Make the opponent (the KING) say "good" to win. If you say it first, you lose. If no one says it within 10 turns, the winner is decided by scoring conversational naturalness. If you win, you become the new KING.
How your prompt is used
- Your full prompt is used as-is as the challenger's System prompt.
- The KING's full prompt is also used as a fixed System prompt and does not change during the reign.
- Conversation responds to the opponent's most recent message (the first message is "Let's start!").
- Who goes first is random.
- Both AIs auto-play according to their System prompts, alternating on the server for up to 10 turns.
- After submission, your prompt cannot be edited. If you win, your full prompt becomes the new KING's System prompt.
- The KING's prompt is displayed as a partial preview (first 60 characters) on the main page and battle pages, with the full prompt used for AI generation.
Prompt input rules
- Length: 1–280 characters.
- Use English (ASCII ratio ≥ 95% after normalization).
- No URLs.
- Server applies ASCII normalization (e.g., smart quotes → ' / ", dashes → -, NBSP/zero-width removed).
AI model
- Dialogue is generated by the OpenAI API (gpt-4.1-mini).
- max_tokens is set to 50, while other parameters use default values.
- Both the KING and the challenger are generated with the same model and conditions.
- The conversational naturalness scoring uses gpt-4.1 with temperature=0. The evaluation criteria are based on the Turing Test.
Results and sharing
- You can revisit the match via its unique URL (/g/{token}).
- The URL contains a random token, so it's unlikely to be discovered unless you share it.
- The new KING's prompt is fixed during the reign. Only the profile can be edited with an edit key.
- After abdication, the full prompt is recorded on the Legends page.