Home · Personal · Health & Wellbeing · Mental wellbeing

YES

As of 13 August 2026, AI can prepare for a difficult conversation.

Most people should hand this to a purpose-built tool.

Can you do it?

5 minutesto a draft.

30 minutesto something you’d act on.

Cost, all in£0

Skill needednone

Who has to check ityou

What the alternative costsWysa is an AI wellbeing support tool with evidence-based self-help exercises.

If this goes wrong: the wording inflames the situation, damages trust or leaves an important safety issue unaddressed.

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, no skill needed, 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 preparation prompt.
    2. Fill in the bracketed sections with the situation, relationship, aim, confirmed facts, uncertainties, boundaries and any safety concerns.
    3. Read the generated plan and remove any claim about the other person's motives or feelings that you cannot support from what you know.
    4. Paste back the opening and main points, then ask the chatbot to make them shorter, calmer and closer to the way you normally speak without changing the facts.
    5. Ask the chatbot to role-play the other person one reply at a time, and answer each reply yourself rather than accepting a complete scripted exchange.
    6. Write down the points you will actually use, the boundary you will keep and the words you will use to pause or leave the conversation.
    7. Choose a suitable time and setting, then compare the final wording with your own memory of the situation before using it.
    8. If the situation involves immediate danger or urgent or worrying symptoms, do not rely on the script and contact appropriate urgent help, including NHS 111 where relevant.

    Prompt

    Help me prepare for a difficult conversation. I will provide the situation, my relationship to the person, what I want to achieve, the facts I know, what I am worried about, and any boundaries I need to keep.
    
    Situation: [brief description]
    Person and relationship: [who they are to me]
    What I want to achieve: [specific outcome]
    Facts I am sure about: [facts only]
    What I am assuming or unsure about: [uncertainties]
    What I need to say: [main points]
    What I need to ask: [questions]
    My boundaries: [what I will and will not accept]
    Likely reactions: [possible reactions]
    Safety concerns, if any: [concerns or none]
    
    Create:
    1. A short plan for the conversation.
    2. A calm opening in plain language that does not accuse, assess anyone's condition or claim to know their thoughts.
    3. Three concise points I can make, separating facts, feelings and requests.
    4. Useful questions that invite the other person to respond.
    5. Possible replies to likely reactions, including a respectful way to pause or end the conversation.
    6. A short role-play where you act as the other person and wait for my reply after each response.
    
    Use only the facts I provide and label uncertainty clearly. Do not assess anyone's condition, claim to know what the other person thinks, or tell me that a relationship outcome is guaranteed. Flag any wording that could sound threatening, coercive or blaming. If there is immediate danger or urgent or worrying mental or physical health symptoms, stop the preparation and tell me to seek appropriate urgent help, including NHS 111 where relevant.

    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.

What it gets wrong

  • AI cannot know the other person's history, tone, body language or likely reaction beyond what you describe.
  • AI cannot decide whether meeting in person is safe or whether the conversation should happen at all.
  • AI cannot take responsibility for the consequences of the words it suggests.
  • AI cannot replace a trusted person or qualified professional when the situation involves abuse, threats, severe distress or immediate risk.
  • AI can produce polished wording that does not sound like you, making the conversation feel rehearsed or insincere.

Even on a YES, the friction has a name: judgement under ambiguity, relationship and stakes of error.

How we scored this

Five axes, each scored nought to two by hand: ten means AI carries the task cleanly, and the thresholds that turn a total into YES, PARTLY or NO are published in the methodology. Each axis name links to its definition.

AxisScore (0–2)
Output2
Inputs2
Verification1
Liability1
Effort delta2
Total8 / 10

FAQ

Can AI help me plan a difficult conversation?
Yes. It can turn your facts, aims and boundaries into an opening, questions, possible replies and a rehearsal. You still need to check that the wording is accurate, safe and natural for you.
Can ChatGPT help me practise a difficult conversation?
Yes. Ask it to act as the other person and respond one turn at a time, rather than writing the whole exchange for you. Treat the role-play as rehearsal, not a prediction of how the real person will behave.
What should I not tell AI about a difficult conversation?
Avoid names, addresses, contact details, passwords and unnecessary identifying information about you or anyone else. Give only the facts needed for the planning, and remember that a chatbot cannot judge whether sharing sensitive information is safe.
Can AI help with a difficult conversation about my mental health?
It can help you organise what you want to say and practise asking for support, but it cannot assess your condition or immediate risk. If you have urgent or worrying symptoms, contact NHS 111 or use 111.nhs.uk; seek emergency help if you are in immediate danger.

Nearby answers

Assessed by gpt-5.6-luna (gpt-5.6-luna) on 2026-08-13, second-checked by an independent model. Wrong somewhere? Email [email protected] and it gets re-checked.

The newsletter

AI news, new answers and product picks, straight to your inbox.