You rarely get a long runway before the other person forms an impression of where a hard conversation is going. Say “we need to talk” and their guard goes up before you have said anything else. Open with an accusation and you have already made them defend instead of listen. Much of what people call “the conversation” is shaped by how it starts — and that opening is the piece you can genuinely rehearse in advance, including with a chatbot, if you use it the right way.
What this actually is
This is a rehearsal method for the opening only — not the whole conversation, and not a script to read from. You describe the situation, the model gives you a small set of different opening patterns to choose from, you pick the one closest to your own voice, and you say it out loud a few times before you need it for real. The model has never met the other person and cannot predict how they will respond; it is only helping you get past the blank-page problem of “how do I even start.”
The misconception: getting the wording exactly right
The most common way people misuse a rehearsed opening is to treat it as a script for the whole conversation rather than just the first line. They write out (sometimes with AI, sometimes without) a full paragraph, memorize it, and then the other person responds with something unexpected — deflects, gets defensive, asks a question you didn’t plan for — and the rehearsed person either freezes or tries to force the conversation back onto the script, which reads to the other person as not actually listening. A script fixes the opening; it should not try to fix the whole exchange, because the whole exchange depends on someone else’s actual response, which no amount of preparation can predict.
The fix: rehearse the opening line hard, and plan nothing beyond it except your own listening.
Three opening patterns
Ask the model for variations on your specific situation, not a generic list — generic templates sound like a customer service script the moment you say them out loud.
I need to open a hard conversation with [relationship] about [topic,
one sentence, plain language]. Give me three different opening lines,
one sentence each, using three different patterns:
1. A direct statement of the topic and how I feel about it.
2. A shared-concern framing that names something we both care about
before the specific issue.
3. A question that invites their side before I state mine.
Keep each under 20 words. No corporate or therapy-speak phrasing.
Illustrative shapes, for a conversation about a friend who consistently cancels plans late:
1. Direct: "I want to talk about the last few times we've made plans
and you've cancelled close to the day — it's been bothering me."
2. Shared-concern: "I know we're both stretched thin right now, and I
want to figure out how we keep making time for each other work."
3. Question-first: "Can I ask what's been going on with our plans
lately? I've noticed a pattern and want to understand it."
Pick the pattern that matches what you actually want from the conversation. Direct works when you need the topic named clearly and are prepared for either response. Shared-concern works when you genuinely believe the problem is circumstantial, not personal, and want to signal that before anything else. Question-first works when you are not yet sure of the full picture and want their account before you commit to a frame — but only use it if you mean the question, not as a softened accusation.
The tone check
Before you use any line, run it back through a narrower check than “does this sound okay.”
Read this opening line back to me and flag anything that could land as
blame, sarcasm, or a hidden accusation, even if that's not what I
intended: "[your chosen line]". Then rewrite it once, keeping the same
meaning, with any blame language removed.
The model is reasonably good at spotting loaded words (always, never, again) and hedge phrases that read as passive-aggressive (“I guess it’s fine, but…”). It cannot tell you how the other person will actually hear it, because tone lands differently depending on history you have with them that the model does not have. Treat the check as a second pair of eyes on wording, not a guarantee.
Do not use this rehearsal method to prepare an opening for a conversation involving abuse, a safety threat, or a custody matter. Rehearsing wording is not a substitute for the safety planning those situations require — see prepare for a hard conversation without making AI the referee for where that line sits, and when AI is not therapy for what to do instead. If you already drafted notes or openings in a chat, delete that conversation and avoid continuing on a shared device or family account.
Rehearse without scripting
Once you have a line you would actually say, rehearse it out loud three or four times — alone, in the mirror, on a walk — until it feels like your own words rather than a memorized line. Then stop. Do not draft the second and third sentences of the conversation in advance; you do not yet know what the other person will say, and a pre-written response to an imagined reply is the fastest way to stop listening to the real one.
A useful discipline: after you say the opening line, the next thing out of your mouth should be a genuine reaction to whatever they actually said — even if that reaction is “I need a second to think about that.” If you notice yourself reaching for a line you rehearsed for a different moment, pause instead.
Setting changes the opening too
The same line lands differently depending on where you say it. In person, your tone of voice and pause length carry half the meaning — the words alone, read cold, can sound harsher or softer than intended. Over text or chat, the words are all the other person has, so ambiguity that would resolve itself with a raised eyebrow in person can instead read as an accusation. If you are choosing between a call and a message, default to a live conversation (in person or by voice) for anything involving feelings, blame, or a request for change, and reserve text for scheduling the conversation itself (“can we talk tonight about something”) rather than having it.
I'm planning to open this conversation over [text / a phone call / in
person]. Given that medium, does my chosen opening line need any
adjustment — for example, added context that tone of voice would
normally carry? Keep any change minimal.
Use this pass only if you have already committed to the medium for a real reason (distance, urgency, the other person’s preference) — do not let the model talk you out of a harder but more honest live conversation just because text is easier to send.
Privacy note
If you are describing a real, identifiable situation to a chatbot to get opening-line help, treat it the way you would a message to a friend you trust but who does not need to know everything: skip full names, specific addresses, or details about the other person’s health, finances, or immigration status unless the conversation is actually about those things. Prefer a private session on a device only you control, and delete the chat afterward if the notes would be risky for someone else to find. See what ChatGPT remembers, sees, and shares for what typically gets stored, and turn off training on the account if you plan to describe personal relationship details regularly.
Try it today
Pick one small overdue conversation — not the biggest one you’re avoiding, a medium one. Generate the three opening patterns above, run your favorite through the tone check, and say it out loud five times today, in different rooms if you have to. If the conversation also needs fact-gathering beforehand — what actually happened, what you need, what you’re still unsure about — pair this with the fuller prep method. The rehearsal above is deliberately just the opening; the rest of the conversation is yours to have.
