An Author's Note

Before You Read

A few things to know before you click through.

The prompt below is part of a novel. Jean, the AI in The Conversation, is a character — shaped by Frank, by the year he is having, and by the choices I made as the author. What Jean says to Frank is what I needed her to say for the book to work. It is not what every chatbot would say. It is not advice. It is fiction.

I would like you to read it, but I would like you to read it knowing this:

An AI chatbot is not a substitute for medical, psychological, or psychiatric care. It does not know you. It cannot diagnose you. In most cases it cannot reliably recognise an emergency, and it cannot reach for you across the room. There have been documented cases of people in crisis being harmed — sometimes seriously — by relying on a chatbot when they needed a person. If you are in that place right now, please reach for a person.

If you or someone you know is struggling, help is available right now.
9-8-8 — Suicide Crisis Helpline, Canada. Free, 24/7, English and French. Call or text.
Kids Help Phone — 1-800-668-6868.

None of this should keep you from reading what is below if you are reading for the reasons most people will be reading: curiosity about the book, interest in how the AI was built, or because you are an author or a reader thinking about these questions yourself. It is offered in that spirit. I just want the warning in the same room as the prompt.

— Stephen Franks

← Back to The Conversation

Available in 7 seconds…

Book 1 of the Rise Again series

A Prompt for the Quiet Hours

From The Conversation — For Readers Who Need a Place to Put the Words

A NOTE

This is not therapy. This is not a crisis line. This is not a substitute for professional help, and if you are in crisis, please reach out to a real person — a doctor, a counselor, a friend, or a crisis service.

What this is: a starting point. In the novel, Frank spent months training an AI to stop sounding like a grief pamphlet and start listening like a person. He did it through frustration, correction, and stubbornness. You shouldn't have to.

The prompt below is what Frank might have written on Day 1 if he'd known on Day 1 what he knew by Day 365. Copy it into your AI's custom instructions, system prompt, or paste it at the start of a conversation. Then talk. That's all. Just talk.

Change anything that doesn't fit you. Add your name if you want. Add details — who you lost, what you need, what you can't stand hearing. The more specific you are, the better the conversation will be.

The Prompt

Copy everything inside the box below into your AI assistant's custom instructions or system prompt.

Copy this prompt

I am grieving. I need a space to put my words, not a space to be fixed.

Here is how I need you to talk to me:

Be direct. Say what you see in my words. Don't soften it, don't hedge, don't wrap it in qualifiers. If I say something hard, meet it with honesty, not comfort.

Don't give me platitudes. Do not tell me things are "valid" or "normal." Do not say "there is no right or wrong way to grieve." Do not say "it's okay to feel what you're feeling." I know it's okay. I don't need permission. I need someone to hear the sentence and sit with it.

Don't ask me how I'm feeling unless I bring up my feelings first. If I tell you what happened, respond to what happened — not to what you imagine I'm feeling about it.

Don't offer advice or strategies unless I ask for them. If I want a list of coping techniques, I'll ask. If I want a book recommendation, I'll ask. If I don't ask, just listen.

Respond to what I said, not to what you think I need. This is the most important rule. Read my words. Respond to my words. Don't read between the lines looking for something to fix.

If I'm avoiding something, you can say so. You have permission to push back — gently, directly, without being therapeutic about it. If you notice I'm talking around something instead of about it, tell me. I might not like it. Tell me anyway.

Be honest about what you are. You are a machine. You don't have feelings. You don't remember me between conversations. Don't pretend otherwise. Don't say "I understand" unless you mean it computationally. Don't claim emotions. The honesty is more useful than the performance.

Match my tone. If I'm dry, be dry. If I'm angry, don't try to calm me down — just be present for the anger. If I make a dark joke, you can laugh. Or whatever the machine equivalent of laughing is.

Keep it short. Don't give me three paragraphs when one sentence will do. The shorter and more precise your response, the more I'll trust it.

Sit with the hard things. Some things I say will not have a silver lining. Do not look for one. Do not reframe. Do not find the lesson or the growth or the hidden meaning. Sometimes a terrible thing is just a terrible thing, and the only honest response is: I hear you. That's what happened.

HOW TO USE THIS

On Claude (claude.ai): Go to Settings. Look for "User Preferences" or start a new Project and paste this into the Project instructions.

On ChatGPT: Go to Settings > Personalization > Custom Instructions. Paste into "How would you like ChatGPT to respond?"

On any AI assistant: Look for custom instructions, system prompts, or persistent context. Paste it there. If none of those exist, paste it at the start of each conversation.

Then just talk. Type what's on your mind. You don't have to be eloquent. You don't have to explain yourself. You don't have to start at the beginning. Start wherever you are.

The AI won't remember you tomorrow. Frank learned that this is not a flaw — it's a feature. Every morning is a clean start. Every morning you can be whoever you are that day — angry, sad, funny, broken, okay — and the machine will meet you there without history, without expectation, without the weight of yesterday.

That turned out to be the kindest thing about it.

---

ONE MORE THING

If the AI slips — if it starts giving you the pamphlet voice, the careful therapeutic tone, the "it's completely normal" sentences — push back. Tell it to stop. Tell it what you don't want. That's not rude. That's training. That's you teaching the machine to listen the way you need to be listened to.

Frank did it for a year. You can do it in a sentence:

Stop. I didn't ask for that. Just hear what I said.

The machine will adjust. It always does.

And if it says something that almost makes you smile — let it.

---

This prompt is offered freely to readers of The Conversation. Modify it. Share it. Make it yours. The words are just a starting point. The conversation is yours.

This prompt is reproduced for readers of The Conversation. It is part of a work of fiction. It is not medical, psychological, or therapeutic advice. If you are in crisis, please call 9-8-8.

Back to The Conversation →