Game Jam AI Prompts: Come Up With and Scope Your Game

Published 2026-10-15 · by AgentKit Co

The theme drops. The clock starts. In the first hour of a game jam, you'll make the decisions that decide whether you ship or not: which idea, how big, and what to cut. AI can help with all three, if you ask it the right questions. These prompts are built for that first crucial stretch.

Replace the [brackets] and paste into any chat assistant.

Note on originality: use AI for brainstorming and planning, and make the final creative decisions yourself. Check your jam's rules about AI use and assets before you start. Rules differ from jam to jam.

Start with a design brief

Give the assistant your constraints once, at the top of the chat:

Design brief: Engine: [Unity/Godot/Unreal/GameMaker/other].
Time: [hours]. Team: [solo/people]. My skills: [code/art/audio].
Theme: [theme]. Keep every answer scoped to what I can build in
this time. If a number is an estimate, label it as an estimate.

Without constraints, AI happily suggests an open-world RPG for a 48-hour jam.

1. Theme to ten hooks

Don't ask for "a game about [theme]." Ask for range:

Give me 10 ideas that use the theme "[theme]" in different ways:
literal, metaphorical, as a core mechanic, as a rule applied to
the player, as a visual style. One sentence each. Then pick the 3
I can actually build in [hours] with [skills] and say why.

Why it works: the first idea everyone has is the literal one. Forcing five different interpretations gets you to ideas other jammers won't have.

2. Find the core loop

For this idea: [one sentence], describe the core loop: what the
player does every 10 seconds, every minute, and over the whole
session. What's the one verb the game is about? Suggest the
smallest version that is still fun.

If you can't state the core loop in a sentence, you can't build it in a weekend.

3. The scope killer

Here's my feature list and my total hours. Estimate each feature,
cut until the total fits with a 25% buffer, and suggest a cheap
substitute for each feature you cut.
[paste list + hours]

Why it works: a finished small game beats an unfinished big one every time. The 25% buffer is there because things always take longer than you think.

4. Mechanic specs before code

Write a one-page spec for this mechanic: [mechanic]. Include
inputs, rules, edge cases, and how I'd test it in 5 minutes.
Keep it implementable by [skills] in [engine].

A short spec prevents the two-hour detour where you discover you never decided what happens when the player does something odd.

5. Guide players without text

My players keep missing [goal/exit/item]. Suggest 5 ways to guide
them without tutorial text: light, color, framing, sound, level
layout. Rank by effort.

Try a lamp above the exit, one accent colour used only for climbable surfaces, or a doorway that frames the goal. These cost far less than building a tutorial.

6. Playtest plan in 15 minutes

I have 3 friends and 20 minutes. Write a playtest plan: what to
watch for, what NOT to say, 5 unbiased questions to ask afterward,
and how to turn the notes into a prioritized fix list.

7. The jam timeline

Build a timeline for a [hours]-hour jam: when to decide the idea,
when to have the core loop playable, when to stop adding features
(feature freeze), when to polish and when to build and upload.
Mark the points where I should cut scope if I'm behind.

8. Game page copy and final checks

Here's my game: [summary]. Write a game page description under
150 words, three screenshot caption ideas, and a checklist for my
final build: export settings, file sizes, controls listed on the
page, assets and licenses credited.

Many jam games lose votes because the page is unclear, or the download didn't work. A checklist avoids that.

Rules of thumb for AI in jams

1. Stay original. Use AI to explore, not to copy another game.
2. Check the rules. Some jams restrict AI-generated assets or code. Read before you start.
3. Verify tech advice. Check anything engine-specific in the official docs. AI sometimes invents function names.
4. Label estimates. Hours and balance numbers from AI are guesses. Test them.

If you're thinking about mods rather than jam games, the same approach applies: check each game's modding rules and EULA first.

Go deeper

The Prompt Pack for Game Design and Modding follows a project from theme announcement to post-jam retro with 40 prompts across eight sections: ideation, mechanics, level design, narrative, balancing, modding, playtesting and shipping. It works with any engine, and most prompts also work for tabletop games.

Building an agent to help with this? See how to write a system prompt for an AI agent for the template.

Keep reading