How to Judge a Source: Peer Review, Preprints and Predatory Journals

The short answer

A source is credible enough to cite when you can identify who wrote it, what evidence it rests on, and what quality control it passed before publication — and when none of those answers is "unclear". Peer review in an indexed journal is the strongest single signal, but it certifies that specialists found no obvious flaw, not that the finding is true. Judge every source on its evidence and its independence, not on how authoritative the page design looks.

This is the companion to how to find academic sources, which covers where to search. Searching is the easy half. The hard half is deciding which of the 400 results you are allowed to build an argument on.

What does peer review actually guarantee?

Peer review means the manuscript was sent to researchers in the same field, who read it and recommended acceptance, revision or rejection. It is quality control by people who know the subject, and it catches a lot: unsupported claims, missing controls, statistics that do not follow from the design.

What it does not do:

• It does not verify the data. Reviewers almost never see raw data or rerun the analysis. They evaluate what the authors report.
• It does not guarantee replication. A peer-reviewed finding can fail to replicate, and in several fields many have. Where a claim matters to your argument, check whether anyone has repeated the study.
• It is not uniform. Review at a leading journal in your field and review at a venue that accepts almost everything are both called peer review.

Practical consequence: treat peer review as a floor, not a verdict. A peer-reviewed article still needs you to read the methods section. If a paper's conclusion is load-bearing in your thesis and its sample was twelve undergraduates, that belongs in your text.

Checking whether something is peer reviewed is usually easy. Most databases have a filter for it, the journal's own site describes its review process under "about" or "for authors", and if you cannot find any description of a review process at all, you have your answer.

How do I spot a predatory journal?

Predatory journals collect a publication fee and provide little or no meaningful review. They matter to you because their articles surface in general search results looking exactly like legitimate research, and citing one weakens whatever it is supporting.

No single sign is conclusive. The pattern is what gives it away:

• Unsolicited invitations that flatter, promise rapid publication, and address a field slightly adjacent to your actual work.
• Implausible speed.Peer review takes weeks to months. "Review completed within 72 hours" describes something other than review.
• Fees disclosed late. Legitimate open-access journals publish their article processing charges openly; predatory ones reveal them after acceptance.
• An unverifiable editorial board. Pick two or three names and check whether those people exist at the institutions listed and mention the role themselves. Names are sometimes used without consent.
• A near-identical title to an established journal, or a scope so broad it spans unrelated disciplines in one publication.
• No indexing. Check whether the journal appears where your field indexes — Scopus, Web of Science, PubMed, or the Directory of Open Access Journals for open-access venues. The Think. Check. Submit. checklist is a good neutral reference, and your university library almost certainly has field-specific guidance.

Be careful with the opposite error too. Open access is not a warning sign — many of the strongest journals in several fields are fully open access, and charging an article fee is a normal funding model. Small, new, non-English or regional journals are also not automatically weak. Judge the review process and the indexing, not the business model or the country.

Can I cite a preprint?

A preprint is a manuscript posted publicly before, or instead of, peer review — on arXiv, bioRxiv, medRxiv, SSRN, PsyArXiv and similar servers. In fast-moving fields it is where the current work is, and ignoring preprints can leave a literature review a year behind.

Three conditions for using one:

1. Label it.The reference entry should make clear it is a preprint, and the text should not present it as an established finding. "In a preprint, Amari and Lund (2025) report …" costs four words and is accurate.
2. Check for the published version. Search the exact title. Most preprint servers link the journal version once it appears. Cite what you read — and if a published version exists, read that one.
3. Weight it accordingly. A preprint is fine as supporting evidence or as an indication of where a field is heading. Building a central claim on a single unreviewed manuscript is a risk your examiner will notice.

Some departments restrict preprint use in student work. If yours does, that rule overrides everything in this section.

Is grey literature allowed?

Grey literature is everything published outside commercial and academic publishing: government reports, statistics agency releases, NGO studies, industry white papers, working papers, standards, dissertations. It is not second-class evidence — for many questions it is the only evidence, because nobody publishes national employment figures in a journal.

Judge it on two questions:

• Who produced it, and what do they gain from the conclusion? A statistics office and a trade association can both publish careful work, but one has a position to defend. That does not disqualify the source; it means the interest belongs in your text when you use it.
• Is the method described? A report that states its sample, its period and its definitions can be assessed. One that presents a number with no method behind it cannot be assessed, and an unassessable number is not evidence.

Two specific cases. Working papers from a university or research institute are effectively preprints in economics and the social sciences — treat them the same way. Other students' thesesare usually a poor citation: a bachelor's thesis is coursework, not vetted research. Mine its reference list, cite what it cited.

Primary or secondary — and why it matters

A primary source presents original evidence: the study that collected the data, the interview transcript, the archival record, the statute itself. A secondary source reports on or interprets that evidence: a review article, a textbook, a meta-analysis, a news piece about a paper.

Both are legitimate, but they answer different needs. Secondary sources are excellent for orientation and for showing how a debate is structured. A specific empirical claim should cite the study that produced it.

The reason is not formal. Claims drift as they are retold. A cautious finding about a specific population, repeated three times through review articles and press coverage, arrives as a general fact — and it is the person who cites it fourth who owns the error. Chasing a claim back to its origin also reveals things a summary never shows you: how large the effect was, how narrow the sample, whether the authors themselves hedged.

If you genuinely cannot obtain the original — an untranslated work, a document that no longer exists — cite it as a secondary citation and say so: (Vygotsky, 1978, as cited in Harper, 2020, p. 44). Only the source you read goes in the reference list. What is not acceptable is citing the original as though you had read it. See the APA 7 guide for the mechanics.

How do I know a paper has not been retracted?

Retracted papers keep circulating. The PDF on someone's course page does not update when the journal withdraws the article, and citing a retracted study as support is a genuinely damaging mistake.

• Open the article on the publisher's site, not only the PDF you downloaded. Retraction notices appear there, usually as a banner or a watermark.
• Check the Retraction Watch database for anything central to your argument, and note that Crossref surfaces retraction status against the DOI.
• Look at recent citing articles.If a paper has been questioned, someone has usually written about it, and Google Scholar's "cited by" list is the fastest way to see that.

You do not need to do this for every source. Do it for the handful your argument actually rests on, and for anything whose result seemed surprisingly clean.

Do citation counts and impact factors tell me anything?

Something, but less than they appear to, and in a specific direction.

• Citation count shows that a paper has been engaged with. It does not show agreement — influential papers are cited by their critics. It is also strongly biased by age, so a 2025 paper with four citations may be more current than a 2011 paper with four hundred.
• Journal impact factor is an average across a journal, not a property of the article you are holding. A weak paper in a strong journal is still a weak paper. The figure is also field-dependent to the point that comparing across disciplines is meaningless.
• Absence of citations is weak evidence of anything in a small subfield or a non-English literature.

Use metrics to decide what to read first when you have forty candidates. Do not use them to decide what is true.

The five-minute check

Run this on any source before it enters your reference list:

1. Who wrote it? A named author with a traceable affiliation, or a named institution. If neither, stop here.
2. Where did it appear, and does that venue describe a review process?
3. When? And is that recent enough for the claim it is supporting in your field?
4. What is the evidence? Open the methods or the data section. If there is none, the source is an opinion — which may still be worth citing, as an opinion.
5. Who benefits from the conclusion? Check funding and author declarations. An interest does not disqualify a source; concealing it in your write-up does.
6. Does the identifier resolve? Open the DOI. This catches broken references and fabricated ones in the same motion.

That last step deserves emphasis in 2026. Language models produce references that look immaculate and do not exist — real authors, real journal, plausible title, a DOI that goes nowhere. If you did not retrieve a source yourself, verify it before it reaches your bibliography.

Once the sources are in, the remaining risk is structural: entries in the reference list that never appear in the text, and citations in the text with no entry. Our citation check reconciles those two lists against each other and flags quoted passages with no attribution nearby. It returns no similarity score and no percentage — it cannot measure that, and it does not pretend to.