I’m seeing in the room is one I want to double-check before I use it. The LinkedIn Learning AI coaching example comes from a live university HR communications page describing real employee use of the tool, not a press release. Nothing here is estimated, and nothing is reworded from someone else’s summary of the original research.
Weekly tips. Real examples. Practical help for busy professionals.
We care about your data, check out our privacy policy.