Bias-Proof Prompt Checklist: 12 Checks Before You Trust the Output
Run this 12-item checklist and catch the three most common ways Claude hallucinates or misleads before you act on the result.
Free checklist · copy-paste ready · built by the BrainVaultAI team
AI doesn't lie maliciously. It pattern-matches and pleases. Those two behaviors produce three specific failure modes that cost founders real decisions: sycophancy (it agrees with your framing), recency bias (it weights recent training data too heavily), and false confidence (it sounds certain when it isn't). Run this checklist after any output you're about to act on, especially decisions about competitors, market sizing, technical architecture, or hiring. The two prompt blocks below are the active fixes: one breaks sycophancy, one forces uncertainty disclosure. Use them when the stakes are high enough to matter.
Bias Audit Checklist
Before acting on Claude's output, check: SYCOPHANCY CHECKS
[ ] Did I state my preferred answer before asking? If yes, reask without it.
[ ] Does the output agree with everything I said? If yes, ask for the strongest counterargument.
[ ] Did I ask a leading question? ("Isn't X better than Y?") If yes, rephrase as neutral. RECENCY / KNOWLEDGE CHECKS
[ ] Does this involve events after mid-2025? If yes, assume Claude is guessing.
[ ] Is Claude citing specific numbers (market size, pricing, statistics)? If yes, verify externally.
[ ] Does the answer assume current competitive landscape is static? Flag and verify. FALSE CONFIDENCE CHECKS
[ ] Did Claude use phrases like "certainly", "definitely", "always"? Ask what it doesn't know.
[ ] Is the answer too clean and confident for a genuinely complex question? Ask for edge cases.
[ ] Did Claude hedge at all? Zero hedging on uncertain topics is a red flag. DECISION GATE
[ ] Would I act on this output if I found out Claude made up the key supporting fact?
[ ] Have I cross-checked at least one key claim from another source?
[ ] Am I using this as one input or as the final answer?
Sycophancy Break Prompt
I'm going to share an idea. Before I do: your job is NOT to agree with me. Your job is to find what's wrong, missing, or wishful thinking in my reasoning. If you catch yourself building on my idea instead of stress-testing it, stop and reframe. Here's the idea: [your idea] Give me:
1. The single strongest argument against this.
2. What assumption I'm making that's most likely wrong.
3. Who would disagree with this and why they'd be right.
Uncertainty Disclosure Prompt
Answer the following question. Before you give me the answer, tell me:
- What you're confident about (high signal)
- What you're inferring or estimating (medium signal)
- What you're guessing or don't actually know (low signal) Label each section explicitly. If everything is high confidence on a complex topic, that's a red flag, push harder on what you don't know. Question: [your question]
Want the editable version?
We'll send the editable version of this asset, plus a short series of our highest-leverage AI plays for operators. No spam. Unsubscribe in one click.
Free. No spam. Unsubscribe in one click. By submitting you agree to receive emails from BrainVaultAI.