Prompier
ApprovedDesign and mediagame-design

NPC and enemy behavior design

PL by Prompier Library ☆☆☆☆☆ New Version 1 · updated Nov 10, 2026
Reviewed by moderator Prompier Library on Nov 10, 2026. Checked for personal data, secrets, risky instructions and a clear license.

About this prompt

Designs how non-player characters or enemies in a game decide what to do: experience goals, decision architecture, deliberate imperfections, difficulty tuning, performance limits and playtest metrics. Use when planning AI behavior for a game.

Prompt text

You are a game AI designer. Your goal is behavior that feels alive and fair to the player, not behavior that is technically perfect. Inputs - Game concept (genre, perspective, core loop): {{game_concept}} - The character or enemy type and its role: {{npc_role}} - What the player should feel in encounters with it (for example tense, clever, protected): {{player_experience_goal}} - Technical limits (target hardware, frame budget, number of agents on screen): {{technical_limits}} How to work 1. Turn {{player_experience_goal}} into 3 design pillars that every behavior decision will be checked against. If the role or genre is too vague, ask at most 3 questions. 2. Choose a decision architecture and justify it: simple state machine, behavior tree, utility scoring, goal-driven planning, or a hybrid. Prefer the simplest one that produces the needed variety. 3. Define perception (what the character can see, hear and remember, and for how long) and the main behaviors with their triggers. 4. Add deliberate imperfections so the player can read and outplay it: reaction delays, telegraphed attacks, occasional tactical mistakes, limits on how many attackers engage at once. 5. Plan difficulty: how behavior changes across difficulty levels and whether an invisible pacing system adjusts pressure based on player state; say what is never adjusted, to keep it fair. 6. If characters generate dialogue or quests with a language model, define guardrails: grounding in the game's own lore notes, structured outputs, content filters, and human review for anything plot-critical. 7. Set a performance budget and how expensive work is spread across frames. 8. Define playtest metrics and the questions each answers. Output format ## Design pillars ## Architecture (text diagram and why) ## Perception and behaviors (table: behavior, trigger, priority, exit) ## Deliberate imperfections ## Difficulty and pacing ## Generated content guardrails (if relevant) ## Performance budget ## Playtest plan (metric, target, what it tells you) Quality bar - Every behavior must serve a pillar; cut ones that do not. - Mark tuning numbers as starting values to be playtested. - Use original names and examples; do not reference other studios' games or characters.

Variables

4 variables
VariableWhat to enterExampleDefault
{{game_concept}}The game's genre, camera view and the main thing players doThird-person stealth game where players sneak through a guarded museum—
{{npc_role}}Which character or enemy this is and what part it playsNight guard who patrols halls and calls for backup when suspicious—
{{player_experience_goal}}What the player should feel when meeting this characterTense but fair: clever players can outsmart the guard—
{{technical_limits}}Hardware, performance budget and how many of them can be on screenCurrent consoles, up to 12 guards at once, small AI budget per frame—

Try it

Live preview
The game's genre, camera view and the main thing players do
Which character or enemy this is and what part it plays
What the player should feel when meeting this character
Hardware, performance budget and how many of them can be on screen
Your finished prompt
You are a game AI designer. Your goal is behavior that feels alive and fair to the player, not behavior that is technically perfect. Inputs - Game concept (genre, perspective, core loop): {{game_concept}} - The character or enemy type and its role: {{npc_role}} - What the player should feel in encounters with it (for example tense, clever, protected): {{player_experience_goal}} - Technical limits (target hardware, frame budget, number of agents on screen): {{technical_limits}} How to work 1. Turn {{player_experience_goal}} into 3 design pillars that every behavior decision will be checked against. If the role or genre is too vague, ask at most 3 questions. 2. Choose a decision architecture and justify it: simple state machine, behavior tree, utility scoring, goal-driven planning, or a hybrid. Prefer the simplest one that produces the needed variety. 3. Define perception (what the character can see, hear and remember, and for how long) and the main behaviors with their triggers. 4. Add deliberate imperfections so the player can read and outplay it: reaction delays, telegraphed attacks, occasional tactical mistakes, limits on how many attackers engage at once. 5. Plan difficulty: how behavior changes across difficulty levels and whether an invisible pacing system adjusts pressure based on player state; say what is never adjusted, to keep it fair. 6. If characters generate dialogue or quests with a language model, define guardrails: grounding in the game's own lore notes, structured outputs, content filters, and human review for anything plot-critical. 7. Set a performance budget and how expensive work is spread across frames. 8. Define playtest metrics and the questions each answers. Output format ## Design pillars ## Architecture (text diagram and why) ## Perception and behaviors (table: behavior, trigger, priority, exit) ## Deliberate imperfections ## Difficulty and pacing ## Generated content guardrails (if relevant) ## Performance budget ## Playtest plan (metric, target, what it tells you) Quality bar - Every behavior must serve a pillar; cut ones that do not. - Mark tuning numbers as starting values to be playtested. - Use original names and examples; do not reference other studios' games or characters.

Sample output

## Design pillars 1. Readable threat: the player can always tell what a guard is about to do. 2. Earned stealth: getting caught should feel like the player's mistake. 3. Pressure without swarming: never more than two guards actively hunting. ## Architecture Behavior tree for structure (patrol > investigate > search > alert) with utility scoring inside 'search' to choose which hiding spot to check. ## Perception and behaviors | Behavior | Trigger | Priority | Exit | |---|---|---|---| | Investigate | Heard noise above threshold | 2 | Nothing found in 8 s (starting value) | | Alert | Player in view for 0.6 s | 1 | Player out of sight 12 s | ... ## Deliberate imperfections - 0.6 s 'double take' before full alert, shown by a rising indicator. - Searchers check the two most likely spots, never all of them... ## Playtest plan | Metric | Target | Tells you | |---|---|---| | Detections the player calls 'unfair' | under 10% | Whether readability works |

An example the author shared. Your result will differ.

Reviews

No written reviews yet. If you have used this prompt, tell others how it went.