As of 13 August 2026, AI can only partly have a difficult conversation.

What to actually do

1. Use a tool built for this

   The route this page recommends

2. Do it yourself

   Second choice

   A chat interface, chat-fluent skill, and roughly 30 minutes until you can act on the result.

   How to actually do it

   1. Open a chatbot such as ChatGPT, Claude or Gemini and paste the full prompt.
   2. Fill in each bracketed section with the facts you know, separating what happened from your interpretation of why it happened.
   3. Ask the chatbot to produce the opening, questions, boundaries and pause plan, then remove any sentence you would not genuinely say.
   4. Compare every factual claim in the draft with your own messages, memories or records, and delete claims you cannot support.
   5. Read the draft aloud and shorten it until the request and boundary are clear without sounding like a prepared speech.
   6. Choose a setting where you can leave or end the call, tell a trusted person where you will be, and do not arrange a private confrontation if you feel unsafe.
   7. Use the opening and one or two questions in the conversation, then follow the pause plan if the exchange becomes abusive or unproductive.
   8. If the situation includes urgent or worrying symptoms, contact NHS 111 through https://111.nhs.uk or call NHS 111; seek emergency help if there is immediate danger.

   Prompt

   Help me prepare for a difficult conversation. Do not provide a medical assessment of me or the other person, and do not assume you know their motives. If this involves threats, abuse, coercive control, self-harm, or immediate danger, prioritise safety over having the conversation and tell me to seek appropriate urgent help. If I describe urgent or worrying symptoms, direct me to NHS 111.
   
   My relationship to the person: [for example, partner, family member, friend, colleague]
   The issue, in factual terms: [what happened, without labels or guesses about motives]
   What I want from the conversation: [specific outcome]
   What I need to say: [main points]
   What I am willing and unwilling to do: [boundaries]
   What I am worried might happen: [concerns]
   Where and when I might have it: [setting and timing]
   
   Produce:
   1. A short opening in plain, natural British English using specific observations, how I feel, what I need and a clear request.
   2. Three calm questions that help me understand the other person without putting words in their mouth.
   3. Two boundary statements I can use if the conversation becomes insulting, pressuring or unsafe.
   4. Brief replies to likely reactions, including denial, anger, silence and a reasonable disagreement.
   5. A practical plan for pausing or ending the conversation safely.
   6. A final checklist of factual claims and assumptions for me to check before I use any wording.
   Keep the language non-accusatory, do not promise a particular outcome, and do not tell me to confess, threaten or manipulate anyone.

   Open it prefilled in ChatGPT or Claude, or copy it into Gemini, which takes no prefill link.

3. Hand it to a person

   The distant third

   A person who owns the outcome does this end to end, worth it when the failure is dear.