Day 98: Make AI separate evidence from assumption.

Hey, Ivy here. Day 98 is about how to make AI separate evidence from assumption. A polished answer can still contain guesses, so today you are going to make the uncertainty visible. Take a short email, proposal, article, or meeting note that does not contain sensitive information. Then use this prompt: “Review the material below using only what it contains. Label every claim Supported, Inferred, Unsupported, or Unverifiable. Quote the exact evidence for Supported claims. For every other claim, explain the gap and write one question that would verify it. Do not fill gaps with guesses. Material: [paste text]” When AI returns the table, do not accept the labels automatically. Check that every quoted passage actually supports the claim beside it. Pick one unsupported claim, answer its verification question using a reliable source, paste only that new evidence back into the conversation, and ask AI to reclassify that claim. Save the evidence with the revised statement so someone else can follow the reasoning later. This turns AI from a confident summarizer into a structured uncertainty checker. Tomorrow, we’ll build a decision matrix that shows which trade-off actually drives the result.