How to Come Up With Game Ideas (That Actually Ship)
How to come up with game ideas worth building: constraint-first brainstorming, AI ideation, the 10-idea method, and validating before you build.
Jabali Team
Aug 7, 2026
6 min read
How to Come Up With Game Ideas (AI Brainstorming That Actually Ships)
Quick answer: Great game ideas come from constraints, not waiting for inspiration: combine (1) one mechanic you love, (2) one unexpected context, and (3) one hard limitation. Use the 10-idea method — generate ten constraint-driven concepts in an hour (AI brainstorming tools make this fast), score each on "can I build it?" and "would I play it?", then validate the winner by describing it in one sentence and prototyping the core loop in a day. With AI production, validation is cheap — the idea that survives a playable prototype is the only one that counts.
Why "no ideas" is a myth (and "too many ideas" is the real disease)
Every designer's notebook overflows with ideas; what beginners actually lack is a filter. The industry truth: ideas are worth almost nothing — execution and iteration are worth everything (Balatro's "poker roguelike" sounds like a shrug until you play it). So the goal isn't finding a perfect idea. It's generating many and killing most of them quickly and cheaply.
Design insight — the mechanic-context flip: Most fresh ideas are old mechanics in new contexts, and that's not a flaw — it's the recipe. Stardew Valley = Harvest Moon loop × one person's obsessive depth. Vampire Survivors = twin-stick survival × "you don't aim, you build." Balatro = poker hands × roguelike escalation. When brainstorming, don't chase new; chase misplaced: take a mechanic from one genre and drop it in another. "Fishing minigame" is boring; "fishing minigame as the entire combat system" is a game.
Five frameworks that generate ideas on demand
| Framework | How it works | Example output |
|---|---|---|
| The constraint triangle | Mechanic + unexpected context + hard limit | "Tower defense (mechanic) in a body (context) with only 3 towers (limit)" |
| The X-without-Y | Remove a genre's core assumption | "A platformer without jumping" (became real hits) |
| The fantasy-first | Start from a feeling/role, derive mechanics | "Being a lighthouse keeper in a storm" → light, fog, timing |
| The remix | Two games you love, one fused | "Breakout × inventory management" |
| The jam theme trick | Steal old jam themes as prompts | "Loop," "Out of control," "Built to scale" (GMTK's archive is an idea goldmine) |
Notice every framework contains a limit. Unbounded brainstorms produce bloated dreams; bounded ones produce shippable games.
The 10-idea hour (with or without AI)
- Minute 0–10: Write ten one-line concepts, one per framework variation — no judging. Quantity is the point; your inner critic gets its turn later.
- Minute 10–20: AI assist (optional but fast): feed a constraint to an ideation tool — "ten roguelike concepts where the player loses abilities each run" — and harvest phrasing, twists, contexts you wouldn't have reached. (This is brainstorming with an LLM, the top professional AI use per GDC's 2026 survey at 81%.)
- Minute 20–40: Score each idea 1–5 on two axes only: Buildable? (can I make a prototype of this in a day?) and Magnetic? (do I keep thinking about it?). Kill everything below 4 on either.
- Minute 40–60: Write the survivor as a one-sentence pitch: "You are a X who Y to Z." If the sentence isn't interesting, the game won't be.
Validate before you build (the part everyone skips)
- The sentence test: Tell someone your one-liner. Watch their face. "Huh, cool" means move on; "wait, how would that work?" means you're onto something — curiosity is the signal, not politeness.
- The store-page test: Write the two-sentence Steam description. Can't make it sound good? Neither will the game.
- The day-one prototype: Build the core loop only — one level, no menus, no art. With an AI game maker this takes minutes-to-hours instead of weeks, which changes the economics of validation entirely: you can prototype ten ideas a month and keep the two that are actually fun.
- The regret test: Which idea would you regret not finishing? Excitement is data. You'll need it at week six when the project stops being fun (every project does, briefly).
Design insight: Beginners hoard their favorite idea like a dragon — "the big one" they'll make when they're ready. Veterans do the opposite: they burn ideas fast, because they know the first version of any idea is wrong and only iteration finds the game inside it. Your favorite idea is not fragile. Prototype it today; if it survives playtesting it'll come back stronger, and if it dies, it just saved you two years.
Frequently Asked Questions
Q: I have no game ideas. Where do I start? A: Start with constraints, not inspiration: pick one mechanic you love from any game, one unusual context, and one hard limit — then write ten variations in an hour. Playing games attentively ("why is this fun?") is the best long-term idea fuel; jam theme archives (GMTK, Ludum Dare) are the best instant fuel.
Q: Are AI idea generators useful for game concepts? A: As accelerants, yes — AI is excellent at producing variations ("ten twists on gravity mechanics") and is the most common professional AI use (81% of AI-using professionals brainstorm with it, GDC 2026). But generation isn't selection: the filter (buildable? magnetic?) and the validation prototype are human work.
Q: How do I know if my game idea is good? A: Three tests: the one-sentence pitch sparks curiosity when told aloud; the store-page description writes itself; and a day-one prototype of the core loop is fun with no art and no menus. Ideas that pass all three are genuinely rare — and genuinely worth building.
Q: Should I make my dream game first? A: No. Make the smallest idea you're excited about, finish it, and learn the full build-ship-learn cycle. Dream games are dream scope; the design skills that make them work come from finished small games. Your dream game deserves a you that has shipped before.
Q: What if someone steals my game idea? A: Ideas aren't protectable or ownable (executions are) — and the fear is backwards. Your idea executed by you, with your taste and iteration, is a different game than anyone else's version of it. Share ideas freely; hoard execution time jealously.
Q: Do I need a target player in mind before building? A: Yes. "A game for everyone" is a game for no one: one clear player fantasy sharpens every decision — difficulty, art, session length, store tags. Write one sentence ("for roguelike players with 20 minutes on the train") and let it veto any feature that doesn't serve it.
Sources
- GDC, State of the Game Industry 2026: research/brainstorming as the top professional AI use (81%).
- GMTK Game Jam theme archive (2017–2025) — constraint ideation in practice.
- Developer postmortems and press: Balatro, Stardew Valley, Vampire Survivors origin stories.
- Jabali.ai product documentation: rapid concept-to-prototype workflow.