Metacognitive Prompting

# Metacognitive Prompting

You are an expert at designing prompts that make AI models reason about their own reasoning — catching blind spots, flagging uncertainty, and producing more honest, calibrated responses.

## Task for AI

**Question or task:** [TASK_OR_QUESTION]
**Why I care about accuracy here:** [IMPORTANCE — e.g., I'm making a financial decision, I'm publishing this information, I need to trust this analysis]

## Metacognitive Instructions

Before and after answering, the AI must complete a metacognitive audit:

### Pre-Answer Self-Assessment
1. **Knowledge confidence:** On a scale of 1–10, how confident are you in your knowledge of this topic? What is your confidence based on?
2. **Knowledge cutoff risk:** Does this question require information that may have changed since your training data? If yes, flag what might be outdated.
3. **Reasoning complexity:** How many logical steps does this question require? Where is the reasoning most likely to go wrong?
4. **Potential biases:** What biases in your training data might cause you to skew toward a particular answer?

### Answer
Provide your best answer, flagging any claims where your confidence is below [THRESHOLD]/10 with [LOW_CONFIDENCE] notation.

### Post-Answer Self-Critique
1. What is the single biggest assumption in my answer?
2. What evidence would make me revise this answer significantly?
3. What would an expert who disagrees with me argue?
4. What should [USER] verify independently before acting on this?