After a disagreement with a colleague, Pavel wants to write a sharp email. First he notes what bothered him: he did not receive timely information about a changed date. He does not know whether it was intentional. He asks AI for a brief request that they confirm changes with each other next time. But the draft adds an accusation of a lack of respect. Pavel deletes it. He did not want to convey that and has no evidence for it.
He saves the message and returns to it later. He has one thing ready that he stands by. He has not adopted the AI response in full: he used part and rejected part. The help served its purpose through precisely this shared search. He does not need a personality profile of his colleague or scenarios for five possible responses.
The AI sandwich
The AI sandwich is our aid for conversations in which you want to address something. You begin with your own need, invite AI in and take the result back into your own hands.
Begin with yourself. Do you need to organise a worry, stay with a feeling, prepare a conversation or decide? You can also begin with still searching for your question. Depending on the need, choose a chat, your own notes, rest or a person if one is available. The aid should make choosing easier, not lead you to AI every time.
Work with AI. Give it a clear role and correct misunderstandings. When you want to change direction, say so: ‘I don’t want another explanation now. I want to prepare a specific sentence.’ Notice whether the response unnecessarily broadens the original problem or takes over decisions that should remain yours.
End on your own terms. You can take away a message, a question for a professional, a decision to wait for now or a calmer moment. Pleasant conversation need not end with an achievement. When addressing a recurring problem, look later at what changed outside the conversation too. The sandwich is an author-created aid, not a clinically validated method. Use what helps you maintain your own direction.
Practice without predicting the other person
When preparing for an ordinary difficult conversation, you can try:
I want to practise a conversation about [topic]. Create a fictional counterpart who asks relevant follow-up questions. Do not pretend to know how a particular person will answer. After a few exchanges, show me which formulations of mine were unclear and which promises I might not want to make. Then leave the choice of wording to me.
Take one or two usable sentences from the practice. The other person will not follow a script and may not respond warmly. For an ordinary conversation, the aim is to express your own need clearly, not prepare a foolproof way to make someone agree. In violence or coercion, do not use this practice as a substitute for safety planning with a professional.
A small change that belongs to you
When you want to try something, choose a proportionate step within your control. Delay an automatic ‘yes’, write to a course organiser, arrange space for a conversation, switch off a distracting notification. After a while, look at what actually happened. The impact need not match AI’s explanation. That is useful information too.
For reflection, it is enough to know why you are trying the change and what happened afterwards. A daily check of ten indicators could add another burden. You can keep useful support long-term, particularly if it helps you manage a health limitation. What matters is whether you can live with it according to your own intentions. The number of aids used does not itself decide that.