List the claims and possible hallucinations to verify in an AI draft
Extracts every factual claim, number, quote, and citation from an AI draft, flags likely hallucinations, and gives a verification checklist.
The prompt
I have a draft written with AI (report, email, article, proposal, or research summary). Pull out every claim I need to verify before I use it.
The draft: [PASTE]
Where it will be used: [e.g. client email, board pack, LinkedIn post, internal memo]
Sources I gave the AI, if any: [PASTE or LIST]
Steps:
1. Extract every checkable claim: facts, statistics, dates, names, titles, quotes, citations, product or policy details, legal or regulatory statements, and causal claims ("X led to Y").
2. For each, note whether it is supported by the sources I provided, common knowledge, or unsupported.
3. Flag statements that sound convincing but may be wrong, and mark possible hallucinations: very precise numbers with no source, confident generalizations, quotes attributed to real people, recent events, specific names, URLs, or studies, and citations that may not exist.
4. Rate the risk if each claim is wrong, given where the draft will be used.
Return:
1. A verification checklist table: # | Claim (short quote) | Type | Support (source / common knowledge / unsupported) | Risk (H/M/L) | How to verify (where to look).
2. The top 5 claims to check first.
3. Suggested softer wording for claims I can't verify in time, or a note to remove them.
4. A copy of the draft with each claim marked [VERIFY #n].
Do not confirm any claim from your own memory as if it were checked; you can say what is commonly reported, but I must check the source.
Highlighted [brackets] are blanks to replace with your own details.
Fill in the blanks and use this promptMore Working with AI prompts
What to fill in
- PASTE
- e.g. client email, board pack, LinkedIn post, internal memo
- PASTE or LIST
- VERIFY #n
How this prompt is built
This prompt follows the RTF structure (Role, Task, Format). RTF tells the model who to be, what to do and what shape the answer should take. It is the quickest structure that still gives noticeably better results than a bare request.
Related prompts
- Write a responsible-AI policyAcceptable use, data handling, review requirements, and accountability for AI in your organization.
- Write a law firm policy for using generative AI on client workA practical firm policy on generative AI that covers confidentiality, verification of citations, supervision, billing, and client disclosure.
- Write custom instructions for my AI assistantCustom instructions, a project brief, or a custom GPT/Claude Project setup so the AI works the way you do.
- Write an SOP for my team to use AI safelyA step-by-step operating procedure so a whole team can use AI on the same task consistently, safely, and with review built in.