Fact-check a claim: separate evidence, interpretation, and unknowns

When to use this prompt

Use this before publishing an article, report, or research note. It helps organize evidence; you still need to open and read the original sources.

Copyable prompt

Review the material below.

1. Extract specific, verifiable claims and preserve their context. Separate observed facts, interpretations, and predictions.
2. If you can browse, prioritize original studies, official documents, or first-hand records. Provide the source title, full URL, publication date, and relevant evidence.
3. If you cannot browse, evaluate only the sources I supply. Do not invent links, papers, quotations, or research findings. A link you have not opened is not verified evidence.
4. Label each claim: supported by the supplied evidence, insufficient evidence, conflicts with the source, or currently unverifiable. Explain why. Check dates, denominators, units, and the population studied.
5. Return: original claim; assessment; evidence; unresolved question. Finish with three specific checks a person should perform before publication.
   Material: [paste the text]
   Sources: [paste source text or complete URLs]
   Relevant time period: [enter dates]

Example input

The following material and sample answer are fictional teaching examples, not an actual tool trial or a completed model evaluation.

A draft says: "This tool saves everyone 30% of their writing time." The attached record describes 16 volunteers completing similar tasks. Average time was 8 minutes before and 6 minutes after. There was no randomized assignment or control group. Individual timings and writing-quality scores were not supplied. Check the claim using only these details.

Example output

1. "Saves 30%": conflicts with the supplied numbers. The relative reduction is (8 − 6) ÷ 8 = 25%, or 2 minutes from the original 8-minute baseline.

2. "Everyone": insufficient evidence. A volunteer sample of 16 cannot establish a result for all users. An improved average does not show that every individual improved.

3. "The tool caused the reduction": currently unverifiable. Practice, differences between tasks, or participant selection may explain part of the change.

4. "The writing is equally good": currently unverifiable. No quality measure was supplied.

A defensible rewrite is: "In this uncontrolled record of 16 volunteers, average time fell from 8 to 6 minutes, a 25% reduction. The record does not establish what caused the change or whether it applies to other users."

Human review

Check the complete individual timing records, whether the tasks and timing rules were comparable, and whether quality, outliers, and a suitable comparison group were assessed. Do not add an imaginary study to make the answer sound authoritative.

For a framework for evaluating reasons and evidence, read Critical Thinking.