Which prompt to choose
Conflict, feedback, and refusal call for different registers. In conflict you have to lower the tension; in feedback you have to be clear without wounding; in saying no you have to protect the boundary without breaking the relationship. Choose based on what's ahead of you.
- Tackling a conflict or a tense disagreement: first prompt, which looks for the meeting point.
- Giving uncomfortable feedback: second prompt, which makes it clear and receivable.
- Saying no without guilt: third prompt, which gives you the gentle, firm formula.
How to do it
- Open the AI assistant.
- Describe the situation from both sides: what happened from your point of view and, as far as you can, from theirs. The AI prepares a better conversation if it understands the other person too.
- Copy the prompt and send.
- Ask for the simulation: "now you play the other person and test me with the hardest replies."
The operational syntax for a conflict:
I have to face a tense conversation with this person: "[who they are and what your relationship is]."
What happened from my point of view: "[the facts]."
What I think they see from theirs: "[their likely perspective]."
What I want to achieve: "[the real goal, not being right]."
Help me to:
1. Open the conversation by lowering the tension, not raising it.
2. State my part without accusing (speaking about facts and how I feel, not about what "they are").
3. React if they get defensive.
Give me sentences I could actually say, in my tone, not from a psychology manual.
The operational syntax for giving feedback:
I have to give uncomfortable feedback to: "[who, and what role they have relative to me]."
The concrete problem: "[specific behavior or result, with an example]."
What I'd like to change: "[the result I expect]."
Help me phrase it so that it:
- Starts from a precise fact, not from a judgment about the person.
- Is clear about what has to change, without dancing around it.
- Leaves room for their version.
Constraints: no fake "sandwich" (praise-criticism-praise) that waters down the message. Direct but respectful.
The operational syntax for saying no:
I have to say no to this request: "[what they're asking and who]."
Why I want to refuse: "[the real reason]."
I want to say no without ruining the relationship and without apologizing endlessly.
Give me 3 versions:
- A short, gentle one for those who don't deserve explanations.
- One that briefly explains the reason, for someone close to me.
- One that proposes an alternative, if I want to soften the no.
Constraints: no rapid-fire apologies, no "I'm terribly sorry but." A clear no is a favor to both.
After the preparation, do the simulation. Asking the AI to play the other person and throw the hardest objections at you is what takes away the trembling: the real conversation won't be the first time you face it.
A real example
Anna had to tell a friend who kept borrowing money from her that she couldn't do it anymore. She felt guilty just thinking about it. She used the third prompt, explained the real reason (it was putting her in financial difficulty) and asked for the version for a close relationship.
The AI gave her a line that started from affection and arrived at the boundary without accusations: it acknowledged the friend's need, stated her own limit, proposed a help different from money. Then Anna asked for the simulation: the AI played the insistent friend, and Anna practiced holding her ground gently. She arrived at the real conversation calm. The no went through, the friendship stayed.
When it does NOT work (and how to fix it)
If the sentences sound fake or not yours
The AI by default writes in a slightly stilted register. Paste a message you wrote in the past and ask "use my way of speaking, not this one." And remember: it's a track, not a script. Reciting from memory shows and stiffens you; take the concepts and say them in your own words.
If it prepares you only to win, not to understand
If you frame the request as "how do I convince them I'm right," the AI goes along with you and gives you ammunition, not solutions. Reframe the goal: not being right, but resolving. Add "help me understand their reasons too, not just defend mine": the conversation changes tone in the preparation itself.
If the situation is truly serious (abuse, harassment, safety)
Here the AI isn't enough and it has to be said without dancing around it: for situations of abuse, workplace harassment, or danger, a prepared conversation isn't the right tool. You need the right people, a human resources office, a lawyer or, if there's danger, the authorities. Use the AI at most to put the facts in order before turning to someone who can really help you.
A tip from someone who actually uses it
Prepare the conversation the night before, not a minute before. The difference isn't the perfect sentence: it's having already heard the objections in the simulation, so in the moment you don't panic. The AI is a gym where you can make mistakes for free before the real match.
Frequently asked questions
Isn't it dishonest to prepare a personal conversation with AI?
No: preparing is respect for the other person, not manipulation. Thinking beforehand about how to say something difficult, so it lands clear and doesn't wound more than necessary, is what someone who cares about the relationship does. Dishonest would be acting out emotions you don't feel, and no prompt asks that of you.
Can the AI replace a therapist or a mediator?
No. It can help you order your thoughts and rehearse what to say, but it doesn't know you, doesn't hear the tone, has no responsibility toward you. For deep conflicts or real suffering you need a competent person. Treat the AI as a notebook to reason in, not as someone who heals you.
If the other person finds out I prepared this way, won't I look bad?
That's the fear to dispel. No one reproaches you for thinking about how to say something important: anyone preparing for an interview, a speech, or a raise request does it. What gets noticed and rejected isn't the preparation, it's sounding fake. If you take the concepts and say them in your own words, from your tone and your eyes all that shows is a person who cared enough to think about it beforehand.