You know the conversation you have been putting off. Maybe it is about money, a repeated broken promise, a boundary that keeps getting crossed, or something smaller that has quietly built up over three months of not-quite-saying-it. You have rehearsed it in the shower a dozen times, and every version ends with you sounding either too harsh or too vague.
This is a preparation method, not a script-writing service. You bring the facts and the feelings; a model helps you sort them into a one-page brief — what actually happened, what you need, what you are still unsure about, and one sentence to open with. The conversation itself stays a conversation between two people. Nothing here replaces that, and nothing here should be used to prepare for a conversation involving abuse, coercive control, custody disputes, or any situation where you feel physically unsafe.
Where this fits, and where it does not
The workflow below is for the conversations that are hard because they are awkward, overdue, or emotionally loaded — not because someone is in danger. A useful boundary test: if your instinct is “I need this to go well” rather than “I need to be safe,” you are likely in scope. If any part of you is afraid of what happens after you speak, or the other person has threatened you, controlled your money or movement, or a child’s custody is at stake, this is the wrong tool entirely.
Do not use a chatbot to prepare for a conversation involving domestic abuse, coercive control, threats, or an active custody or protective-order matter. A model has no way to assess your safety, cannot detect manipulation in real time, and may draft language that increases risk. Contact a local domestic-violence support service (in Estonia, the state victim support service and its 116 006 helpline at palunabi.ee; in the US, the National Domestic Violence Hotline’s safety-planning resources), a qualified lawyer, or emergency services, depending on the situation. If you used a shared device or account, use a safer device if available and ask an advocate how to protect privacy and preserve evidence; do not automatically delete material that may be relevant to safety or legal proceedings.
For ordinary relationship friction — a partner, sibling, adult child, roommate, or close friend — the model’s job is narrow: help you organize what you already know into a form you can actually use in the room.
Why a short brief can help
In some hard conversations, an accusation-first opening, an untested inference about intent, or an unclear request can make the discussion harder. A model can sort text into categories, but it cannot decide what is true about another person’s motives. It was not in the room and has only the account entered into the prompt. A brief may help expose gaps; it does not guarantee a good outcome.
This is the same failure mode covered in your AI agrees with you too much: a chatbot will tend to validate whichever framing you give it, including an unfair one, because agreement is the path of least resistance in a fluent reply — a measured tendency the research literature calls sycophancy (Sharma et al., 2023). The fix is not to distrust the tool; it is to feed it a narrower job than “tell me if I’m right.”
The four-part prep brief
Do this the day before, or at most a few hours before, the conversation — not in the heat of it. Write freely first, in your own words, for five minutes: what happened, how you feel about it, what you’re afraid will happen if you raise it. Then run that text through four separate passes.
Pass 1: Facts only
Here is my account of a situation I need to raise with [relationship,
e.g. "my partner" or "my brother"]:
[paste your freeform notes]
List only the observable facts: things that were said, done, agreed to,
or that occurred, with dates or approximate timing where I gave them.
Put anything that is my interpretation, assumption, or guess about their
motives into a separate "not a fact" list instead. Keep the facts list
under 120 words.
This step alone often changes what you plan to say. A grievance that feels like one solid memory (“he never listens to me”) usually turns out to be three or four specific incidents plus a large interpretive leap. Naming the leap does not mean it is wrong — it means you now know it is a leap, and you can decide whether to state it as an observation (“I’ve noticed X three times this month”) instead of a verdict (“you never listen”).
Pass 2: Your actual need
Based on the facts above, help me write one or two sentences stating
what I actually need going forward — not what I want the other person
to admit, feel, or apologize for, just the concrete change or agreement
I'm asking for. If my notes only describe a complaint with no stated
need, point that out and ask me what I want to happen next time.
A prep brief without a stated need turns into a complaint session. This pass forces the distinction between venting (legitimate, but not what this conversation is for) and asking for something specific.
Pass 3: What you are still unsure about
List the open questions I genuinely do not know the answer to about
this situation — things only the other person can tell me, not things
I could figure out myself. Do not guess at answers or suggest what they
"probably" think.
This pass is the antidote to walking in with a verdict already decided. If the honest answer to “why did they do that” is “I don’t know,” write the question down and plan to actually ask it, rather than asking a rhetorical version of it that is really an accusation.
Pass 4: One opening line
Using the facts, my stated need, and my open questions above, draft
three short opening lines (one sentence each) I could use to start this
conversation. Each should name the topic plainly, without blame
language, and leave room for their response. Do not soften the topic
into vagueness — I still need them to know what we're discussing.
Pick the version that sounds most like you, not the most polished one. A line that sounds like a corporate mediator wrote it will feel wrong the moment you say it out loud.
Read your chosen opening line aloud once before the conversation. If it makes you wince, it is probably either too harsh or too rehearsed-sounding — adjust the wording yourself rather than asking the model for a fourth draft.
Validation: read it back before you use it
Before the conversation, reread your four-part brief and check it against two things it cannot self-verify: did you leave out a fact that makes you look worse, and is your stated need actually about you, or is it secretly a demand that the other person change how they feel? A brief that only survives if the other person agrees with your account is not ready. See the emotional clarity check for a companion method that separates what happened from the story you told yourself about it — useful the night before, on the same material.
The escalation rule
If, while preparing, you notice fear rather than dread, a pattern of control rather than a pattern of forgetfulness, or facts that describe threats, monitoring, or restricted access to money or contact with others — stop the prep exercise. Use a safer device if available and ask a qualified advocate how to protect privacy and preserve relevant evidence; do not make a blanket decision to delete records from this article. That is not a hard conversation to prepare for with a template; it is a safety situation to plan with a person. When AI is not therapy and stop the chat, contact a person both cover how to recognize that line and who to contact instead.
For the ordinary case, the model’s part ends when the brief is done. It does not sit in on the conversation, and it should never be quoted back to the other person as though it delivered a verdict (“even ChatGPT agrees you’re being unreasonable” is a way to lose the conversation and the trust of the person you’re talking to).
Use it today
Pick the one overdue conversation you have been avoiding. Run the four passes above on it tonight, using the difficult conversation prep sheet to keep the four sections on one page. If you also want help with exactly how to open your mouth in the first two minutes once you’re in the room, pair this with the first two minutes of a difficult conversation.
This communication worksheet is not a validated psychological intervention, abuse assessment, mediation method, or legal strategy. For coercion, abuse, custody, threats, or danger, use appropriate qualified safety, legal, or mental-health support instead.
