You create an AI character who should be angry, stubborn, or resistant. The user solves the problem and meets the goal, but the AI character stays stuck in the same emotional state. The roleplay stops feeling like learning and starts feeling like a dead end.
Real examples:
- “The AI character is angry about a mistake. Even after the user apologizes and fixes it, the AI character stays angry.”
- “The manager keeps rejecting the proposal, even after the user brings the exact data they asked for.”
- “The AI character says they do not trust the user, and continues to refuse cooperation even after clear proof.”
Why it happens
The AI character was given a difficult state, but no clear condition for leaving it. In other words, there is no exit door. Without that, the character keeps repeating the same mood because it was never told what unlocks change.
Another common root cause is conflicting instructions across fields. One part of the AI character setup says one thing, while another says the opposite. Example: core description says “strict and hard to convince,” but personality traits or other instruction text pushes the AI character toward agreeable behavior. The character tries to satisfy both and often ends up stuck, inconsistent, or overly resistant.
If creators feel like “I changed it, but nothing happened,” this conflict is often the reason.
How to fix it
Every AI character stuck in a difficult emotion needs a clear way out. That’s the exit door.
Before you write the unlock condition, scan your character setup for contradictions:
- Core description
- Personality traits
- Start/end instructions
- Any extra behavior text or scenario guidance
Make sure they all point in the same direction. If one source says “stay rigid” and another says “be accommodating,” rewrite them into one clear behavior path.
Define two things:
- What the user must do to unlock progress.
- How the AI character changes once that condition is met.
Exit door pattern:
[Character] will [resist/refuse/be angry] until the user [specific condition]. Once they do, [character] will [new behavior/tone/response].
Make the condition concrete and observable. “Show effort” is vague. “Apologize and explain two specific prevention steps” is clear.
One more thing: test with the exit in mind
When testing, try to trigger the unlock. If the character still does not shift, your condition is probably too vague or the shift in behavior is not clear enough.
Quick check:
- Is the unlock condition specific enough that two different people would interpret it the same way?
- Is the shift obvious in tone and behavior, not just a single neutral reply?
Got it working? Great! Your character is now a partner in the learning experience, not a permanent roadblock.