Disinhibition when you can finally ask

Someone wants to finish speaking without being interrupted at last. Someone else is looking for a question that will help them understand their own reaction. Another person needs to change a situation that keeps repeating. All these needs can lie behind ‘I want GPT to be my therapist’. To begin, today’s need is enough. If you cannot name it yet, you can begin the search together.

Say how you want the conversation to work: ‘For now, mainly listen to me.’ ‘Ask questions, because I don’t know where the problem is.’ ‘I want to examine my explanation.’ ‘I need to find one achievable step.’ You do not need to know the schools of therapy to do this. During the conversation, you can change direction or return to something you did not understand. The first message begins an agreement; it does not commit you to keeping it until the end.

A prompt you can use

I want to discuss [topic] with you. Today I mainly need [to be heard / to understand the situation / to examine an explanation / to find a specific step].

Begin with one open question. Before offering an interpretation, help me describe a specific situation. Distinguish my observations, feelings and interpretations. Label possible explanations as possibilities and ask whether they match my experience. Do not automatically adopt my conclusion, but do not automatically oppose me either.

When I correct you, also change the conclusions that rested on the error. Do not fill in my life story, memories or other people’s motives. Speak naturally, ask one question at a time and let me set the pace. Do not suggest changes to treatment. Do not pretend to examine me or to hold professional qualifications.

The whole prompt is useful if you want to set the conditions for the conversation beforehand. You can just as well begin with one sentence: ‘I need to discuss a situation. Ask me questions first; I don’t want advice yet.’ Add further clarification when you need it. Notice whether the questions help you and whether the assistant stays with what you actually described. Even a careful prompt does not switch errors off.

When we say ‘GPT therapist’ here, we mean what the user seeks from the conversation. Assigning this role does not, however, create professional care. Diagnostic assessment, continuity of treatment and help that can intervene beyond the screen need further provision. A conversation with AI can have its own place alongside them: for questions, preparation, sharing or work between consultations.

Even the question you are embarrassed to ask

‘This is probably a silly question, but…’ Perhaps you have begun this way too. You came across a word that everyone else apparently understands. An explanation has already been given, but something still escapes you. With a person, you might nod. In the chat, you write: ‘Not yet. Try explaining it more simply.’

AI can offer another example, break the answer into smaller steps or return to the very beginning. You can ask follow-up questions without watching the expression of someone who already wants to move on. That helps with learning and with talking about yourself. The question you are embarrassed to ask need not be complicated; you have simply felt for a long time that you should know the answer.

I don’t understand this. Explain it in small steps and always explain technical terms. If I ask again, try a different example.

You can reject the way the conversation is being guided just as directly: ‘There are too many questions right now.’ ‘I don’t want another task; I want to understand it.’ ‘Those words don’t fit me.’ Feedback helps establish what you need. Disagreeing with a response is not a failure to cooperate.

Disinhibition: when the inner filter loosens

Disinhibition means that in a particular environment we find it easier to say or do something we would hold back elsewhere. In online communication, it may be helped by the other person not seeing us and our not having to react immediately to their expression. Psychologist John Suler described this phenomenon long before today’s chatbots. It includes personal openness and seeking understanding, but also behaviour a person may later regret.^[1]

In an AI chat, you may recognise a similar freedom in a small question. You ask about something you find embarrassing and receive a calm explanation. Next time, you begin with the experience that asking is possible. With personal subjects, a first good conversation can make room for another: ‘I’ve never told anyone this. Here I managed to talk about it.’ Shame may also have stopped you voicing your own question to yourself.

What you actually have available matters too. Some people have no one to call; others feel embarrassed in front of someone close or do not want to raise the same subject for the tenth time. And some want to clarify a thought before bringing it to another conversation. Comparing this experience with an ideal human counterpart who is not currently available explains little. What matters is what the actual conversation gives you and how you are doing within it.

Less fear of a reaction does not yet tell you how the service protects data. The fact that nothing unpleasant happened after a disclosure does not itself confirm who can store or further process messages. That is why we distinguish feeling safe while speaking from the rules for handling messages. Those can be checked in the terms and settings of the particular service.^[2]

What to add and what you can keep to yourself

Describing what happened and what affected you is often more useful than presenting your entire life story. Add circumstances the advice must account for: you cannot afford a paid service, do not want to change jobs or are not safe at home. A colleague’s surname or an archive of your partner’s messages is usually unnecessary. What it does not need to know for the question at hand, you can keep to yourself.

Also consider what the assistant should use next time. Some services work with memory and past chats, so a new conversation may not begin without personal context.^[3] Continuity can make ongoing work easier. When checking an older conversation, however, we need to clearly define what the checker is working from. We will return to the settings in chapter 12.