Library › Research & AI Search › Finance and accounting

Mark each claim in a deep-research draft Keep, Fix or Cut

  • For Finance and accounting
  • You practice Catching mistakes and made-up facts
  • Time about 5 minutes

About this exercise

A deep-research agent drafted a background note on savings account rates for your finance committee, and you have the source excerpts it used. Write exactly 5 numbered lines, one per claim, each labeled Keep, Fix or Cut: Keep if the source says it, Fix if the source gives a different number (give the right number), Cut if no source supports it. Plain text, under 100 words.

Common mistakes when you ask AI for this

  • Asking for a review. The AI rewrites the draft in nicer words without checking it against the sources. Try: Say: "Check each claim only against its source excerpt."
  • No fixed labels. The AI says things like "mostly accurate", which doesn't tell you what to do. Try: Add: "Label each Keep, Fix or Cut, and define each."
  • Missing the overreach. A bank's own page proves its rate, not that it's the highest anywhere. The AI often lets that slide. Try: Add: "Cut any part the source doesn't actually say."
  • Missing a flipped claim. The AI skims the column and misses that it says the opposite of the draft. Try: Add: "For Fix, say what the source actually says."
  • Long or formatted output. AI adds bold labels and a rewritten paragraph at the end. Try: Add: "Only the 5 lines, each under 18 words, plain text."

What a strong prompt includes

  • Checks against the sources
  • Gives exact labels
  • Defines each label
  • Asks for the right number
  • Sets the shape

Practice this now ↑

Related exercises