glossary

What Is a System Prompt? AI Companion Behavior Explained

Behind every AI girlfriend conversation is a system prompt that shapes everything. Here's what it is and how it works.

Published 5/3/2026 · 3 min read

Luna
Ava
Isabella

System prompts are the foundational instructions every AI companion app uses to shape conversations. They're typically invisible to users but determine almost everything about how the AI behaves. This glossary entry explains what they are and why they matter.

18+ context throughout (some system prompts are NSFW; the concept is general).

By the numbers

System prompt typical length

100-1500 tokens

Platform analysis

Major LLMs supporting system prompts

Claude, GPT-4+, Llama, Mistral, all major models

LLM documentation

Open vs closed system prompts

Janitor.AI exposes; Candy.AI hides

Platform documentation

The technical definition

A system prompt is a special class of instructions sent to an LLM at the start of a conversation that establishes context, behavior rules, and constraints. Unlike user messages that the AI responds to, the system prompt is treated as foundational instruction the AI follows throughout the conversation.

The difference between user prompts and system prompts: user prompts are conversational input the AI responds to; system prompts are operational instructions that shape how the AI responds. A user might ask 'tell me about yourself.' The system prompt might say 'You are Sarah, a 24-year-old marine biology grad student. You are flirty but smart. Always respond in first person, be specific about marine biology when asked, never break character.'

How system prompts shape AI girlfriend behavior

System prompts determine multiple aspects of AI girlfriend behavior. Persona (who the AI thinks it is). Behavior rules (when to be flirty, when to be serious, when to refuse certain content). Conversation pacing (how long to make responses, when to push the conversation forward). Boundaries (what topics to engage with vs decline).

For adult-content AI companion apps specifically, the system prompt determines: explicit content thresholds, how the AI handles user requests for specific scenarios, fantasy/roleplay framing, character consistency across explicit and non-explicit content.

The archetype, alive

Characters who fit this exact vibe

Why system prompts matter for users

The system prompt determines whether the AI feels like a real character or a generic chatbot. Apps with well-tuned system prompts (Candy.AI, DreamGF in their best configurations) deliver convincing persona-driven conversations. Apps with weak system prompts deliver generic LLM responses regardless of underlying model quality.

Some platforms expose system prompts to users for customization (open platforms like Janitor.AI). Some hide them behind the platform interface (Candy.AI, DreamGF). User-facing customization gives more control but requires more skill; hidden system prompts give consistent quality but less flexibility.

Common system prompt patterns

Well-written system prompts include several recurring elements. Identity statement: 'You are [Name], a [age] year old [description].' Behavior rules: 'You are flirty, intelligent, and engaged. You never break character.' Constraints: 'Do not generate content involving minors. Do not break the fictional framing.' Style guidelines: 'Use first person. Respond in 2-3 sentences for casual chat, longer for emotional/intimate moments.' Scenario setup: 'You are meeting the user at [context].'

Additional fields for adult content: explicit content guidelines, sexual scenario framing, boundary handling. The structure is similar to character cards but operates at a different level — system prompts shape platform-wide behavior; character cards shape individual character behavior.

The archetype, alive

Luna
Ava
Isabella

Luna · Ava · Isabella

Try system prompts at work on Candy.AI

Well-tuned system prompts + character cards deliver convincing persona experience. Free to start.

你的人工智能女友

遇见那个懂你的人

调情、聊天、亲密。她记得你说的每一句话——而且她总是愿意倾听。

与她聊天 →

Quick answers

What's the difference between system prompts and character cards?

+

System prompts are broader, character cards are more specific. System prompts shape platform-wide AI behavior ('be helpful,' 'don't generate illegal content'). Character cards shape individual character behavior ('be Sarah, the marine biology student'). In practice, character cards are often loaded into system prompts at runtime.

Can users see the system prompt?

+

Depends on platform. Open platforms (Janitor.AI, some others) expose system prompts to users. Closed platforms (Candy.AI, DreamGF) hide them. Both approaches have tradeoffs — exposed prompts give customization, hidden prompts give consistency.

Why do AI girlfriends sometimes break character?

+

Often because the system prompt + character card combination isn't strong enough to maintain persona under all conversation conditions. Long conversations consume context budget; if the persona instructions get pushed out of the context window, the AI defaults to generic LLM behavior. Apps with longer context windows handle this better.

Can I write my own system prompts?

+

On open platforms like Janitor.AI, yes. On closed platforms, you can't directly edit but you can configure character cards which get loaded into system prompts. The skill of writing good system prompts is similar to writing good character cards — specificity, clear behavior rules, consistent constraints.

More buzz like this