AI fact-check checklist
For dates, figures, names, quotes and citations supplied by an AI assistant — in donor letters, grant proposals, board summaries and policy briefs. From Lesson 1: generation is not verification.
Before you rely on an AI-supplied fact
- Is it checkable? Specific dates, figures, names, quotes and citations are exactly what plausible-sounding patterns imitate best. Flag every one in the draft.
- Would the internet have written much about it? Niche funders, local programmes, internal policies and anything recent are thin territory — treat those answers as well-dressed guesses (Lesson 2).
- Was it looked up or generated? If no source is cited, assume the frozen pattern answered. If a source IS cited: open it — a citation is a pointer, not a proof.
- Verified at an authoritative source? Funder's own guidelines, the official website, your finance system, the primary document. Never "the model said so twice".
- Never ask the model "Are you sure?" Its yes is another generated continuation, not confirmation. Model self-reports are generation too.
- Stakes check: trivial — enjoy; useful — spot-check the load-bearing fact; going to your board, a funder or forty inboxes — verify every checkable claim, then sign.
Record (for anything that shipped)
- Claim: ____________________ · Verified at: ____________________ · Date: ________ · By: ________
- Claim: ____________________ · Verified at: ____________________ · Date: ________ · By: ________
- Claim: ____________________ · Verified at: ____________________ · Date: ________ · By: ________
education.yougenius.ai · The Prediction Machine · free to copy and adapt within your organisation.