Fact-Checking and Evidence

Scientific Claim Review

Evaluate a scientific claim against study design, population, measures, evidence strength, uncertainty, relevance, replication, and risk of overstatement.

Free editable Markdown · Science editors, fact-checkers, and health communications teams · Updated July 30, 2026

Download Markdown

Accessible HTML preview

Blank template

The downloaded file contains the same fields in editable Markdown.

Claim and consequence

Claim as drafted: [Copy exact sentence]
Claim type: [Description / Association / Cause / Prediction / Mechanism / Safety / Effect]
Population implied: [Enter]
Exposure or intervention: [Enter]
Comparison: [Enter]
Outcome and period: [Enter]
Reader consequence if wrong: [Low / Medium / High]
Required specialist: [Role/name]

Study record

Citation, identifier, and version: [Enter]
Study design and setting: [Enter]
Participants or material studied: [Sample, inclusion, exclusions]
Exposure/intervention and comparator: [Enter]
Primary and secondary outcomes: [Enter]
Recruitment, allocation, and blinding: [Describe]
Attrition and missing data: [Describe]
Funding and conflicts: [Record]
Protocol, data, correction, or retraction check: [Record]

Result interpretation

Effect size and units: [Enter]
Absolute values and denominator: [Enter]
Uncertainty interval: [Enter]
Analysis limitations: [Confounding, multiple tests, subgroup, other]
Practical or clinical relevance: [Explain cautiously]
Applicability to claimed population: [Strong / Partial / Weak and why]
Replication or synthesis: [Describe]
What the study does not establish: [List]

Editorial decision

Decision: [Keep / Qualify / Rewrite / Remove / Escalate]
Final wording: [Write]
Expert reviewer and date: [Name, YYYY-MM-DD]
Recheck trigger: [New review, replication, correction, guidance]

The full study, not only an abstract or release, was reviewed.
Association and causation are distinguished.
Effect magnitude, denominator, and period are not hidden.
Animal, laboratory, surrogate, and human outcomes are distinguished.
Uncertainty and important limitations appear near the claim.

How to use this template

Convert the proposed sentence into a precise population, exposure, comparison, outcome, and claim type.
Read the full primary study, protocol, corrections, and relevant evidence synthesis when available.
Evaluate design, measures, analysis, effect magnitude, uncertainty, bias, funding, and applicability.
Compare the wording with what the evidence directly establishes and revise overstatement.
Record unresolved questions, expert review, final language, and a recheck trigger.

Translate the published sentence into an evidence question

Copy the exact claim and identify whether it asserts description, association, causation, prediction, mechanism, safety, or effectiveness. Define the population, intervention or exposure, comparison, outcome, period, and setting implied by the wording. Then compare those elements with the study rather than its press release. A result in cultured cells, animals, a narrow volunteer sample, or a short surrogate outcome cannot automatically support a broad claim about people or lasting benefit. Mark whether the sentence reports one study, a body of evidence, or expert interpretation; each requires different support and attribution.

Inspect design, analysis, and magnitude

Record study type, recruitment, allocation, controls, blinding, attrition, measurement method, prespecified outcomes, analysis choices, and funding. Examine effect size and uncertainty, not only whether a threshold was crossed. Ask whether multiple comparisons, subgroup analysis, missing data, confounding, or selective reporting could change interpretation. Statistical significance does not establish practical importance, and an observational association does not by itself establish cause. Check whether the abstract matches the full results and whether corrections, retractions, protocols, data, or independent replications exist. Absence of replication should appear as uncertainty, not as proof that a result is false.

Match language to the whole evidence record

Describe what the study found in its tested conditions and attribute limits clearly. Avoid “proves,” “breakthrough,” “safe,” “works,” or “no risk” unless the evidence and qualified review truly justify that language. Use absolute values where they help readers understand relative changes, and preserve denominators and time frames. When studies disagree, inspect differences in population, design, outcome, dose, and method instead of counting papers. A systematic review can be valuable, but its quality depends on its question, included evidence, bias assessment, and currency. Record the final wording and have an appropriate expert review consequential conclusions.

See the fields in context

Fictional example: short sleep intervention

The intervention, journal, sample, and findings below are invented and must not be cited as research.

Draft claim: “A two-minute audio exercise cures insomnia.”
Imaginary evidence: One unblinded, 20-person pilot measured self-reported sleep onset for seven nights without a control group.
Limit: The design cannot establish a cure, comparative effect, durability, or safety for other populations.
Decision: Remove the cure claim and describe only that the fictional pilot explored feasibility.
Escalation: A qualified sleep researcher reviews any health-facing wording.

Frequently asked questions

Is peer review proof that a claim is correct?

No. Peer review is one quality process. Study design, analysis, reporting, replication, corrections, and the broader evidence still require evaluation.

What is the difference between statistical and practical significance?

Statistical measures describe evidence under a model. Practical significance asks whether the effect's size and context matter to real decisions.

Can a single study support a headline?

Sometimes it can accurately report that a study found something, but broad conclusions usually require context, limitations, and the wider evidence record.

How should preprints be handled?

Label their status, review the methods carefully, check for later versions, and use proportionate caution because formal review and correction may still occur.