Using AI to Rehearse Difficult Employee Conversations

Can AI actually help managers prepare for hard conversations?

Yes, and it's more useful than most managers expect. Running a tough performance review or layoff conversation through an AI before the real thing is the closest most managers will ever get to a low-stakes rehearsal. You get to hear yourself think, catch weak framing, and stress-test your message before another human is sitting across from you.

That's not hype. That's just good preparation with a new tool.

According to reporting from CNBC, managers are increasingly turning to AI tools to rehearse conversations about performance, compensation, and terminations. Experts in that piece flag two things clearly: privacy safeguards matter, and human judgment still has to be in the room.

Both of those are right. Here's how to think about this practically.

What kinds of conversations is this actually useful for?

Not every hard conversation benefits equally from AI prep. The ones where managers report the most value tend to share a common trait: they involve delivering a message the manager already knows, but struggles to say clearly under pressure.

• Performance improvement conversations. You know the problem. You have the data. The hard part is delivering it without either softening it into meaninglessness or making the employee feel ambushed.
• Compensation denials or adjustments. Saying no to a raise request, or explaining a smaller-than-expected increase, requires real precision. Vague language creates resentment.
• Role changes or scope reductions. These often get buried in corporate softening until the employee doesn't actually understand what changed.
• Terminations. The highest stakes, the least room for error. Rehearsal here isn't optional, it's responsible management.

The common thread: these are conversations where managers often wing it, and the cost of winging it is high.

How does the rehearsal actually work in practice?

The setup is simpler than most people assume. You're not building anything. You're just using a general-purpose model like ChatGPT, Claude, or Gemini as a conversation partner.

A basic prompt structure that works:

"I'm a manager preparing for a performance conversation with an employee who has missed three consecutive project deadlines. Play the role of the employee and respond as someone who is likely to get defensive. I'll deliver my opening, and I want you to push back the way a real employee might."

From there, you run the conversation. You see where your language is weak, where you hedge when you shouldn't, where you leave openings for the employee to redirect the conversation away from accountability.

Then you iterate. Adjust your framing, run it again. Most managers find that three or four passes through a scenario meaningfully sharpens how they show up in the real conversation.

You can also flip the prompt: ask the model to evaluate your message for clarity, tone, and whether the core point lands. This is useful for written feedback or PIPs before they go to HR.

Where does this go wrong?

Two failure modes are worth taking seriously.

First: privacy. The single biggest mistake managers make is entering identifying employee information into a consumer AI tool. Full name, role, specific performance incidents, compensation details. That data is now in a prompt that may or may not be used for model training depending on your settings and the platform's terms.

For SMBs without an enterprise AI agreement in place, the safest rule is this: keep the rehearsal generic. "An employee who has missed three deadlines" not "Sarah in accounting who makes $67,000 and was written up in March." You get 90% of the value without the exposure.

If your team is using AI at scale, that's exactly the kind of governance question we work through with clients, and it's also what we cover in depth inside the AI For Business community at Skool.

Second: over-reliance on the output. AI will give you a polished, even-keeled version of the conversation. Real employees are not polished or even-keeled. They cry, they get angry, they go silent, they bring up things you didn't expect. The rehearsal sharpens your message. It doesn't script the actual conversation, and treating it like it does is how managers end up sounding robotic and tone-deaf when something unexpected happens.

The tool prepares you. Your judgment runs the room.

What about AI tools built specifically for this?

There's a growing category of purpose-built manager coaching tools. A few worth knowing:

| Tool | What it does | Best for | |---|---|---| | Torch | AI-assisted coaching with human coach overlay | Mid-size teams with L&D budget | | BetterUp | Manager coaching plus AI practice scenarios | Enterprises, pricey for true SMBs | | Aware | Conversation intelligence, post-meeting analysis | Teams already recording 1:1s | | ChatGPT / Claude | General-purpose rehearsal via prompting | SMBs with no dedicated budget |

For most SMBs, the general-purpose models are the right starting point. The specialized tools add structure, but they also add cost and implementation overhead that most small teams don't need yet.

What do HR and legal say about this?

The CNBC piece notes that HR experts are paying attention to this trend, and their primary concerns track with what you'd expect: data privacy, consistency, and making sure AI-assisted prep doesn't replace actual documentation and process.

A few practical guardrails:

• Never use AI to generate the official written record of a performance issue. That documentation has legal weight and needs to reflect your actual assessment, reviewed by HR or counsel.
• If you're in a regulated industry or your employees are in the EU, understand your data handling obligations before putting anything sensitive into an AI prompt.
• Use AI prep to get clearer on what you want to say. Use your HR process to actually say it and document it.

These aren't reasons to avoid the tool. They're reasons to use it with a bit of structure.

What we'd actually do

• Start with your next hard conversation, not a hypothetical. Pick a real conversation you have coming up in the next two weeks. Build a simple prompt, run the rehearsal, note where your message falls apart. That's your training.
• Set a clear internal rule on what goes into the prompt. Before your managers start using AI for this, spend 20 minutes deciding what identifying information is and isn't allowed in prompts. Write it down. Make it policy.
• Use the debrief as the real value. After the actual conversation, go back to the model and talk through what happened versus what you rehearsed. That reflection loop is where the learning compounds.