BRAINLAB
Log in

← Field notes

How to write a game prompt your 7-year-old will love (5 patterns that actually work)

2026-05-04 · Louis · 8 min read

Most kid-generated AI prompts fail in one of three ways: too vague (a cool game), too crowded (a six-clause run-on with two protagonists and four mechanics), or too uncommitted (a game with maybe a cat or maybe a dog). The result is usually a game the kid is disappointed by, followed by a request to try again, followed by their losing interest before the second attempt finishes.

This is fixable. After six months and roughly four hundred family prompts, the patterns that produce oh wow reactions consistently follow one of these shapes. We use them ourselves. Copy them.

Pattern 1: Noun–verb–feel

The smallest, most reliable prompt. Three pieces: a noun (who or what), a verb (what they do), a feel (the mood). Comma-separated. Done.

a sleepy bear, collecting berries, very cozy

a robot, dodging lasers, fast and shiny

a hot dog, running from a cat, silly

The reason this works: each slot constrains a different layer of the game (character, mechanic, tone). The AI doesn't have to guess at any of them. The result lands closer to what the kid pictured because the kid actually said it.

Pattern 2: Familiar-thing-with-a-twist

Take a thing the kid already knows — Flappy Bird, a maze, hide-and-seek, a quiz — and add one specific twist. Not three twists. One.

flappy bird, but the bird is a slice of pizza

a maze game, but the walls move

a memory match game, but the cards have animal sounds

Why it works: the familiar piece gives the AI a clear archetype to instantiate. The twist gives the kid ownership. The kid feels like they invented it because, in the only way that matters to a five-year-old, they did.

Pattern 3: Subject-of-week

Take whatever your kid is currently obsessed with — dinosaurs, mermaids, the Titanic, soccer, our cat Nori — and use that as the protagonist. The genre doesn't matter much; the protagonist does.

a maths quiz where every question is about dinosaurs

a platformer starring Nori the cat, who collects fish

a paint-the-mermaid colouring game, lots of pinks and blues

This is the most reliable way to make educational content not feel educational. A maths quiz is broccoli; a maths quiz about dinosaurs is broccoli wrapped in a steak.

Pattern 4: Tiny-emotional-arc

The advanced move. Instead of describing a mechanic, describe what the kid should feel at the start, middle, and end. The AI is surprisingly good at translating arc into mechanic.

a game that starts cute, gets scary, then ends with a hug

a game where you feel proud at the end because you helped someone

a game that gets faster and faster until you laugh

Kids over seven really shine here. They have ideas about how a game should feel they couldn't articulate as a six-year-old. This pattern unlocks them.

Pattern 5: Iteration-by-instruction

Not a starting prompt — a follow-up prompt. After the first game lands, the magic move is to not ask for a new game. Ask for one specific change to the one you have.

make the cat purple

add a score that goes up when you catch a mouse

make the music slower and the background nighttime

This is where the actual creative skill compounds. Kids who learn to iterate end up understanding cause and effect in design at an age where most of their peers are still describing things in finished form. Pattern 5 is the single highest-leverage habit we've seen.

Anti-patterns to avoid

The wishlist prompt. a game with a dragon and a princess and a knight and a wizard and also a robot and the dragon can fly and breathe fire and the princess can shoot bows. The AI tries to honour all of it and the result is a chaotic mess. Help your kid pick the one piece they care about most and leave the rest for the second iteration.

The unspecified noun. a game with cool stuff in it. The AI's idea of cool stuff is unlikely to match a six-year-old's. Force one concrete noun into the prompt.

The negation prompt. a game that's not boring. The AI cannot reliably reason about negation, and your kid can't either. Replace every not X with a positive: instead of not boring, write silly or fast.

Prompts in a different voice. When parents type "for" their kids, they often quietly upgrade the language and remove the kid-isms. The kid then feels less ownership of the result. Type what the kid actually said. Misspellings, weird grammar, the noun they invented. The AI handles it fine and the kid recognises it as theirs.

A worked example

My six-year-old said, last Tuesday: "I want a game where the cats win the war against the dogs but the dogs are bigger so the cats have to be very clever."

That's not bad. It's pattern 1 + pattern 4 stitched together. Noun (cats), verb (win the war), feel (clever, asymmetric). I typed exactly that. The AI built a stealth puzzle where small cat sprites sneak past large dog sprites by hiding behind objects. She played it for forty minutes. Two iterations later (pattern 5: add a level where it's nighttime, then make the dogs sleep sometimes) it was a complete game.

I did not edit her prompt. I did not "fix" the grammar. The cats won the war. She got to bed at 8:15.

— Louis, prompt nerd / part-time editor of his daughter's intent.