Do not classify a source with one yes-or-no test. A peer-reviewed paper may be outdated or methodologically irrelevant, while an official dataset without peer review may be the strongest primary source for a particular statistic. Scholarly usefulness depends on quality, transparency and fit with the claim you intend to make.
“Scholarly” describes a process that others can examine
A robust source makes authorship, purpose and evidential basis visible. Empirical work explains how material was selected, collected and analysed. A theoretical paper shows the concepts and earlier arguments on which it builds. A research report identifies its commissioning body, institution and limitations. In each case, readers should be able to follow the reasoning and trace important statements to data or earlier publications.
This does not require a source to be infallible. Scholarship advances through criticism and correction. A publication that reports uncertainty, documents versions and corrects mistakes can be more trustworthy than one that claims complete certainty. Principles of good research practice emphasise traceability, documentation and honest attribution of other people’s work.
Peer review is evidence, not a warranty
Review can identify weaknesses and improve disciplinary framing. It cannot guarantee correct results, honest data or an appropriate source for your purpose. Continue to inspect methods, sample, conflicts and post-publication notices. Crossmark and publisher records may show whether an article has been corrected, updated or retracted.
A preprint can be timely and important, but its status differs from a version of record. Label it, look for later versions and limit the confidence of your wording. Theses, conference papers, standards, legislation and official statistics each have their own publication pathways. Evaluate their role rather than pretending that every useful source should resemble a journal article.
A strong source can still be wrong for your sentence
Imagine a carefully designed study of English essays written by school pupils. You want to cite it for a claim about German-language doctoral theses. The journal and method may be excellent, yet the studied population, language and text type do not match. Narrow the sentence, find corroborating evidence or state the limits of transfer.
Currency matters in the same contextual way. An older foundational work can be the right source for an established concept or historical argument. The same age may be unacceptable for a software version, AI model, current law or contemporary prevalence estimate. “Recent” is not a fixed number of years; it depends on the rate at which the subject changes.
Worked example: a striking percentage
A magazine reports that a particular percentage of all students use AI for writing. Its link leads to a press release, which summarises a survey commissioned by a technology provider. Before repeating the number, locate the complete report. Inspect participant recruitment, wording of the question, fieldwork dates, definition of “using AI”, commissioning interest and groups that were not sampled.
The percentage might accurately describe respondents in that online panel but not students as a whole. You can report it with that boundary. If the full report cannot be found, the number is unsuitable for a strong prevalence claim. The press release may remain useful as an object of media analysis, but it is not a replacement for the underlying research.
Document the judgement you made
- Write down the exact proposition that needs evidence.
- Identify source type and publication status.
- Inspect responsibility, method, data and citation trail.
- Look for corrections, retractions and later versions.
- Compare population, period and definition with your sentence.
- Record limitations in the same note as the quotation or paraphrase.
- Word the claim no more broadly than the evidence permits.
Use primary, secondary and tertiary source roles to classify the item. The guide to questionable journals adds publisher checks. The source evaluation hub connects the complete workflow.