How to Reduce Hallucinations

When accuracy matters, treat the model like a helpful intern: require sources, constrain inputs, and verify output.

Related: prompt recipe · prompt lists · daily prompts · random prompt

1) Ground the model in your data

Use only the information in the text below.
If something is missing, say “unknown”.
Text: [paste]

2) Ask for assumptions explicitly

List the assumptions you’re making.
Then answer using those assumptions as bullet points.

3) Ask for uncertainty labels

• “Mark each claim as high/medium/low confidence.”
• “What would you need to confirm this?”

4) Force a verification step

After answering, add a “Verification checklist” of things I should confirm.

Next: evaluate AI answers.

Explore more

More: all guides · suggest a prompt