Rehearse the hard conversation
Have the AI play the other person, push back, then tell you what landed. Before a raise, a landlord or roommate issue, a tough family holiday talk, or a meeting with a teacher, set the role, the stakes, and the objections you expect. Ask for the honest note, not a compliment. You still have the real conversation.
Next step: copy the first prompt that matches your task. Fill in the brackets. Send only that.
A prompt is the message you paste. An approval gate is a stop you control. A checkpoint is a pause after one phase. School pages teach. They do not do the homework. Follow your teacher's rules.
Last updated
How do you do it?
Do the steps in order. Each line is one action, and you stop at the end of that action.
- 1 Set the role, the stakes, and the likely objections. Say who the chat should play, what you need from the conversation, and the pushback you already expect. A blank role becomes a friendly stranger who agrees with you.
- 2 You speak. It plays the other person and pushes back once. Answer in your own words, out loud if you can. Tell it to stay in character for one objection, then stop. A script it writes for you is a speech you will not remember.
- 3 Ask what landed, and ban the compliment. Ask which sentence was clear, which one dodged the ask, and what the other person still does not know. Tell it not to praise you. Agreement is not a rehearsal.
- 4 Keep one line, then have the real conversation. Rewrite that one line yourself. Practice it once more. You still choose the moment, the place, and whether to talk at all. The chat does not attend the meeting.
Work
What prompt fits “Asking for a raise”?
The weak version is "Tell me exactly what to say so my boss says yes." That invites a speech and a fantasy yes. The directed version makes the chat play your boss, with the objection you already fear, then marks what landed. You still have the meeting. A number you have not checked does not go in the ask.
Do not write my script. Do not tell me I deserve the raise. Do not invent a salary, a company policy, or a compliment.
Play my boss. Stay in that role until I ask you to stop.
Stakes: I am asking for [the change, in my words].
Facts I can say out loud: [role, how long, one result I can point to].
Facts I will not claim: [a number I have not checked, or a promise I cannot make].
The objection I expect: [budget, timing, or "not this cycle"].
Ask me to say my opening in my own words. Wait.
Then respond once, in character, with that objection. Do not solve it for me.
Then stop and wait.School
What prompt fits “Talking to a teacher”?
The weak version is "Write the speech that gets me an extension." The teacher will hear you, not the chat. The directed version lets you practice the ask. You still write anything you turn in, and you still follow the late-work rule on the syllabus.
This is for learning. Do not write anything I will turn in. Follow my teacher's rules.
Do not write a speech I could read to my teacher. Do not promise a grade or an extension. Do not invent a policy.
Play my teacher for [class].
What I need to ask: [extension, a confusing grade, or help with one topic].
The rule I already read on the syllabus: [paste it, or "I have not checked the syllabus"].
The objection I expect: [deadline already passed, or fairness to the class].
Ask me to say the ask in one or two sentences, in my words. Wait.
Then push back once, in character. Do not give me a better paragraph to memorize.
Then stop.Life
What prompt fits “A landlord, a roommate, or a family holiday talk”?
The weak version is "Make them agree." The directed version names the person, the one request, and the objection you are avoiding. A holiday table and a lease are different rooms. You pick one conversation. You do not paste a legal threat you have not checked.
Do not tell me I am right. Do not invent a law, a lease clause, or a family fact I did not type. Do not write a speech.
Play this person: [landlord, roommate, or family member].
The one thing I need to say: [the repair, the rent, the chore, or the holiday boundary].
Stakes: [what happens if we do not talk].
Their likely objection: [cost, blame, or "not during the holiday"].
What I will not bring up: [list].
I will type what I would say. Respond once, in character, with the objection.
Do not soften their reply into agreement. Then stop and wait.What mistakes should you avoid?
These are the misses that waste the result. Skip them before you send anything.
- Asking for the perfect script and memorizing it. The other person will not follow your outline.
- Skipping the objection because you want to feel ready. The pushback is the rehearsal.
- Letting the chat play a cartoon villain, or a saint who agrees. Ask for the person you will actually meet.
- Asking "how did I do?" and keeping the praise. Ask what landed and what was missing.
- Putting a salary you have not verified, a lease clause you have not read, or a threat into the practice. If it is not true in the room, it is not practice.
- Using a school rehearsal to get the email, the essay, or the excuse written. You still say it, and you still do the work.
More in this section: Stop the yes-man reply · Ask for options and tradeoffs before you decide · Let the AI interview you first
Practice on a real task: Practice for a presentation · Write a difficult work email · Prep for a job interview with AI · Negotiate a job offer with AI
Also on this site: Write a speech with AI · Write holiday cards and letters with AI · Work · School · Life
Last updated · All director guides
Where does this come from?
These are public pages. The note after each link is what that page says, in plain words.
- OpenAI Help, “Prompt engineering best practices for ChatGPT”. OpenAI says a prompt should be clear and specific, and should provide enough context for the model to understand what you are asking. Who the other person is, what you want, and what they are likely to object to is that context. The same page says to refine the prompt after you see the reply, and to name a tone with ordinary words such as serious or friendly.
What else do people ask?
Open a question. The short answer under it is the one to use.
What is the one-line version?
Have the AI play the other person, push back, then tell you what landed.
Why not ask it to write the script?
A script assumes the other person says the next line you hoped for. OpenAI says the prompt has to give enough context, and that you refine it after you see the reply. The context here is their objection. You answer it in your words.
What if it only compliments me?
That is a yes-man reply. Tell it to stop praising you and to name the sentence that dodged the ask. The page for that habit is Stop the yes-man reply.
Can it decide whether I should have the conversation?
No. Ask for options and the tradeoff on the options page if you have not chosen. This page starts after you know the one thing you need to say. You still choose whether to say it.
Will this do my homework or write the email I send?
No. Practice is spoken. The email you send is a different guide, and you edit it. A class assignment is still your writing. Follow your teacher's rules.