Reading the evidence
How to evaluate a research source.
Before asking what a source says, ask what it is, who checked it, how it was built and who paid for it. Five questions that separate evidence from assertion.
To evaluate a research source, ask five things of it: what kind of document it is, whether it was independently peer-reviewed, how the study was built, who paid for it, and whether anyone else has found the same result. None of the five settles the matter alone. Together they tell you how much weight a source can bear — and whether a confident claim built on it is borrowing more certainty than the source contains.
I. What kind of document is it?
Primary research.
A primary research article reports a study by the people who ran it — the question, the methods, the data and the result. A randomized trial or a cohort study in a peer-reviewed journal is primary research, and it is the closest a reader can get to the evidence itself. A research article is a primary source when it reports original data; its methods section usually makes that plain.
Review articles.
A review summarizes studies that others have published, which makes it a secondary source. A narrative review is an expert’s reading of a field — valuable, and shaped by which studies the expert chose. A systematic review follows a method stated in advance to find and appraise every relevant study, and may pool them in a meta-analysis. Its choices can be checked, which is why it carries more weight.
Preprints.
A preprint is a manuscript posted publicly before peer review, on servers such as medRxiv. Preprints let findings circulate quickly; they have not yet been checked by independent reviewers, and some change substantially before publication or are never published at all. Read them as early signals, not as settled evidence.
II–III. Who checked it, and how was it built?
What makes a source credible.
- Independent peer review. Review by other researchers before publication catches some errors. It is a floor, not a guarantee — peer-reviewed studies are still corrected and retracted.
- Registration in advance. A trial registered before it begins — on ClinicalTrials.gov or a similar registry — has its planned outcomes on record, so a reader can see whether the outcome reported is the outcome that was planned.
- Transparent methods. Randomization, blinding, the number of participants and how many dropped out should all be stated. What a study does not report is part of what it tells you.
- The right people and outcomes. A finding in cells or animals, or on a laboratory marker, is not the same as a finding in people like you on outcomes you would notice.
IV. Who paid for it?
How funding and conflicts of interest are checked.
Most major medical journals ask authors to declare who funded the work and any financial interest in its result, under standards such as the ICMJE recommendations. Look for the funding statement and the competing-interests disclosure, usually at the end of the article or beside the author list. Then ask whether the funder, or any author, stands to profit from one answer rather than another.
This matters because the pattern is measurable. A Cochrane methodology review of 75 papers found that drug and device studies sponsored by the manufacturer more often reported favorable efficacy results and favorable conclusions than studies funded by other sources — a bias its authors judged could not be explained by the standard risk-of-bias checks alone. Funding is a reason for closer reading, not proof that any one study is wrong.
V. Has anyone found it again?
One result is a lead, not a conclusion.
A finding reproduced by independent teams, in different populations, is worth more than the same finding repeated loudly by one. Before treating a claim as established, look for a second study that set out to test it — and for a systematic review that has gathered all of them.
What is not evidence.
A press release, a podcast, a social post, a brand’s own white paper, or a claim that cannot be traced back to a primary document may point toward evidence; none of them is evidence. Encyclopedias, Wikipedia included, can be a useful map of a subject — follow their citations to the studies and judge those. Our own source tiers and exclusions are published on evidence, and the way they feed a rating on how we rate.
Common questions
- What is a primary research article?
- A primary research article reports a study by the people who carried it out: the question, the methods, the data and the results. A randomized trial or a cohort study published in a peer-reviewed journal is primary research. It is the closest a reader can get to the evidence itself.
- Is a research article a primary source?
- When it reports original data — a new trial, a new cohort, a new experiment — yes. When it summarizes other people's studies, as a review article does, it is a secondary source. The methods section usually makes the difference plain.
- What is a review article in research?
- A review article summarizes and interprets studies that others have published. A narrative review is an expert's reading of the field; a systematic review follows a pre-stated method to find and appraise every relevant study, and may pool them in a meta-analysis. Systematic reviews are the more reliable of the two because their choices can be checked.
- What are preprints in research?
- A preprint is a manuscript posted publicly before peer review. Preprints let findings circulate quickly, but they have not yet been checked by independent reviewers, and some change substantially or are never published. Read them as early signals, not settled evidence.
- What are credible sources for health research?
- Peer-reviewed primary studies with transparent methods, systematic reviews that state how they searched, trial registries, and documents from regulators. Credibility rises when a study was registered in advance, discloses its funding, and has been replicated by an independent team.
- How do I check who funded a study?
- Look for the funding statement and the competing-interests or conflict-of-interest disclosure, usually at the end of the article or beside the author list. Most major journals require both. Industry funding is not proof of error, but a Cochrane methodology review found that industry-sponsored drug and device studies more often reported favorable efficacy results and conclusions.
- Is Wikipedia a reliable source for health research?
- It can be a useful map of a subject, but it is not itself evidence. Follow its citations back to the primary studies and reviews, and judge those.
- Is this medical advice?
- No. This page explains how to read a source; it says nothing about whether any treatment, supplement or practice is right for you. Decisions about your health belong with you and a qualified clinician.
This page explains how to read a source. It does not say whether any treatment, supplement or practice is right for you — discuss that with your physician. See our medical disclaimer.
Next: what a meta-analysis can tell you