Use Chain-of-Verification so AI fact-checks itself first
AI-written facts often hide confident mistakes. Follow Meta's CoVe method: get a draft, get verification questions, answer them in a fresh chat, then keep only what agrees.
Start step 1 ↓The problem
You ask ChatGPT or Claude for background, a bio or a timeline, and it reads as complete, but a date or a cause in it was made up. If you just say "please double-check," it usually agrees with its own answer.
AI tools used
Tap a tool to see what it is and which other cases use it
What to prepare
- A question whose facts you really need checked
How to do it
- 1
Get a draft as usual
Ask the way you normally would. Don't add "please verify" yet; this step is just for a complete draft.
- 2
Have it write only verification questions
Questions must be open-ended ("When did X happen?"), not yes/no ("Was X in 1845?").
Prompt - 3
Switch to a new chat and paste only the questions
Open a brand-new chat and paste only the question list. Don't paste the draft, so it can't copy its own mistakes (in the paper, this factored version usually beats checking in the same chat).
Prompt - 4
Back in the draft chat, keep only what agrees
Paste the new chat's answers back and ask it to rewrite and list what it removed. Check three key dates or numbers against an outside source before you hand it in.
Prompt