How to write a research paper you can defend

Most guidance on writing a research paper treats drafting as the hard part, as though the problem were getting words onto the page. The harder problem is keeping the manuscript inspectable: every claim traceable to a source, every citation resolving to the passage it is supposed to support, every submitted file matching what the analysis actually produced.

This article describes a source-first workflow for a research paper, from the working question through the final preflight. It produces a scoping note, a traceable source library, a draft with passages attached, and a verification log. It does not decide whether your research question is worth asking, choose a target journal, or replace your field's reporting standard.

1. Scope the claim before you search

Write the working question as one sentence that can be tested, not as a topic label. Then list three to five claims that might answer it and mark which ones could be supported by primary sources in the time available. Claims resting on unqualified novelty are the ones that collapse under review; claims tied to a specific population, setting, or method are the ones you can defend. Guidance on framing a research gap covers the difference between the two.

A lot of writers search before they know what they are looking for, which produces volume rather than evidence. Define the claim first, then search to find out whether it survives. If the literature already closes the gap, or if no primary source supports the claim, narrow it or drop it and record that decision. A scoping note is not a summary of the literature. It is a record of what the paper will and will not claim, and it is what you return to when the draft starts to drift.

The rejection test is blunt: can this claim be supported, in the time available, by sources a reviewer would accept as primary evidence? If not, shrink it until the answer is yes. Every later section has to defend the same boundary, so the cost of an oversized claim compounds.

2. Build the literature set by chaining, not by volume

Start with three seed papers from a recent review, then trace backward through their reference lists and forward through papers that cite them. Chaining surfaces both the earlier foundations and the later developments, and it is more robust than keyword search when a field's terminology is fragmented. A practical walkthrough of the method appears in this literature review guide.

Treat the abstract as a gate rather than a destination. If the paper belongs in the working set, extract only the material that bears on your argument. Group notes by theme rather than by source, so the eventual map mirrors the structure of the paper instead of the order in which you happened to read. A paper on measurement error may belong beside a paper on sampling bias if both support the same point; the reader will never see your reading order, so the notes should not preserve it. This literature review workflow sets out the same sequence.

Each retained source should answer three questions: what does it claim, how does it get there, and which exact passage could carry a sentence in the draft? A source that cannot answer all three is not ready for the manuscript, and a paper you have only skimmed should not be cited as though you had read it closely. Record the search itself as you go, using a reproducible literature search log, and run a deduplication pass once the same paper starts arriving through more than one path.

3. Hold the IMRaD lines on purpose

Introduction, Methods, Results, and Discussion each carry a different kind of evidence, and a draft reads as confused mainly when those boundaries blur. The structure is not decoration; it is what lets a reviewer find the part they need to check. APA papers use a fixed sequence of title page, abstract, introduction, method, results, discussion, and references, with the title page, abstract, references, tables, and figures on their own pages, as summarized in this APA format guidance. The formal article took this shape historically because structured sections made findings easier to inspect and compare, a development traced in this history of the scientific article.

The introduction names the gap and signals the contribution. It is not a second literature review. Methods do something different: they let another team repeat the work, which means stating the data source, extraction date, version or sampling strategy, software and version, inclusion rules, and any analytic decision that would change the outcome if altered. A reporting framework for healthcare database studies catalogues that class of decision in detail; the principle generalizes even where the specific fields do not. A simple test separates the two: a sentence explaining why the study exists belongs at the front, a sentence explaining how it was done belongs in Methods.

Results should report what was found and stop. Interpretation belongs in Discussion, as the UCI scientific writing guide makes explicit. If a sentence in Results contains an evaluation verb — shows, suggests, proves — check whether it has crossed the line. Figures and tables should carry the data load, with the text pointing toward them rather than restating every value. Reporting a mean without a measure of spread is the common failure here: a single number can conceal the variation that makes the result interpretable, which is why Purdue OWL's statistics guidance emphasizes showing variability where possible.

4. Keep citations portable across styles

Citation work is mechanical until a source is added mid-draft, at which point it becomes a source of errors. The defense is to keep the underlying source record clean and let the format be applied late. The same set can be reformatted without retyping when the metadata is consistent.

Numbered styles need careful renumbering when the source set changes: backshift the later numbers, then check every bracket against the bibliography. Author–date systems fail differently — the text stays stable while the reference list quietly drifts — so the final pass matters in both. Normalize DOIs as well, stripping the resolver prefix and the trailing punctuation that different databases introduce, so a reference that looks complete also resolves.

Three mistakes recur. Numbered and author–date systems get mixed in one draft. Access dates get dropped for online-only sources in styles that require them. A preprint gets cited as though it were the published article, with no note of its status. None are hard to fix, and each one makes the manuscript look unfinished.

5. Draft against the passage, not against memory

A fluent paragraph can still be untethered from its evidence. The pattern that causes it is familiar: remember the idea, write from memory, add the citation afterwards. By then the wording may no longer match what the source said, and the mismatch is hardest to catch in exactly the places it matters most — definitions, thresholds, and distinctive terminology.

Source-grounded drafting inverts the order. Identify the claim, attach the supporting passage, paraphrase against the attached text, then confirm the citation resolves to that same passage. Keep the passage visible until the sentence is final. This also guards against a clean rewrite silently changing a claim: a model can produce a sentence that reads well and means something different, and the manuscript remains the author's responsibility either way.

A draft is defensible when every sentence has that trail. Each in-text citation points to a saved source record. Each paraphrase has a parent passage. Each number in Results matches a statement in Methods. Each quotation appears exactly as it does in the source. Anything skimmed but not read in full stays out of the core claims. The list is strict because a polished paragraph without a traceable passage is a liability, not an asset.

6. Preflight the sources, not just the prose

Read the manuscript once for unsupported claims and orphan citations specifically, rather than for style. Confirm every figure and table is cited in the text and numbered correctly, that data and code statements match what was actually archived, and that the reference list has not drifted from the in-text calls. A claim-by-claim method for that pass is set out in this submission audit guide.

Then check the sources themselves. A reference that was sound during drafting may have been retracted or corrected since, and that change matters more than any wording. Re-run the reference formatting after the journal style is confirmed, with DOIs normalized and broken links repaired. The remaining items are administrative but still block acceptance: length limits, figure requirements, supplementary file names, a cover letter matching the scope statement, ORCID details, and conflict-of-interest disclosures.

The standard for the last mile is simple enough to state. A paper is ready when it is not only written but inspectable: when a reader who asks “where did this come from?” already has an answer on file. Everything above is in service of being able to produce that answer without reopening the search.