The sandbagging eng lead, the sales VP furious about a cut feature, the CEO who wants a date you can't give — you'll have that conversation once, live, with your credibility on the table. This playbook rehearses it first with AI playing the counterpart, steel-manning their position before you ever open your mouth.
Workflow area: Stakeholder Communication · Tier: Foundation · Hands-on time: ~20 minutes
Most PMs walk into hard conversations having rehearsed only their own side. The result is a monologue that collapses on the first objection they didn't anticipate. An AI role-play flips the preparation: you make the model argue the other side at full strength, discover where your position is actually weak, and iterate your framing in a sandbox where a bad answer costs nothing. The conversation you eventually have is your third or fourth attempt, not your first.
Write the one-paragraph situation brief. Who is the counterpart, what do they want, what do you want, what constraint is non-negotiable, and what does a good outcome look like. If you can't write this paragraph, you're not ready to rehearse — you're not ready for the meeting either.
Steel-man them first. Before any role-play, make the AI argue their side as its best self:
You are [ROLE — e.g. "a VP of Sales at a B2B SaaS company"]. Here is
the situation: [SITUATION BRIEF — anonymized, no real names].
Argue YOUR side of this disagreement as persuasively as an
experienced, intelligent person in this role would. Do not strawman
yourself. Include:
1. Your three strongest arguments, in order of strength.
2. The legitimate business pressure driving your position.
3. The thing the PM is probably underestimating about your world.
4. What you would need to hear to change your mind.
Read the output slowly. If argument #1 surprises you, the rehearsal has already paid for itself.
Run the live role-play. Start a fresh conversation:
Role-play a difficult work conversation with me. You play [ROLE],
whose position is: [PASTE THE STEEL-MAN SUMMARY]. I play the PM.
Rules:
- Stay in character. Push back the way this person realistically
would — interrupt, redirect, get frustrated if I'm vague.
- Do not cave early. Only move if I give you a genuinely good reason.
- If I say "PAUSE", break character and give me coaching: what
worked, what sounded defensive, what I missed.
I'll open the conversation now: [YOUR OPENING LINE]
A realistic exchange looks like:
[as Sales VP]: "So let me get this straight — I committed this
integration to two seven-figure renewals based on YOUR roadmap,
and you're telling me it's moving a quarter? What am I supposed
to tell them, exactly?"
Answer as you would in the room. When you flounder, say PAUSE and take the coaching.
Iterate your opening and your concession ladder. Rerun the conversation two or three times. Fix the opening ninety seconds first — that's where tone gets set. Then work out your concession ladder: what you can offer, in what order, and the line you will not cross. Ask the model directly: "Break character. Rank the three openings I've tried. Which one made you least defensive, and why?"
Extract the prep sheet and delete the drama. End with:
Break character. Summarize into a half-page prep sheet: their three
core arguments, my strongest responses, the concession ladder we
converged on, likely emotional flashpoints, and my opening line.
The prep sheet goes into your meeting notes. The role-play transcript does not go anywhere.
Never put real colleagues' names, performance details, or verbatim internal disputes into a consumer AI tool — rehearse with roles and anonymized situations in your organization's approved tenant, and treat the transcript as disposable, not shareable.
Book the rehearsal 24 hours before the real conversation, not 30 minutes — the gap lets the steel-man arguments sink in, and the strongest sign it worked is walking in genuinely curious about their answer to the one question you found they can't easily dodge.