The rejection was three sentences long and perfectly polite. Well written. Thorough coverage of the literature. Not a sufficient contribution for this venue. You had spent seven months on it, read a hundred and forty papers, and cited ninety-one of them.
That is the standard failure, and it is not a writing failure. You wrote a survey and submitted it as a paper. A paper is not what you found out. It is one claim someone could disagree with, and the evidence that they should not.
Write the claim before you write anything else
One sentence, on its own line, at the top of an empty document. It must pass a single test: a competent person in your field could read it and say "I don't think that's right." If nobody could disagree, you have a topic rather than a claim.
A topic
"This paper examines the relationship between remote work and employee productivity."
Nobody can disagree with an examination.
A claim
"Productivity gains attributed to remote work in 2020–2022 studies are largely an artefact of measuring output rather than coordination cost."
Someone will want to argue. That is the point.
The claim does more work than any other sentence in the project. It tells you which papers belong in the literature review, which data you must collect, which objection you have to answer, and, mercifully, which interesting side-question you are allowed to ignore.
The literature review is an argument, not an inventory
Most rejected literature reviews are organised as a list: Ahmed (2019) found this, Park (2020) found that, Rossi (2021) found the other. Each paragraph is accurate and the whole thing is unreadable, because the reader cannot tell what any of it is for.
A working review is organised by the shape of the disagreement. Three or four positions exist in your field, and your claim lives in a gap between them. Say so:
The four-move review.
1. Here is what the field broadly agrees on, in one paragraph with grouped citations.
2. Here are the two camps that disagree about the part that matters, and why.
3. Here is the assumption both camps share without testing it.
4. Here is the question that assumption leaves open — which is my claim.
This also fixes the citation problem. In an inventory review you cite everything you read, because you cannot justify leaving anything out. In an argument review you cite what the argument needs, which is usually between thirty and sixty sources rather than a hundred and forty, and reviewers read it as confidence rather than as gaps.
Write the sections out of order, on purpose
The reader's order is introduction, methods, results, discussion, conclusion. The writer's order is nothing like it, and following the reader's order is the most reliable way to lose a month on an introduction you will rewrite anyway.
First
Methods
The only section where you already know everything. It is also the section that is checked hardest, and writing it early exposes gaps in your design while you can still fix them.
Second
Results
Facts only, no interpretation. Build the tables and figures first, then write the sentences that point at them. If a figure needs a paragraph to be understood, the figure is wrong.
Third
Discussion
Where the claim is defended. Three obligations: what the results mean, what the strongest objection is, and what the study genuinely cannot show. The third one buys you more credibility than any other paragraph in the paper.
Fourth
Introduction
Easy once the discussion exists, because you now know exactly what the paper turned out to be about. Four moves: the field, the gap, the claim, the shape of the paper.
Last
Abstract and title
The most-read and least-written parts of a paper. An abstract is a miniature of the whole: context, question, method, result, significance, in about 200 words with no citations and no suspense.
Why papers come back
Reviewer reports vary enormously in tone and very little in substance. Five objections cover most of what gets written, and four of them are avoidable before submission.
"The contribution is unclear." There is no claim, or the claim appears for the first time on page nine.
"The conclusions are not supported by the data." Usually one verb. You wrote "shows" where the design only permits "is consistent with".
"Insufficient engagement with recent work." Often means one specific recent paper, sometimes the reviewer's. Search the last two years of the venue you are submitting to before you send.
"Out of scope." A venue problem, not a paper problem, and entirely preventable by reading the last two issues.
"Methodological concerns." The only one that can require real new work. It is also why the methods section is written first.
When a report arrives, write the response letter two days later rather than the same evening, and answer every point in order, including the ones you disagree with. A polite, itemised disagreement with reasons is accepted far more often than silence.
Two of these five are partly tooling problems — a reference you lost and a version you cannot reconstruct. Our comparison of academic writing software covers what to pair with a reference manager and why.
Academic prose is precise, not ornate
The belief that scholarly writing must be heavy costs readers and citations. Long nominalisations, stacked hedges and a passive voice in every sentence do not make a claim more careful, only harder to check.
"It could potentially be suggested that there may exist a relationship between the variables under investigation."
Better: "The two variables appear to be related, though the design cannot establish direction."
Hedge once, precisely, and then stop. Three hedges in a sentence read as evasion rather than caution, and a reviewer will ask you to say what you actually mean.
The passive voice has one good use here: when the actor is genuinely irrelevant. "Samples were stored at four degrees" is correct. "It was decided that" is hiding who decided. The general principle is the same one that governs clear nonfiction anywhere, which our notes on professional writing rules set out in more detail.
Draft it in sections you can move
Writing out of order is painful in a single long file. Writer Atlas has a research project type where each section is its own page in a tree, so you can draft methods first and reorder afterwards, search across everything you have written, and export to DOCX, PDF or LaTeX when the venue tells you which it wants. Keep your references in Zotero or Mendeley — we do not manage citations, and pretending otherwise would cost you a submission.
Start writing freeQuestions about research papers
How long should a research paper be?
The venue decides, and the range across fields is enormous: many computer science conferences cap papers at eight to twelve pages in a fixed template, while humanities journals routinely publish eight to twelve thousand words. Find the author guidelines before you draft, because a paper written at the wrong length is not trimmed, it is restructured.
How many sources should I cite?
Enough that every claim which is not yours is attributed, and no more. For a standard journal article this usually lands between thirty and sixty. Padding is visible: reviewers notice citations that appear once, in a list, doing no argumentative work. What they do notice as a gap is a missing recent paper from the venue you are submitting to.
Can I use "I" in a research paper?
In most fields today, yes, and increasingly it is preferred where it improves clarity about who did what. Some journals still ask for impersonal constructions, and a few disciplines remain strict. Check three recent articles in your target venue; they will settle the question faster than any style guide.
What is the difference between a research paper and a thesis?
A paper defends one claim to people who already work in the area. A thesis demonstrates competence as well as contribution, to a committee, at ten to thirty times the length, and it must show its working: the literature review is exhaustive by design, and methods are justified rather than merely reported. Many theses contain two or three papers inside them.
How do I choose a journal or conference?
Work backwards from your citations. If five of the papers you argue with appeared in the same venue, that venue is your conversation. Then check turnaround time, open access fees, and whether recent issues contain anything methodologically similar to yours. Aiming one tier above your evidence costs six months; aiming one tier below costs readers.
Is using AI acceptable in academic writing?
Policies vary by publisher and are changing quickly, so read your venue's statement rather than assuming. The common position is that AI may assist with language, clarity and structure but cannot be an author, cannot generate results, and must be disclosed where it materially contributed. Never let a model supply citations: fabricated references are the fastest way to lose a reviewer's trust permanently.