Invent a Toy and Write Its Instruction Manual

Anyone can invent a toy. Writing instructions somebody else can follow is the hard part, and it is the one skill in this project that transfers to everything.

How it goes

1. Read a real manual together. Anything in the house. Note the parts list, the numbered steps, the warnings, the diagrams with arrows.
2. Your child invents the toy and draws it with every part labelled.
3. Write the manual by hand, in the real one’s format: what is in the box, how to set it up, how to play, how to put it away, one safety warning.
4. Test it on a person. You follow the instructions exactly as written, with no interpretation. Step three says “put it together” and you stand there doing nothing. This is the funniest and most instructive part of the afternoon.
5. Ask the AI to be a second tester: Read my instructions and tell me which step could be understood two different ways. Do not rewrite them.
6. Your child rewrites the ambiguous steps.

Who does what

AI finds ambiguity, which is a genuine strength, and formats a parts list without complaint.

Your child invents, draws, writes every step, watches you fail, and rewrites.

You follow the instructions with malicious literalism and refuse to guess what they meant.

Make it easier or harder

For a six-year-old: cut the manual to three numbered steps and a parts list, and let them dictate while you write. The comedy of you following the instructions literally still lands, and dictation removes the writing barrier that would otherwise end the project at step three.

For a twelve-year-old: add a troubleshooting table, which is the part of a real manual nobody reads and everybody needs. Three rows: what went wrong, why it might have happened, what to try. Writing plausible failure modes for an imaginary toy requires understanding how it is supposed to work, which is harder than describing assembly.

Take it off the screen

Build the toy from card and check whether the assembly steps were right. Instructions written before building are always wrong somewhere.

Swap manuals with a sibling or a friend and build each other’s toys from the instructions alone.

When it goes sideways

If your literal-instruction performance upsets rather than amuses, stop it immediately and switch to asking questions instead: When you say put it together, which bit goes on first? Some children find being deliberately misunderstood funny and some find it humiliating, and you will know which within about fifteen seconds.

If the AI rewrites the manual despite being told not to, ignore its version entirely and ask a narrower question: Which single step is the most ambiguous? One target at a time keeps your child editing their own words rather than accepting somebody else’s.

One thing to watch

If your child asks the AI to write the manual, the project evaporates: the machine writes unambiguous instructions easily, and the whole exercise was learning to do that themselves. Keep it to the tester role.

Related: invent a family gadget supplies the thing to write about, and design a mystery puzzle uses the same “test it on a real person” step.