Custom Prompting
Why I started
I started Project CLEAR because I wanted to understand how a few careful words can change the way a chatbot responds.
A prompt is a set of instructions that tells the AI what kind of help I am asking for, what limits it should respect, and what I still need to decide myself.
Build a safer prompt
When I write a prompt, I name the task, give only the context that is needed, ask for uncertainty, and say what the chatbot should not do.
I avoid sharing names, private records, passwords, or details about another person without permission. I also ask the chatbot to show more than one possibility instead of presenting a confident answer as a fact.
Use it as a writing aid
For reflection, I use a safe CBT journaling prompt template.
CBT means cognitive behavioural therapy, but this template is not therapy and it is not a diagnosis. It is a writing aid I made for finding the link between a situation, my thoughts, my feelings, and the actions I want to choose.
A safe template
One safe template I use is, “I am using you as a journaling aid, not as a therapist. Help me reflect without diagnosing me or telling me what I must do. Ask one question at a time. Help me separate what happened from what I assumed. Help me name the feeling and its intensity from zero to ten. Offer two or three balanced alternative thoughts. End by asking what small action could support my safety or wellbeing. Do not encourage secrecy or dependence. If I mention immediate danger, self harm, abuse, or danger to someone else, tell me to contact a trusted adult or emergency support instead of continuing the exercise.”
Check before using it
Before using it, I remove identifying details and decide whether human support would be better.
I do not paste a private message, a medical record, or someone else’s story into a chatbot. If the prompt makes me feel more panicked, stuck, watched, or dependent, I stop.
A good prompt should give me more room to think, not make the screen feel like the only place I can turn.
Leave space to think
I also ask for a pause after the response.
I write my own answer in my notebook before asking for another suggestion. That keeps the chatbot from filling every silence and helps me notice when I agree just because the words sound calm.
The point of prompting is to make the conversation clearer and keep my own judgment active.
My simple pattern
My basic pattern is simple. I say what I need, what information is missing, what boundaries to follow, and how I will check the result.
Then I decide whether the result is useful. Prompting can improve a conversation, but it cannot make a chatbot private, wise, or responsible for me.