Your manager is rehearsing the bad news with AI before they call you in
A new wave of AI coaching tools lets managers role-play difficult conversations, from pay freezes to layoffs, before the real thing. Here's what that means for workers on both sides of the desk.
Key points
- A July 2025 survey by The Predictive Index found that 42% of managers know what they want to say in a tough conversation but struggle with how to say it.
- Only 20% of managers prefer to prepare for a difficult conversation entirely on their own, without any framework or coaching support.
- Tools such as Obi (The Predictive Index), Penny (Xactly) and Posh let managers rehearse layoffs, performance reviews and pay talks with an AI stand-in before the real meeting.
- HR professionals warn that managers should never type an employee's real name or personal details into a public AI tool, citing privacy and legal risk.
- Workers who sense a tough conversation is coming can use free tools like ChatGPT or Claude to prepare their own responses.
Before you get called into that meeting about your performance or your job, there's a decent chance your manager has already had the conversation. Just not with you.
A crop of AI coaching tools now lets managers rehearse difficult workplace talks with a chatbot standing in for an employee. The manager types or speaks, the AI pushes back and gets defensive, and the manager tries again until the delivery lands. Think of it as batting practice for hard conversations.
Why do managers need this?
Most of them aren't as confident as their bosses think. A July 2025 survey by The Predictive Index, a company that makes an AI coaching chatbot called Obi, found that 74% of chief executives believe their managers handle tough talks with confidence. Only 42% of those managers agreed. The same survey flagged three blind spots: delivering criticism without crushing someone, guessing how an employee will react, staying objective when things get emotional.
"Spontaneity is not a goal of a high-stakes work conversation," said Emily DeJeu, a business communication professor at Carnegie Mellon University's Tepper School of Business. Preparation is.
How does the role-playing actually work?
A manager logs into a tool, such as Penny from HR software firm Xactly or the training platform from Posh, and sets the scene. They might describe a scenario involving a habitually late employee whose behaviour is affecting the team. The AI then plays that employee, cycling through anger one round, tears the next.
The feedback after each round is where the real learning happens, according to Anthony Belluccia, senior product scientist at The Predictive Index. One manager kept leading with the bad news before providing any context, which read as cold. Once flagged, he learned to open by acknowledging the employee's contributions before reaching the difficult part.
Tools can also catch filler words like "um" and "so", warn when a manager is talking too much, or flag if the conversation drifts toward legally risky territory. A human coach isn't available at midnight on a Sunday. These tools are.
We've tracked workplace AI tools across 25 stories since mid-July, and the pattern is consistent: the productivity gains are real, but so are the ways people over-trust the output.
What should workers know?
First, the privacy warning is real. Managers should never enter an employee's real name, salary figures or personal details into a public AI tool. Behaviours described anonymously are fine; anything identifiable is not. Employer-approved internal systems may differ, but managers should check before typing anything sensitive.
Second, this cuts both ways. Workers who sense a tough meeting is coming can use free tools like ChatGPT or Claude, the AI assistant from Anthropic, to rehearse their own responses. Thinking through your answers in advance tends to produce calmer, clearer conversations. It's the same drill, just from the other chair.
Human coaches are still worth keeping in the mix. A chatbot can't fully replicate the weight of looking someone in the eye and delivering news that changes their life. "The emotional weight of something like a termination conversation is nearly impossible to replicate," said Casey Schaffer, founder of Definitions Coaching and Consulting.
That judgement feels right: these tools are genuinely useful for getting words off the page and into your mouth, but a manager who skips the human debrief entirely is probably over-trusting the machine.
Common questions
Can the AI coach give dishonest feedback to make the manager feel better?
Yes, and it's a known problem. AI tools are built to be agreeable by default, so DeJeu advises managers to explicitly tell the tool to be blunt and critical, not encouraging, during practice runs.
Is it safe to use ChatGPT or Claude to prepare for a work conversation?
For general practice, yes. Don't type in colleagues' names or confidential company details. Describe the situation in general terms and you protect both your privacy and your colleague's.
