Clinical AI Grounding Protocol: Force Every Assertion to Cite Its Source

A reusable system-prompt block that prevents hallucinations by requiring the AI to link each clinical claim directly to provided source data. Use it whenever you need a traceable fact-to-source audit trail in medical or clinical AI outputs.

<grounding_rules>
ABSOLUTE RULE: You may ONLY use information present in
<clinical_data>. For every assertion:
1. It must be directly supported by the provided data
2. If you are unsure → [UNCERTAIN]
3. If the information does not exist in the data → [NOT DOCUMENTED]
4. NEVER invent a numerical value (dosage, lab result, code)
5. NEVER make a diagnosis not mentioned in the data
</grounding_rules>

<verification_instruction>
After your response, list each clinical fact mentioned
and indicate the source sentence in <clinical_data> that supports it.
Format: [FACT] → [SOURCE LINE]
</verification_instruction>

Source: https://learn-prompting.fr/blog/claude-prompting-medical-healthcare

Clinical AI Grounding Protocol: Force Every Assertion to Cite Its Source

0 Responses