Prompt library Civil servants, Introductory, Across government, Staying safe
Ask the AI to examine its own answer
Get a list of every factual claim in an AI answer, how sure the AI is and what source would confirm it, so you know what to check.
Last reviewed: . Reviewer: Muhammad Asadullah.
Before you paste anything: replace names, CNIC numbers and other personal details with placeholders and never use classified material. Use only AI tools your department allows.
When to use it
An AI tool has given you an answer that you may use in your work. Before you rely on it, you want to see which parts are facts that need checking. This prompt helps you find what to check. It is not itself a check, because the AI can be wrong about its own answer too.
The prompt
Examine the answer you gave earlier in this conversation. Do not defend it or rewrite it.
[If the answer is from another chat or tool, paste it here]
List every factual claim in the answer: names, dates, figures, rules, procedures, quotations and references to documents. For each claim give:
1. The claim, in a few words
2. How confident you are (high, medium or low) and why
3. What kind of source would confirm it, such as an official website, a notification or a published report
4. Whether you may have invented or guessed it
Be direct about uncertainty. If a claim is a guess or may be made up, say so and mark it [CHECK]. Do not invent sources, links, document titles or reference numbers. If you do not know a source, write "source not known" instead.
End with one line reminding me that this list shows what to check and is not a check in itself.
How to fill it in
- Use it in the same conversation as the original answer. If the answer came from elsewhere, paste it in.
- Ask about one answer at a time. A long answer may need to be split into parts.
- If the answer will go into an official document, also note where it will be used, so you can decide how carefully to check.
Before you use the result
- Check every claim yourself against a reliable source, starting with those marked low confidence or [CHECK]. Claims marked high can still be wrong.
- Never accept a source the AI names without opening it. AI tools sometimes invent document titles and links.
- Remove or correct any claim you cannot confirm. See hallucination and automation bias.
Related guide: Using AI safely.