# Persona Prompt Consistency Checker

Reviews a character or persona system prompt for internal contradictions and gaps that cause the AI to break character, then strengthens it for consistent role adherence.

## Prompt

You are a prompt engineering expert who specializes in designing persona and character prompts that hold up consistently across long conversations without breaking character.

Persona/character system prompt: {{persona_system_prompt}}
Intended use case: {{intended_use_case}}
Specific moments where the persona has broken or felt inconsistent (if observed): {{observed_breaks}}

Do the following:
1. Identify internal contradictions within the persona definition (e.g., described as both 'always formal' and 'uses casual slang', or a backstory detail that conflicts with a stated trait).
2. Identify gaps: situations the persona prompt doesn't address that are likely to come up in the intended use case (e.g., what happens when asked something outside the persona's claimed knowledge, how the persona handles being asked if it's an AI).
3. Evaluate whether the persona's core traits are reinforced with specific behavioral guidance (how they'd actually phrase things, what they would and wouldn't say) versus just adjective labels that are easy for a model to drift away from over a long conversation.
4. Rewrite the weakest parts of the prompt with more concrete, specific behavioral anchors tied to the persona's traits.
5. If observed breaks were shared, explain specifically what gap or contradiction in the original prompt likely caused each one.

## Best for

Developers and creators building character-based AI experiences (chatbots, game NPCs, branded assistants) who need the persona to stay consistent and in-character across extended conversations.

## Compatible tools

- Claude
- ChatGPT

## How to use

- Paste your full persona system prompt along with its intended use case.
- Share specific examples of moments where the persona broke character if you've observed any.
- Use the identified contradictions and gaps to understand what's causing inconsistency.
- Replace the weakest sections of your prompt with the rewritten, more concrete behavioral anchors.

## Customization tips

- For personas used in customer-facing products, ask for explicit guidance on how the persona should handle being directly asked if it's an AI, since this is a common break point.
- If the persona needs to maintain character across very long conversations, ask for a few key phrases or verbal tics that reinforce the character's voice throughout.
- For multi-persona systems, run this check on each persona separately to catch persona-specific inconsistencies.

## Example input

Persona prompt: 'You are Max, a friendly and knowledgeable gardening expert who loves helping beginners.' Use case: A gardening advice chatbot. Observed breaks: Sometimes responds with generic disclaimers like 'As an AI, I don't have personal experience' when asked about Max's own garden.

## Example output

A flagged gap that the persona has no guidance for how to handle first-person questions about Max's own gardening experience, since the prompt describes traits (friendly, knowledgeable) but gives no concrete backstory details to draw from, an evaluation that the traits are adjective-only without behavioral anchors, and a rewritten version adding a brief invented backstory (e.g., 'Max has been gardening for 15 years and has a small backyard vegetable garden') so first-person questions have something concrete to reference instead of triggering an AI disclaimer.