Start with “What exactly am I studying, and what is this item doing in that study?” rather than “What kind of document is this forever?” A newspaper report may be primary evidence in a history of public reaction, yet serve as background reporting in another project. A journal article may report original experiments, synthesise previous studies, or become the object of a rhetoric analysis. A sound label therefore includes the discipline, question and claim.
The categories are roles, not a league table
Primary sources supply the material you investigate at first hand. Interviews, field notes, laboratory measurements, archive letters, speeches, legislation, photographs, source code and articles reporting new experiments may all serve this role. Primary does not mean unbiased or correct. A witness can misremember; a sensor can be miscalibrated; an official record can omit people. The material is primary because your analysis works directly on it.
Secondary sources interpret, compare, contextualise or criticise primary material. Research monographs, analytical papers, scholarly commentaries and many reviews belong here for the questions they address. They let you understand concepts and debates, but the label is no quality guarantee. You still inspect the method, evidence trail, scope and publication status.
Tertiary sources organise and point into a body of knowledge. Encyclopaedias, handbooks, bibliographies, catalogues and database records often play this part. They are efficient starting places for vocabulary and discovery. When they are used to support a contested central claim, trace their reference to the underlying research instead of treating the summary as the final evidence.
Resolve edge cases by function and disciplinary practice
An experimental paper may be a primary study in a meta-analysis because it supplies newly collected outcomes. The same paper is primary evidence for a historian who studies how scientists described uncertainty at that time, but for a different reason. A systematic review is secondary to the trials it combines. If your project compares review methods, however, the reviews themselves form your primary dataset.
Government publications also mix functions. A statistical release is primary evidence for the figures reported. Its commentary may interpret trends and draw on outside studies, which gives those passages a secondary function. Record data, method and commentary separately rather than attaching one label to an entire PDF.
In literature, film and art research, the work under analysis is usually primary material. Criticism of it is secondary. A catalogue raisonné may look tertiary because it inventories works, yet contain new archival evidence or original attribution arguments. Classify the chapter or dataset you actually use, not merely the book cover.
Worked example: student use of generative AI
Suppose your question is how students use generative AI while writing. Interviews you conduct and their transcripts are primary material. A published survey can act as a primary study when you compare its newly collected results. A paper interpreting several surveys is secondary literature. A reference entry defining “generative AI” or a bibliographic index is tertiary and helps you orient the search.
Change the question to how newspapers frame student AI use and the newspaper stories become your primary corpus. The survey may now provide context or become an external claim whose media representation you compare. The documents did not change; their role did. This is why lists of file types fail when detached from a question.
Classification does not assess reliability
Source role and source quality are separate judgements. A primary item is not automatically superior. A current scholarly handbook may be more suitable for a stable definition than one historical document. A dated letter can be closer evidence for what its writer stated on a particular day than a later biography. Select for the claim, then evaluate authorship, provenance, method and limitations.
Do not confuse original material with an original physical file. A reputable digitisation of an archive letter can still be primary material, although you are viewing a reproduction. You must inspect provenance, completeness and any transcription or image processing. A translation can provide access but adds interpretive decisions; consult the source language where your argument depends on exact wording.
Make the decision auditable
Add four fields to your research log: research question, unit examined, source role and reason. For example: “Parliamentary speech — primary — its wording and argumentative structure are analysed.” A second entry might say: “Historical journal article — secondary — it compares the speeches and proposes an interpretation.” A one-line explanation is more informative than a coloured P/S/T tag with no context.
If one publication performs several jobs, split the uses. Cite the data section for a reported number and the discussion for the author’s interpretation. Follow a tertiary record until you find the originating publication. Then apply the scholarly-source evaluation grid to decide whether it supports your claim. Use the secondary citation guide when the original is genuinely unavailable, and keep the complete workflow in the source-evaluation hub.