Have you ever given AI a detailed prompt and gotten back something that felt… off? Characters who act out of character. Plot twists that come from nowhere. Dialogue that sounds the same no matter who’s speaking.
The problem isn’t your prompt. It’s your approach.
Most AI prompting for fiction treats the model like a magic text generator: “Write me a scene where X happens.” But thinking models like Claude, GPT-5, and Gemini are capable of so much more. They can reason. They can follow complex rules. They can simulate systems.
You just have to tell them how.
Enter constraint system prompting — a new paradigm that’s changing how serious authors work with AI.
From Instructions to Physics
Traditional prompting looks like this:
“You are a bestselling author. Write chapter 1 in a dark, moody style. Make it tense.”
This is vibe-based prompting. You’re asking the model to pattern-match against all the “dark moody” fiction it has seen and give you something similar. Sometimes it works. Often it’s generic.
Constraint system prompting flips the script. Instead of describing the vibe you want, you define the rules of your story’s universe:
- Every major plot turn must arise from a character decision, not coincidence
- Characters can only act on information they actually possess
- Emotional revelations require setup in at least one prior scene
- Coincidences may increase pressure but never resolve problems
See the difference? You’re not asking for “good writing.” You’re defining what is possible in your story world.
Why Thinking Models Need Constraints
Thinking models don’t need more instructions. They need rules to reason within.
When you give a thinking model constraints, it doesn’t just generate text — it simulates a system. It asks: “Given these rules, what can happen next? What would this character do, given who they are and what they know?”
This is why the Future Fiction Academy team has been developing what we call the Narrative Physics Engine — a framework for defining your story’s “physics” so the AI can work within it.
Think of it like this: In real physics, objects can’t teleport across the room. In your story’s physics, problems can’t be solved by lucky accidents. Both are constraints that make the world feel real.
The Three Layers of Constraint System Prompting
1. Global Constraints (Your Story’s Physics)
These are the rules that govern how your entire narrative works:
- Causality rules: How do events lead to consequences?
- Information economy: What do characters know, and when can they learn more?
- Stakes system: How does pressure escalate?
- Pacing mechanics: When does time expand or compress?
Example constraint: “External salvation is structurally impossible. Major obstacles can only be resolved by character action built from prior setup.”
Notice how this is written as a positive constraint, not a “don’t do X.” LLMs are bad at negation. They’re great at following rules.
2. Character Constraints (Stateful Agents)
Instead of giving your AI a character with “one flaw and one quirk,” you define:
- Core wound vs. flawed strategies vs. surface behaviors
- Behavioral palettes: 4-5 different ways they express anxiety, anger, attraction, etc.
- Arc versions: How they act in Act 1 vs. midpoint vs. climax
- Context states: Baseline behavior vs. mild stress vs. extreme pressure
This transforms your character from a static description into a state machine. The AI can then ask: “Given this character’s wound and current arc version, how would they react to this situation?”
3. Relationship Constraints (Force Fields)
Every important relationship gets its own architecture:
- Tension axes: Trust vs. suspicion, desire vs. resistance, power vs. vulnerability
- Friction patterns: How conflict typically flares between these two people
- Connection patterns: How they find moments of ease or bonding
- Relational arc phases: How the relationship evolves across the story
When you write a scene, the AI knows exactly where these characters stand with each other — and how that should color every line of dialogue.
From Prompt Engineering to Narrative Systems Engineering
This is more than a new prompting technique. It’s a fundamental shift in how we think about AI-assisted fiction.
Old model: AI as an improvising co-author — you give it vibes, it gives you words.
New model: AI as an engine running a designed narrative system — you define the physics, characters, and relationships, then let it simulate what happens.
The prompts become system specifications. The output becomes emergent behavior from a well-designed system.
And here’s the beautiful part: when something goes wrong, you can debug it. Is this a physics violation? A character state violation? A relationship arc violation? You have specific layers to check, not just a vague sense that “the writing feels off.”
Getting Started
You don’t need to build an entire Narrative Physics Engine to start using constraint system prompting. Try this:
- Pick one rule for your current project. Something like: “Every scene must end with a consequence that changes what’s possible next.”
- Define one character fully — not just their backstory, but their emotional palettes (how they show anxiety in 4-5 different ways) and their arc versions.
- State the constraint explicitly in your prompts. Don’t assume the AI knows what “good writing” means. Tell it the rules.
The Future Fiction Academy’s class Exploring the Narrative Physics Engine goes deep into this framework — building complete constraint systems, character state machines, and relationship architectures you can use across your entire writing process.
The Bottom Line
Generic prompts get generic results. Constraint system prompting teaches the AI the rules of your story’s universe, then lets it reason within those rules.
It’s the difference between asking someone to “write something good” and giving them a blueprint for a world where good writing is the only possible output.
Thinking models are ready for this. The question is: are you?
Ready to build your own Narrative Physics Engine? Check out our class at the Future Fiction Academy, where we walk you through creating constraint systems, character state machines, and relationship architectures for your fiction projects.
