# Check the answer. Keep your judgement.

Original 5HO practice material. Fictional example. Reviewed 5 October 2026.

## Useful result

You'll finish with a checked response with claims linked to evidence or marked uncertain.

A neat answer can still be wrong. Today, you will catch three kinds of trouble: changed facts, invented details and numbers that do not add up. Nice formatting gets no extra marks.

## Set up

Plan for 15 minutes. Keep the supplied evidence visible beside the AI answer. Use a calculator for arithmetic rather than asking the same model to reassure itself.

## Practice input

Source: 12 attendees; catering quote AUD 18 per person; delivery AUD 15; dietary responses incomplete. Draft: Lunch for 14 is confirmed at AUD 231, including delivery. All dietary needs are covered.

## Do the task

1. List each factual claim in the draft.
2. Match each claim to the source and check the arithmetic yourself.
3. Ask for a correction that preserves what is known and openly marks what is not.

## Prompt

Audit the draft against the source. Return Claim, Supported or unsupported, Reason, Corrected wording. Do not assume a quote is a confirmed booking.

## Refine

Separate the arithmetic check from the booking and dietary checks. Show the multiplication and addition.

## Expected checks

12 × 18 + 15 = AUD 231. The total happens to match the 12-person quote, but 14 attendees, a confirmed order and completed dietary information are unsupported.

## Save

Keep the corrected wording and your checking table. Use this check before sharing your next AI-written summary.

## Sources

- Writing clear prompts: https://help.openai.com/en/articles/10032626-prompt-engineering-best-practices-for-chatgpt