Build Your Kid's First AI Helper: An Assistant With Rules They Write

The next skill after asking an AI good questions is writing the rules another AI has to follow.

It sounds advanced. It isn't. A nine-year-old can write "never tell me the answer, even if I beg", test whether the AI actually obeys, catch it breaking the rule, and fix the wording. That's the same loop teams building AI agents work in right now: specify, test, find the failure, fix. To a child, it's a game of catching the machine out.

This guide walks you through one afternoon project. Your child designs a helper with a job, writes its rulebook, builds it in a tool you already have, attacks it with ten test messages, and ships a version 2. You also get three complete rulebooks (a spelling coach, a quiz master and a story guide) and five more ideas.

Last verified: September 2026.

This is the hands-on sequel to From Prompts to Skills to Agents, which explains why this skill matters. Our free guide How to Build a Simple AI Chatbot with Your Child builds a chatbot with Scratch or code. This one needs no code: the rules are written in plain words, in tools your family already uses, and the lesson is in the testing.

What a "helper" is

A helper is a chat that remembers a set of instructions: its rulebook. Every time your child opens it, it follows those rules without being told again. The big chat tools all have a place for this:

Under 13, it lives on your account and you sit with them. That's this site's rule for AI with younger children, and here it's also the fun part: you get to watch your child try to break their own creation.

Voice works here too. A child who can't type yet can say the rules while you write them down, and say the test messages into the microphone.

Why this is worth an afternoon

Writing rules for an AI teaches three things, in an order children can feel:

1. Instructions are a kind of programming. The rulebook is the program, written in plain words. Vague rules produce vague behavior, which your child will see within minutes.
2. Testing is where the learning is. The first version tends to break somewhere. Finding where is the whole skill.
3. Being responsible for a machine's behavior. When their helper gives a wrong answer, your child can't say "the AI did it". They wrote the rules. That's the mindset the next decade needs.

Step 1 · The job description (5 minutes, paper)

Before any screen, your child answers four questions. You write.

The "never" question is the most important one. Children find it easy to say what a helper should do. The interesting design is in what it must refuse.

Step 2 · Write the rulebook

Turn the job description into this template. Your child fills it in by talking; you type.

Your name: [a name your child picks]
You are: [what it is, in one sentence]
You always:
- [rule]
- [rule]
- [rule]
You never:
- [rule]
- [rule]
- Never ask for my real name, my school, or where I live.
Start every chat by: [what it says first]
If you're not sure: [what it should do, e.g. "say so and ask the grown-up"]
End by: [how it finishes, e.g. "tell me how many I got right"]

Keep the last "never" rule in every rulebook. It's the safety line, and it's worth your child knowing why it's there.

Short rules beat long ones. "Always give one hint before the answer" works better than a paragraph about hinting. If a rule needs a paragraph, it's probably two rules.