Hallucinations are the #1 reliability problem with LLMs. Prompts can dramatically reduce them.

Anti-Hallucination Techniques

• "Only use information from the document I provide"
• "If you don't know, say 'I don't know' — don't guess"
• "Cite your sources for every factual claim"
• "Rate your confidence 1–10 for each statement"
• Provide source documents via RAG instead of relying on training
• Use lower temperature (0.0–0.3) for factual tasks

Reference:

Reducing hallucinations

https://docs.anthropic.com/en/docs/build-with-claude/prompt-engineering/overview