How to Evaluate a Source

A source is not the same thing as evidence

An article, book, television program, scientific paper, or video is a source of information. The mere existence of a source, however, is not evidence for everything it claims. If a news article says that “a new study proved” an effect, the article is our route to the claim; evaluation requires the original study, its methods, data, limitations, and later scrutiny.

A good source therefore usually allows us to trace a path back toward the origin of the information. If there is no link to the study, document, recording, or dataset, independent checking becomes harder. Traceability does not guarantee truth, but it greatly improves our ability to evaluate a claim.

Who actually created the information?

The web can hide authorship remarkably well. Professional design, an institutional-sounding name, and confident prose are not evidence of expertise. We should first ask whether the author is identified, what relevant knowledge they have, and whether their identity and qualifications can be checked outside their own website.

The same applies to organizations. A description such as “leading research institute” is self-presentation until we establish who runs it, what it actually publishes, how it is funded, and how independent sources describe it.

Do not stay only on the site you are evaluating

Wineburg and McGrew compared 10 PhD historians, 10 professional fact-checkers, and 25 Stanford undergraduates. Historians and students often stayed on the website and evaluated its appearance and content, while fact-checkers quickly left it, opened other tabs, and investigated who stood behind it and what other sources said. This strategy is known as lateral reading.

In practice, when we encounter an unfamiliar site we do not have to begin by reading everything it says about itself. We can open another tab and check the organization, ownership, funding, expertise, and correction history. The sources used for that external check also need evaluation, but this prevents the original site from controlling the entire context in which we judge it.

Expertise must match the question

A title is not a universal passport to authority. If someone is introduced as “Dr. X,” the next question is: doctor of what? Archaeological dating requires one kind of expertise, molecular biology another, and interpretation of an ancient text yet another.

The ACRL information-literacy framework describes authority as constructed and contextual: different communities may recognize different forms of authority, while the information need helps determine the type and level of expertise required. This does not make credentials irrelevant; it means they should carry only the evidential weight justified by the question.

A primary source matters — but it is not automatically true

If we want to know what someone actually said, a full recording is normally closer to the source than a social-media summary. If we are studying a historical letter, a scan is closer to the original material than a modern article about it. If we are checking a study result, the original paper is closer to the data than a news story.

But “primary” does not mean “infallible.” An eyewitness can be mistaken, a chronicler can exaggerate, a company can present its own data selectively, and a researcher can make a methodological error. Primary source means proximity to the origin; truth still requires evaluation.

A secondary source can sometimes be more useful

For the words of a historical figure, their own document may be excellent evidence. To understand how that document fits among hundreds of sources from the period, a strong historical monograph may be more useful. A secondary source can compare primary material, identify limitations, and add context unavailable to an individual participant.

There is therefore no universal rule that primary sources are good and secondary sources are inferior. The right hierarchy depends on the question we are trying to answer.

How was the information created?

An opinion, news report, official statistic, scientific study, systematic review, court document, and advertisement are produced through very different processes. They differ in editorial control, access to data, expert review, correction mechanisms, and purpose.

The ACRL framework therefore also emphasizes “Information Creation as a Process”: evaluating information requires attention to how it was produced, not just its final format. A polished PDF or elegant site may exist without rigorous checking, while a plain technical report may be the product of demanding measurement and review.

“Peer reviewed” does not mean “proved”

Pre-publication peer review is an important mechanism of scholarly communication, but it is not a seal of infallibility. Research on peer review shows that reviewers do not necessarily detect all major errors and that the effectiveness of different review systems is difficult to measure. Peer review is therefore a meaningful quality signal, not the end of evaluation.

The scientific record also changes after publication. Nature Portfolio formally uses corrections, editor’s notes, expressions of concern, and retractions, linking them to the original articles. For an important study, we should therefore check not only whether it was published, but whether it was later corrected, retracted, replicated, or placed in a broader body of literature.

Check what the study actually found

A great deal of misinformation arises not in the original source but in its summary. A study reports an association and a headline converts it into causation; an animal experiment becomes a claim about human treatment; a preliminary result becomes “science has proved.”

For consequential claims, compare the summary with the original conclusion. What did the authors measure? Which population was studied? How large is the uncertainty? What limitations do the authors themselves state? A source can be sound while a later interpretation of it is exaggerated.

Ten articles can still mean one source

Ten websites repeating the same claim may look like ten confirmations. But tracing the links may reveal that B copied A, C copied B, and several others relied on the same wire story or press release. All branches may lead back to a single root.

That is not ten independent pieces of evidence. It is one information stream repeated many times. For widely circulated claims, it can be useful to sketch a source tree and ask which branches are genuinely independent.

Who has an interest?

A company has an interest in selling a product, a political organization in gaining support, a media outlet in audience attention, and a researcher may have financial or professional relationships. An interest does not by itself prove that a claim is false, but it is relevant to the risk of bias.

COPE guidance for reviewers requires disclosure of potential competing interests and explicitly includes personal, financial, intellectual, professional, political, and other relationships. The better question is not “does this source have an interest?” but “is the interest disclosed, and can the claim be checked independently of it?”

Transparency is a strong signal

A trustworthy source does not merely state a conclusion; it shows a path to it. Historical work identifies archives and documents, statistics define what was measured, research describes methods and literature, and strong journalism distinguishes established facts, sourced claims, and commentary.

A good source may also say “we do not know,” “the data are limited,” or “this is our interpretation.” Such limits are not necessarily weakness. They can be evidence that uncertainty is being represented rather than hidden.

Check the date — and the version

Correct information can become wrong by becoming outdated. Laws, health guidance, population counts, scientific classifications, and the facts known during a developing event can all change. A date is therefore not a minor technical detail.

For important sources, also check whether the record was later amended. Modern publishing systems may add corrections, warnings, or retractions. We need to know which version of the information we are actually reading.

Good design and a URL are not methodology

Professional appearance, official-sounding names, logos, and domain endings are easy signals to imitate. Wineburg and McGrew found that such features often misled historians and students, while professional fact-checkers moved more quickly to the question of who actually stood behind a site.

The same caution applies to domains: .org does not automatically mean neutral, .com does not automatically mean unreliable, and a university domain may host a personal page rather than an institutional position. A domain is a clue, not a conclusion.

Wikipedia, AI summaries, and social media can be routes to sources

Wikipedia can be a useful starting point for orientation, terminology, and references. An AI summary, forum post, or social-media thread can likewise alert us to a document, study, or event. For an important claim, however, we should continue to the material on which the summary rests.

This matters especially in an environment where the same summary can be copied through many intermediaries. The closer we move to the origin, and the better documented the path, the less room there is for accidental or deliberate distortion.

Does the source represent its opponents fairly?

A source does not have to be neutral to be useful. But one good indicator of argumentative quality is whether it can accurately represent the strongest version of a position it opposes. If it attributes a claim to opponents that they do not hold, or selects only the weakest examples of the other side, greater caution is justified.

This does not by itself prove the conclusion false. It tells us something about the quality of the reasoning and how much additional independent checking may be needed.

The best source depends on the question

If we ask what a politician actually said, we want the full recording or transcript. If we ask about the economic consequences of a policy, we need statistics, research, and expert analysis. If we ask how a particular person experienced an event, their testimony may be directly relevant.

We do not evaluate a source in isolation. We evaluate the relationship between a source and a claim.

Nine questions for evaluating a source

1. Who is the author or organization, and can the identity be verified outside the source itself?
2. Does the author have relevant expertise for this specific question?
3. Where does the information originate, and can we trace it to the original document, data, study, or recording?
4. How was the information created — as opinion, news, research, statistics, advertising, a report, or something else?
5. What do the cited data actually support, and does the conclusion go beyond them?
6. Are there genuinely independent confirmations, or are multiple publications repeating the same source?
7. What interests and incentives are present, and are they disclosed?
8. Is the information current, and are there corrections, newer versions, or retractions?
9. What do we learn about the source from outside it through lateral reading?

This is not a mathematical scoring formula. It is a checklist for questions that are otherwise very easy to skip.

No badge can replace judgment

Peer reviewed, university, government, doctor, major media outlet, independent researcher, primary source — each label can tell us something useful, but none guarantees truth. A prestigious journal reduces some risks, not all; an expert brings knowledge, not infallibility; a primary document brings proximity to the origin, not a guarantee of accuracy.

The strongest sources are usually those for which we can trace the path from claim to methods, data, context, and independent checks with relatively little ambiguity.

Trust can be provisional

In ordinary life we cannot re-check everything from first principles. Civilization depends on distributed expertise, so trust is necessary. But rational trust can be provisional: it rests on good reasons and changes when better evidence appears.

This is the same posture developed in Openness Without Gullibility, Skepticism Without Dogma: openness without gullibility and skepticism without dogma. We do not require perfect certainty before every action, but we should be able to explain why a source deserves a particular degree of trust.

A source is the beginning of a trail, not its end

When someone says “I have a source,” that is a useful beginning. The next questions are: what kind of source, a source for what, where did its information come from, how strong is its method, does it represent the evidence accurately, are there independent confirmations, and has the record changed since publication?

The goal of How to Evaluate a Source is not to create a white list of sources we always trust and a black list we always reject. That would quickly become another form of dogma. The goal is to improve judgment: whom we trust, for which claim, how much — and why.

The best source is not necessarily the loudest or most prestigious one, but the one that is competent for the question, traceable, transparent, and well connected to evidence.