Difficult Conversation Prep Prompt

What this prompt does

A difficult conversation prep prompt builds the preparation for a hard conversation with someone on your team — your opening lines, the reactions you're most likely to get, and the things they might say that would knock you off balance — while the conversation itself stays yours. The output is a one-page brief you read twice: the night before, and five minutes before.

No prompt has the conversation for you, and this page won't pretend otherwise. What AI does well is the part managers skip: structuring how you'll open, thinking through how it might go wrong, and catching accusatory phrasing before it leaves your mouth. Harvard's Program on Negotiation makes a related point about why prep matters — difficult behavior is often someone trying to express legitimate concerns badly, which makes the conversation partly a listening job, and listening is easier when you've already rehearsed the hard parts.

Two neighbours on this site do adjacent jobs. The tone adjustment prompt rescues a message you already wrote while frustrated; this prompt prepares a conversation that hasn't happened yet. And the discovery call prep prompt is the sales cousin — a brief built from pasted research about a company; this one runs on nothing but your own description, which is why the privacy rule below is the whole design.

The prompt

Help me prepare for a difficult conversation with someone on my team. I'll describe the situation hypothetically — no names, no identifying details — and you'll build me a one-page prep brief. Do not write a script for the whole conversation; build the preparation.

**The situation:** [ROLE + BEHAVIOR PATTERN, DESCRIBED HYPOTHETICALLY — e.g. "a senior engineer, strong technically, who has dominated the last three team discussions"]
**What I've already tried:** [PRIOR FEEDBACK, EARLIER CONVERSATIONS, OR "NOTHING YET"]
**What I need from the conversation:** [THE CONCRETE CHANGE OR AGREEMENT — not "make them see reason"]
**My biggest worry:** [WHAT YOU'RE AFRAID THEY'LL SAY OR DO]

Build the brief in four parts:

1. OPENING LINES — my first two or three sentences, written out. Name the behavior and its effect on the team; no statements about their intent or character. Flag any wording in my situation description that assigns motive rather than describing what happened.
2. LIKELY REACTIONS — the three most probable responses (defensive, hurt, counter-argument, or agreement), each tagged [GUESS], each with one steadying line I can use to keep the conversation on track.
3. MY TRIGGERS — from my "biggest worry," the one or two things they might say that would rattle me, and a reframe for each so I hear it as information about what they value rather than as an attack.
4. THE EXIT — what a good ending looks like in one sentence, and what I should confirm in writing within a day.

Rules:
- Everything stays hypothetical. If I slip and include a name or identifying detail, replace it with a role in your output.
- Reactions are guesses, not predictions — the conversation tests them.
- Keep it to one page.

How to use it

1. Write the situation description like a case study, not a complaint. "A designer who has missed two sprint deadlines and gets short when asked about status" preps a better conversation than "my designer is checked out." The first describes behavior you both witnessed; the second is a verdict you'd have to defend.
2. Take the motive-flagging seriously. If the brief flags that your own description assigns intent ("doesn't care," "is being difficult"), that's the most useful line on the page. You were about to open with an accusation you can't prove.
3. Read the triggers section twice. Stanford's Margaret Neale suggests anticipating what the other person might say that would set you off, and reappraising it before the meeting rather than during. The reframes in section 3 are that technique on paper.
4. Then put it away. The brief is preparation, not a teleprompter — memorize the first sentence, know the shape of the rest, and let the conversation be a conversation.

Example output

Inputs: the situation is a senior engineer, strong technically, who has dominated the last three team discussions — two quieter teammates have stopped contributing. Already tried: a light comment after one meeting. Needed: they keep contributing, and make room for others. Biggest worry: "they'll say the others should just speak up."

Opening lines: "I want to talk about how our team discussions have been running. In the last three, you've carried most of the airtime — your input is some of the strongest we get, and that's part of why it fills the room. I've noticed two people have stopped contributing, and I need your help changing that." (Flag: your description said the engineer "takes over" — that wording assigns intent. "Carried most of the airtime" describes what happened.)

Likely reactions: [GUESS] Counter-argument — "if they have opinions, they should voice them." Steadying line: "That's fair, and I'm working on that side too. Right now I'm asking about the part you control." [GUESS] Hurt — going quiet, hearing it as "contribute less." Steadying line: "I'm not asking for less from you. I'm asking for the same quality with more room around it."

Your trigger: exactly that counter-argument, because part of you agrees with it. Reframe: if they say it, it tells you they value directness and self-advocacy — a value you can build the solution around, not a dismissal of your point.

The exit: you both name one concrete thing to try in the next meeting. Confirm it in a two-line message the same day.

The flag in the first section is the prompt earning its keep — the difference between "you take over meetings" and "you carried most of the airtime in the last three" is the difference between a fight and a conversation.

Variations

A peer, not a direct report

"The person is a peer, not someone I manage — I can't set expectations or consequences, only ask. Rebuild the opening so it's a request between equals, and change the exit: what does a good outcome look like when all I have is the working relationship?"

The conversation you've been avoiding

If it's been weeks, add: "I've put this conversation off for [TIME], so the delay is now part of the situation. Give me one honest sentence to own that at the start, without turning the meeting into an apology for the timing." The procrastination buster prompt is the sibling tool for the avoidance itself.

Delivering a decision, not opening a discussion

"The decision is already made — this conversation informs, it doesn't negotiate. Rewrite the opening so it doesn't invite a debate I can't grant, and make the steadying lines acknowledge the reaction without reopening the decision."

Common pitfalls

• Don't paste the real details. Not the name, not the review score, not the salary conversation from March. Anything typed into a public tool may be retained, and a direct report's situation deserves the same discretion as their personnel file. The hypothetical habit costs nothing and works on any tool.
• Don't treat the reactions as a forecast. They're tagged [GUESS] because that's what they are — a person who walks in certain how the other side will respond has stopped listening before the meeting starts.
• Don't script past the opening. The further into the conversation your script runs, the more wrong it gets, because every line after the first depends on what they actually say.
• Don't skip the written follow-up. What you agreed evaporates within a week unless someone writes it down, and that someone is you — the follow-up email prompt closes the loop while it's fresh. If the meeting needs setting up first, the meeting request email prompt does the invite without leaving them to dread it for two days.

Who uses this prompt

• New managers: the first real feedback conversation is the scariest hour of the first year, and borrowed structure lowers the temperature. More in prompts for managers.
• Experienced managers: not for the routine conversations — for the one that's been sitting on the to-do list for three weeks disguised as something else.
• Team leads without formal authority: the peer variation exists because half of difficult conversations happen sideways.

ChatGPT, Claude, and Gemini all run this prompt unmodified. The design leans on two ideas from Anthropic's prompting best practices: give the model the full context it can't infer (what you've tried, what you need, what you fear), and constrain the output to the job — preparation, not performance. The conversation still belongs to you. The brief just means you walk in having already survived the worst version of it once.