Question

Your team builds an internal HR assistant on Claude that answers employee questions about company benefits and leave policies. During testing, you notice the model sometimes confidently invents specific policy details (like exact reimbursement caps) that don't exist in your documentation. Which change would MOST effectively reduce these hallucinations?

Accepted Answer

Provide the relevant policy documents in the prompt and instruct Claude to answer only from those sources, cite them, and say 'I don't know' when the answer isn't present

Answer

Raise the temperature setting so the model explores more varied phrasings of its answers

Answer

Add a system prompt telling Claude it is an expert HR professional with 20 years of experience

Answer

Increase max_tokens so Claude has more room to fully explain each policy in detail