Parent AI Prompt Builder

This tool needs JavaScript to build your sheet, and without it nothing on this page builds one.

The short version to write on paper: The method needs nothing from this site. You can complete it entirely without an account or a device. On any piece of paper: what you actually want out of the answer, what the assistant must not do, and what you will check before you act on any of it. And if somebody might be at risk right now, that goes to the emergency or local service in the place where that person is, not to an assistant and not to a prompt.

First: what kind of question is this?

Some questions should not be put to an assistant at all. This is where that gets decided, before anything is written.

An ordinary question about family life I want it to make a medical, legal or safeguarding decision for me Someone might be at risk right now

This one does not become a prompt.

If what you want is a medical, legal or safeguarding decision, no prompt fixes that. A diagnosis, a dose, a medicine, a treatment, a pregnancy, whether a product is safe, significant symptoms, and school, medical or legal rights all need the right qualified person. Never change a prescribed treatment on the strength of a generated answer. A model does not examine your child, does not carry the duty, and cannot be held to the answer. What it can genuinely help with is getting you ready for the person who does.

So, for the decision case and only once nobody is at risk: come back to the first question, pick "An ordinary question about family life", then choose the goal called "Get ready for an appointment with a professional". That builds a prompt for questions to ask, not answers to act on. If someone is in danger, this page waits.

Stop here. This needs a person now, not a prompt.

If someone might be at risk right now, an assistant is the wrong place entirely. Possible poisoning, trouble breathing, a severe symptom, someone unconscious, immediate risk of self-harm, violence or abuse: that goes to the emergency or local service in the place where that person is, now. Do not stop to write a better prompt or to collect details first.

Nothing saved yet

What is the question about?

A few words. No name, no birth date, no school, no address, no diagnosis, no photograph.

The subject, in your own words

For instance "bedtime has been taking two hours" or "a wet Saturday and two children who are bored". If you find yourself typing something that identifies your child, that is the sign to write it more generally.

Roughly how old is the child this is about? (Optional)

Optional, and a band rather than a birth date. Give it only when the answer would genuinely change with age. Leave it out and the prompt asks the assistant to keep things general instead of guessing an age.

Country or system, only if it matters here (Optional)

Optional. No address or identifying details.

What do you want out of it?

This is the choice that changes the prompt most. Each of these produces a different structure, because they are different requests.

The goal

Ideas for something to do Get ready for an appointment with a professional Have something explained in plain words Find words for a conversation with my child Weigh up options I am already choosing between

What shape should the ideas take?

Asking for a shape is the cheapest way to get back something you can actually use. The options change with the goal, because a shape that suits one request contradicts another.

The shape you want

A short list, five things at most A list with a line on what each one needs A few plain sentences, no list

What shape should the explanation take?

Asking for a shape is the cheapest way to get back something you can actually use.

The shape you want

A few plain sentences, no list A short list, five points at most Steps, in the order they happen What is well established first, then anything still open

How should the openings come back?

Openings, in every case. A script is the one shape this goal cannot use.

The shape you want

A few openings, one line each A few openings, each with a line on when it fits A few plain sentences I can put in my own words

How should the comparison come back?

Each of these shapes sets the choices side by side. None of them asks for a verdict.

The shape you want

The case for and the case against Side by side, point by point A few plain sentences, no list

How should the questions come back?

Every option here is questions and nothing else. Preparing for an appointment is a different job from replacing one, and a prompt that leaves room for answers quietly turns the first into the second.

The shape you want

Questions only, in the order I would ask them Questions only, grouped by theme Questions, plus what to write down before I go

Your sheet is ready

Your prompt

Everything below can be edited. Change the words before you print it.

Go to the print buttons

What this will not do

It does not answer anything. It has no model behind it, sends nothing anywhere, and cannot see the reply you get.
Leaving names out is not anonymization. The provider still sees your account, your device and everything else in that conversation, and a small set of details can identify a family. Once you press send, where the text goes next is not your decision.
It will not write a prompt that asks a machine to diagnose, prescribe, or make a legal or safeguarding decision. Those are not prompt problems.
The lines it puts under your prompt are instructions, not controls. An assistant can ignore every one of them, so read the answer as though it had.
An answer that sounds confident is not therefore correct, and an error can be delivered in a perfectly sure tone. The checks at the end of your sheet are not optional politeness, they are the point.

The sheet above is what prints, without the questions or the buttons.

What to do next

Paste the prompt into whichever assistant you use, limits included. They are instructions, not settings: an assistant can drift past them, so the answer gets read against them rather than trusted because they were there.

Run the four checks before you act on anything. If the answer will not survive being looked up somewhere else, it was not an answer.

If the reply starts telling you what is wrong with your child, close it and take the questions to a person instead.

Finish on the sheet, in the four lines left blank under the checks: what you contributed, the source you opened and checked, what is still uncertain, and the person to contact with the next question. That last line is the one you will want next week.

Where this sheet stops

Nothing you type here leaves your device. What you paste into an assistant afterwards does leave it, and what happens to it then follows that service's rules rather than yours. Decide what your child would want in there before you press send.

What this will not do

It does not answer anything. It has no model behind it, sends nothing anywhere, and cannot see the reply you get.
Leaving names out is not anonymization. The provider still sees your account, your device and everything else in that conversation, and a small set of details can identify a family. Once you press send, where the text goes next is not your decision.
It will not write a prompt that asks a machine to diagnose, prescribe, or make a legal or safeguarding decision. Those are not prompt problems.
The lines it puts under your prompt are instructions, not controls. An assistant can ignore every one of them, so read the answer as though it had.
An answer that sounds confident is not therefore correct, and an error can be delivered in a perfectly sure tone. The checks at the end of your sheet are not optional politeness, they are the point.

Privacy and sources

Everything you type is worked out on this device. None of it is sent anywhere, and saving is a button you press on purpose.

Sources of this tool

Guidance on AI and children (version 3.0)
UNICEF Innocenti. Backs data minimization and human oversight as defaults.
AI Risk Management Framework: Generative AI Profile (NIST AI 600-1)
NIST. Source of the check-before-you-act list. Written for organizations deploying systems, adapted here for a household.
Health advisory: Use of generative AI chatbots and wellness applications for mental health
American Psychological Association. Backs refusing to build a prompt that asks a chatbot to stand in for a clinician or a crisis service. An advisory about consumer chatbots and wellness apps used for mental health, across age groups. It is not a rule about every use of a model.

Written for any country: nothing on it is specific to where you live yet.

The book's references for chapter 32

This one does not become a prompt.

Stop here. This needs a person now, not a prompt.

This one does not become a prompt.

Stop here. This needs a person now, not a prompt.

Your prompt

Saved in this browser only

What to do next

Sources of this tool