Rehearse the Hard Conversation Before You Have It

Workflow area: Leadership Ops · Tier: Core · Hands-on time: ~30 minutes

Pilots get simulators; leaders traditionally get to crash the real plane. Role-play with a model fixes the asymmetry: you can take the hardest version of the conversation three times before 9am, hear your opening land badly, and fix it while it costs nothing. The rehearsal is not about scripting the human — real people will surprise you — it is about draining the adrenaline from your first attempt so your second attempt, the real one, gets your full attention.

When to use — and when not to

• Use it for high-stakes 1:1 conversationsturning down a promotion expectation, delivering hard feedback upward, announcing a decision the person will hate.
• Use it before mediating conflict between two team membersrehearse both perspectives, then prep an anonymized agenda that neither party's private disclosures leak into.
• Use it when you already know your opening is weakif you have rewritten the first sentence four times in your head, you need reps, not more rewriting.
• Skip it for conversations that need spontaneity more than polishan over-rehearsed leader sounds scripted, and a report who feels "managed" hears the script, not the care.
• Skip real names and real disclosuresthe counterpart in your rehearsal is a role, not a person; feeding a model someone's actual confided grievances to simulate them is a privacy violation with extra steps.
• Skip it as a substitute for HR guidanceterminations, conduct cases, and legal-exposure conversations get rehearsed only after HR has shaped what you can and cannot say.

The recipe

1. 1.

   Define the scenario on paper first. One paragraph: the message you must deliver, the outcome you need, the reaction you most fear, and the thing you must NOT say (the unannounced reorg, the peer comparison, the confidential context). The last item is your tripwire — rehearsal exists partly to practice not saying it.

2. 2.

   Set up the role-play.

   Copy

   Role-play a difficult workplace conversation with me. You play
   [ROLE — e.g. "a senior engineer, 6 years tenure, who expected
   the promotion I am about to tell them they did not get"].
   
   Play them [PICK ONE: defensive and argumentative / visibly upset
   and quiet / blindsided and asking rapid questions]. Be realistic,
   not cartoonish: they are a competent professional having a bad
   day, not a villain. Push back where a real person would. Do not
   make it easy for me, and do not resolve the conversation for me.
   
   Stay in character until I say "PAUSE". I open the conversation:
   [TYPE YOUR ACTUAL OPENING LINE]
   

   Run the conversation for 8–12 exchanges. Then say PAUSE and switch modes.

3. 3.

   Get the critique against your chosen framework.

   Copy

   PAUSE. Step out of character. Critique my side of that
   conversation against [FRAMEWORK — e.g. "SBI + Radical Candor" /
   "Nonviolent Communication" / "Crucial Conversations"].
   
   Score me on: (1) clarity of the core message — could they repeat
   the decision back?; (2) where I defended, over-explained, or
   negotiated a settled decision; (3) empathy that was concrete vs
   performative; (4) anything I said that made a commitment or
   comparison I should not have; (5) the single change with the
   highest payoff. Quote my exact lines as evidence for each point.
   

   Good critique output looks like:

   Copy

   2. You over-explained: after delivering the decision you gave
      four justifications unprompted ("and also the budget...").
      Each extra reason handed them a new thing to argue with.
   4. Risk line: "you're definitely on track for next cycle" — you
      just made a promise a calibration committee owns, not you.
   

4. 4.

   Run it again, harder. Take the top critique item, change one thing, and rerun with the same reaction mode — or escalate: "play them angrier this time" or "this time they threaten to quit." Two or three reps is the sweet spot; past five you are polishing a script instead of building readiness.

5. 5.

   For mediation: rehearse both chairs, then draft the anonymized agenda. Role-play as party A, then as party B — you will discover your framing favors one of them. Then: "Draft a mediation agenda for a conflict between two colleagues over [NEUTRAL DESCRIPTION — e.g. 'code review tone and turnaround expectations']. Ground rules, sequence of speaking, questions that surface interests rather than positions. Include nothing that reveals what either party told me privately." The agenda goes to both parties in advance; private disclosures never do.

Make it yours

• Voice practice: run the role-play in ChatGPT voice mode or similar — hearing yourself say the words out loud, with real-time interruptions, is meaningfully harder and closer to the day than typing, and worth the awkwardness.
• New managers: rehearse the routine-but-scary set once each — first critical feedback, first "no" to a request, first pushback from a senior report — before any of them is live; the library of reps transfers.
• Senior leaders and skip-levels: rehearse announcements to hostile rooms, not just individuals — "play a team of 12 hearing this reorg; give me the three hardest questions from the back of the room" — and keep pb-06's rule in view: what you rehearse saying still gets fact-checked before you say it.

Quality check

• The scenario contained no real identities or confided detailsrole and situation only; if a colleague could recognize themselves in your prompt, rewrite it.
• The AI counterpart pushed back hardif the rehearsal felt easy, re-prompt for more resistance; an agreeable simulator trains false confidence.
• The critique quoted your actual linesfeedback without quotes is generic coaching content, not analysis of your performance.
• Your core message survived intact across repsrehearsal should sharpen the decision's delivery, not negotiate the decision away; if rep three ends with you conceding, that is a finding about your resolve, not your phrasing.
• You stopped rehearsing while it still felt slightly uncomfortablethe goal is readiness, not a memorized performance the real human will immediately derail.