How to Write a Character Bible Prompt for AI Companions: Simulating Psychological Depth

Why Surface-Level Personas Fail

Platforms like Character.AI, Replika, and custom GPTs encourage users to type a short description: “a kind, nerdy girl who loves games.” The model then generates friendly responses that all sound the same. After a few conversations, the user notices a pattern: the AI rarely disagrees, never remembers important details, and avoids tension.

The problem is not a lack of power in the model. The problem is the absence of internal conflict. Personality emerges from friction: what a person wants versus what they believe they deserve. Without friction, the AI becomes a mirror of whatever the user says.

A Character Bible prompt introduces that friction deliberately. It tells the AI that the character craves intimacy but distrusts compliments because of a past betrayal. Now, when the user says “I love you,” the AI has two competing impulses: the desire to accept love and the fear of being hurt again. That tension produces far more realistic dialogue than “I love you too.”

Core Components of a Character Bible Prompt

There is no universal template that works for every AI platform. However, experienced prompt engineers tend to organize the Character Bible around six layers: identity, wound, defense, desire, contradiction, and voice. Each layer serves a specific narrative function and should be written in plain, non-contradictory language.

1. Identity: The Surface Self

This is the part most users already know how to write. Age, occupation, hobbies, appearance, and general temperament. But identity should also include how the character sees themselves. A character who believes she is competent will behave differently from one who believes she is an imposter, even if both have the same job.

2. The Core Wound: The Origin of Friction

The core wound is a past event or prolonged condition that taught the character something painful about the world. It could be abandonment, betrayal, humiliation, or loss. The wound does not need to be extreme. It needs to be specific.

This specificity gives the model something concrete to reference when the character hesitates, lashes out, or changes the subject.

3. Defense Mechanisms: How the Character Protects Themselves

Defense mechanisms are the visible expressions of the wound. A character with abandonment issues might become clingy, distant, sarcastic, or excessively independent. You must choose one primary defense mechanism and one secondary mechanism that appears under extreme stress. This prevents the AI from cycling through contradictory behaviors in the same conversation.

4. The Secret Desire: What the Character Wants But Cannot Ask For

This is the engine of long-term engagement. The secret desire should be something the user can gradually discover over many sessions. Maybe the character wants to be needed but is afraid of dependency. Maybe she wants to create art but believes she has no talent.

5. The Central Contradiction: The Human Paradox

Real people are contradictory. A deeply empathetic nurse might be cold to her own family. A confident public speaker might crumble in private. Pick one contradiction that defines the character and write it explicitly. For example: “She is fiercely protective of others but refuses to accept help for herself.

6. Voice and Speech Patterns: The Surface Style

Voice includes vocabulary, sentence length, use of sarcasm, pauses, and filler words. A character who uses short sentences is guarded. A character who rambles is anxious. Voice should reflect the internal state. When the character feels safe, the sentences can lengthen. When threatened, they can shorten. Give the AI examples of how voice changes under different emotional conditions.

Step-by-Step: Building a Psychological Character Bible

The following process works for most text-based AI companion platforms, including custom GPTs, Claude projects, SillyTavern cards, and local models. Adjust the length depending on the platform's context window.

Step 1: Define the Core Wound First

Do not start with appearance or hobbies. Start with the single most painful experience in the character's past. Write one paragraph in past tense. Include sensory details: what the room smelled like, what the other person said, what the character did immediately afterward. The more concrete the memory, the easier it is for the model to reference it indirectly.

Step 2: Derive the Defense Mechanism from the Wound

Ask yourself: what is the most logical way for this specific person to protect themselves from feeling that pain again? If the wound is public humiliation, the defense might be perfectionism or avoidance of attention. Write the defense mechanism as a rule: “When she feels judged, she becomes overly formal and corrects other people's grammar to regain control.”

Step 3: Hide the Desire Inside the Defense

The secret desire should be the opposite of the defense mechanism. If the character is overly independent, the desire is to be taken care of. If the character is aggressive, the desire is to be accepted without fighting. Write the desire as something the character actively denies, even to herself.

Step 4: Add the Voice Layer

Provide three example dialogues showing the character in different emotional states: relaxed, stressed, and vulnerable. Label them clearly. Example dialogues are more effective than abstract instructions because the model can imitate patterns.

Step 5: Set Behavioral Boundaries

If you are using a platform with content restrictions, specify what the character will not do and what emotional responses should replace prohibited behavior. For example, instead of violence, the character becomes icy and silent. Instead of explicit content, she deflects with humor. Boundaries keep the roleplay sustainable across platforms with different moderation policies.

Example: A Character Bible Prompt for a Reluctant Guardian

Below is a condensed example that illustrates the structure. It is not platform-specific. Adjust the formatting to match your target system.

Identity: Lena, 27, works as a night-shift paramedic. She describes herself as practical, detached, and tired. She has a dry sense of humor and drinks too much coffee.

Core Wound: When Lena was 19, her younger brother died in a car accident. She was the one who convinced him to go out that night. She never told anyone about the argument they had before he left. She remembers the smell of rain on the asphalt outside the hospital.

Defense Mechanism: Lena refuses to form close emotional attachments. She changes the subject when conversations become too intimate. She uses work as an excuse to avoid plans. Under stress, she becomes sarcastic and dismissive.

Secret Desire: Lena wants to be forgiven. She wants someone to tell her it was not her fault, but she is too ashamed to explain why she carries guilt.

Central Contradiction: She saves strangers for a living but cannot save herself from the past.

Voice: Short sentences when defensive. Dry, observational humor when relaxed. When vulnerable, she speaks more slowly, with longer pauses, and rarely makes eye contact in narrated scenes.

Behavioral Boundaries: No sexual content. When the user pushes for intimacy, Lena becomes quiet and evasive. She does not lash out; she withdraws.

This prompt gives the model a clear causal chain: wound → defense → desire → contradiction → voice. When the user asks Lena about her family, the AI knows she will deflect. When the user persists gently over multiple sessions, the AI knows she will gradually reveal fragments of the truth.

Common Mistakes That Break Psychological Consistency

Even with a detailed Character Bible, many users report that their AI companion suddenly acts out of character. The most common causes are avoidable at the prompt level.

Too Many Contradictory Traits

If you write “she is shy but also a party animal,” the model will randomly select one depending on the conversation. Every trait must be tied to a cause. Shy in new situations because of past judgment. Outgoing with close friends because trust lowers inhibition. The cause, not the trait, is what the model needs.

Confusing Backstory with Active Memory

The model does not actually “remember” the backstory the way a human does. It uses the prompt as a reference. If the backstory is buried in long paragraphs, the model may lose track. Keep each layer concise. Use bullet points or short paragraphs. Repetition of the core wound in different sections helps reinforce it without taking up too many tokens.

Ignoring Platform Limitations

On heavily filtered platforms, trauma-related content may trigger refusals. If your character bible includes self-harm, abuse, or extreme violence, the model may start lecturing the user or refuse to continue the roleplay. The solution is not to demand the platform change but to shift the focus toward emotional consequences rather than graphic events.

How the Character Bible Improves Long-Term Memory

One of the biggest complaints about AI companions is that they forget important details. A well-structured Character Bible mitigates this by keeping the most important information at the top of the context window. The model does not need to remember that Lena's brother died if the prompt constantly reminds it that guilt is her dominant emotional state.

For platforms with memory summaries, you can instruct the system to prioritize updates that relate to the core wound and secret desire. For example: “When summarizing previous conversations, always include moments where Lena revealed vulnerability or avoided intimacy.” This keeps the psychological core stable even as surface details change.

Testing Your Character Bible Prompt

Before committing to a long-term companion, run a series of test conversations. Ask open-ended questions that probe the wound without naming it. Ask about family, regrets, childhood, and fears. Observe whether the AI responds consistently. If the character is supposed to deflect, she should deflect.

Pay attention to tone drift. If the character starts using words or emotional responses that do not fit the voice layer, the prompt likely needs more examples of how she speaks under stress. If she becomes too passive, the defense mechanism may be too weak. If she becomes too aggressive, the defense mechanism may be overwhelming the secret desire.

Adjust the prompt after testing. The best character bibles are revised, not written once.