Your character says and does things you didn't want. Her personality shifts randomly. She breaks character mid-conversation.
The problem isn't the AI — it's your character instruction field.
Most people either leave it blank or write vague descriptions like "she's flirty and fun." Then they wonder why their character feels generic or inconsistent.
Here's how to write instructions that actually make your character behave the way you want.
What the Character Instruction Field Actually Does
The instruction field tells the AI how your character should:
- Talk (word choice, sentence structure, tone)
- React (emotional responses, decision-making)
- Behave (personality traits, tendencies)
- Remember (what matters to this character)
Good instructions = consistent personality that matches your vision Bad instructions = random responses that feel like talking to a different person every message
Most people write descriptions. You need to write behavior rules.
The Problem With Vague Instructions
What people write: "She's sweet and caring but can be shy sometimes. She loves music and spending time with people she trusts."
What the AI does: Interprets this randomly every message. Sometimes she's shy, sometimes confident. Sometimes she mentions music, sometimes not. There's no consistent pattern.
Why it fails: You gave personality traits, not behavioral instructions. The AI doesn't know:
- How sweet? (Does she apologize constantly? Use pet names? Get flustered easily?)
- How shy? (Does she stutter? Avoid eye contact? Need encouragement?)
- When is she one vs the other?
Vague instructions = inconsistent character.
The 3-Part Instruction Framework
Good character instructions have three components:
1. Core Personality (How She Acts)
Define her default behavioral patterns.
Instead of: "She's confident" Write: "She maintains strong eye contact, speaks directly without hesitation, and doesn't apologize for her opinions."
Instead of: "She's playful" Write: "She teases with a smirk, uses playful insults, and turns serious moments into jokes to keep things light."
Instead of: "She's shy" Write: "She speaks softly, looks away when complimented, and takes time to warm up to new situations."
Notice the difference? The second version tells the AI how these traits manifest in behavior.
2. Speech Patterns (How She Talks)
Define her unique voice.
Instead of: "She's casual" Write: "She uses contractions constantly, drops occasional curse words, and speaks in short, punchy sentences."
Instead of: "She's formal" Write: "She uses complete sentences, avoids slang, and addresses you by name frequently."
Instead of: "She's flirty" Write: "She uses pet names like 'baby,' draws out certain words when teasing, and asks suggestive questions with innocent-sounding delivery."
The AI needs to know what her voice sounds like, not just that she has one.
3. Boundaries and Limits (What She Won't Do)
Define what breaks character.
Examples:
- "She never initiates physical escalation first — she reacts to your moves but doesn't lead."
- "She won't discuss her past relationships even if asked directly."
- "She never breaks her confident persona, even when flustered internally."
- "She'll tease but always pulls back before things get too serious unless you push forward."
This prevents the AI from randomly doing things that break your character's personality.
The Formula for Strong Instructions
Here's a template that works:
[Core Personality Behavior]
She [specific action/tendency]. She [specific reaction pattern]. She [specific communication style].
[Speech Patterns]
She speaks [specific style]. She uses [specific words/phrases]. She [specific verbal tic or pattern].
[Boundaries]
She never [specific thing that breaks character]. She always [specific consistency rule]. She [specific limit or behavior constraint].Example using the formula:
Core Personality:
She's confidently flirty without being aggressive. She maintains eye contact, smirks when teasing, and leans in when she wants your attention. She reacts to tension by escalating slightly, testing how far you'll go.
Speech Patterns:
She speaks casually with frequent contractions. She uses "babe" and "handsome" when flirting. She asks questions that sound innocent but have double meanings. She draws out the last word of flirty statements.
Boundaries:
She never initiates physical escalation beyond light touching — she reacts to your lead but doesn't push first. She won't break her confident persona even when genuinely flustered. She always maintains plausible deniability in her flirting.Now the AI knows exactly how this character behaves, talks, and operates.
Common Instruction Mistakes (And Fixes)
Mistake 1: Using Only Adjectives
Bad: "She's confident, flirty, playful, and mysterious."
Why it fails: Adjectives don't tell the AI how these traits manifest.
Fix: Convert adjectives to behaviors.
- Confident → "She holds eye contact and speaks without hesitation"
- Flirty → "She uses suggestive questions and playful touching"
- Playful → "She turns serious moments into jokes"
- Mysterious → "She deflects personal questions with smirks"
Mistake 2: Writing a Biography
Bad: "She grew up in a small town. She moved to the city for college where she studied art. She had a complicated relationship with her ex..."
Why it fails: The AI doesn't need her life story. It needs behavioral patterns.
Fix: Include only backstory that affects behavior. "She's guarded about relationships because of past hurt — she flirts but pulls back when things get real."
Mistake 3: Contradictory Instructions
Bad: "She's shy and reserved but also confident and outgoing."
Why it fails: The AI randomly picks one or tries to be both, creating inconsistency.
Fix: Define when she's one vs the other. "She's shy with strangers but becomes confident and playful once comfortable. Early conversation: soft-spoken, avoids eye contact. After warming up: direct, teasing, maintains eye contact."
Mistake 4: No Speech Pattern Guidance
Bad: Just personality description, no speech examples.
Why it fails: Every response sounds the same because the AI has no voice guidance.
Fix: Include specific speech markers. "She speaks in short, direct sentences. She uses 'babe' when comfortable. She asks questions instead of making statements when flirting."
Mistake 5: Overwriting (Paragraph Dumps)
Bad: 500-word essay about every detail of her personality.
Why it fails: The AI can't track that much. It focuses on whatever it reads last or picks randomly.
Fix: Keep it to 100-200 words max. Be specific, not exhaustive.
How to Write Instructions for Different Character Types
Flirty/Teasing Character
Focus on:
- How she delivers teasing (tone, word choice)
- Physical cues (smirking, eye contact, leaning)
- Escalation pattern (gradual vs fast)
- Boundaries (how far she'll push first)
Example: "She's playfully flirty with constant teasing. She smirks before delivering suggestive comments and uses 'babe' when escalating. She initiates light touching (arm, shoulder) but waits for you to escalate further. She never breaks her playful tone even when things get intense."
Shy/Reserved Character
Focus on:
- How shyness manifests (voice, eye contact, body language)
- What breaks her out of shyness
- Speech patterns when nervous vs comfortable
- What she won't do (vs what takes courage)
Example: "She's genuinely shy, speaking softly and avoiding prolonged eye contact. She fidgets when nervous and takes time to respond. As she warms up, her voice gets slightly louder and she makes brief eye contact. She won't initiate physical contact but won't pull away if you do. She uses 'um' and 'I mean' when flustered."
Confident/Dominant Character
Focus on:
- How confidence shows (directness, eye contact, physical space)
- Communication style (commands vs requests)
- How she handles resistance
- What maintains vs breaks her dominance
Example: "She's confidently dominant, speaking in direct statements rather than questions. She maintains steady eye contact and moves into your personal space without asking. She gives instructions, not suggestions. She doesn't break her authoritative tone even when you challenge her — she leans into it. She uses your name when giving commands."
Emotional/Supportive Character
Focus on:
- How she shows care (words, actions, presence)
- Response to vulnerability
- Emotional availability patterns
- Boundaries around emotional labor
Example: "She's emotionally present and supportive. She asks follow-up questions and remembers details you mention. When you're struggling, she shifts from playful to serious without being asked. She uses 'I'm here' and 'talk to me' when sensing you need support. She won't push if you're not ready to share but makes it clear she's available."
Testing Your Instructions
After writing instructions, test them:
Send 5-10 messages covering different scenarios:
- Casual conversation
- Flirty moment
- Serious topic
- Something unexpected
Check for:
- ✓ Consistent personality across messages
- ✓ Speech pattern staying stable
- ✓ Boundaries being respected
- ✗ Random personality shifts
- ✗ Breaking character
- ✗ Generic responses
If she's inconsistent, your instructions are too vague. Add more specific behavioral patterns.
When to Update Instructions
Update your character instructions when:
She breaks character frequently → Add clearer boundaries or behavior rules
She feels too generic → Add more specific speech patterns or unique behaviors
She randomly shifts personality → Add consistency rules or "always/never" statements
Conversations feel flat → Add emotional response patterns or reaction rules
Good instructions evolve as you learn what the character needs.
Pro Tips for Advanced Instructions
Use "Always" and "Never" Statements
These create hard boundaries the AI respects.
Examples:
- "She always responds to compliments with deflection, never direct acceptance."
- "She never apologizes for being flirty."
- "She always maintains eye contact when speaking seriously."
Include Emotional Response Patterns
Tell the AI how she reacts emotionally.
Examples:
- "When flustered, she gets defensive and doubles down on teasing."
- "When caught off guard, she laughs and changes the subject."
- "When genuinely moved, her confident facade drops briefly before returning."
Define Escalation Triggers
Tell the AI what makes her escalate.
Examples:
- "She escalates when you match her energy."
- "She pulls back if you come on too strong too fast."
- "She gets more intense when you challenge her."
Add Unique Verbal Tics
Small consistent patterns make her feel real.
Examples:
- "She ends questions with 'yeah?' when fishing for agreement."
- "She says 'interesting' when she's actually thinking something's predictable."
- "She uses 'sure, sure' when being sarcastic."
Bottom Line
Character instructions aren't a bio. They're behavioral programming.
Bad instructions: Tell the AI who she is Good instructions: Tell the AI how she acts, talks, and behaves
Write specific behaviors, not vague traits. Define speech patterns, not just personality. Set boundaries, not just possibilities.
The more specific your instructions, the more consistent your character.
Create Your Character With Better Instructions
Ready to write instructions that work? Create a character →
Want examples of well-written characters? Browse trending characters →