Desk Rejection Criteria: What to Screen, in What Order, and What to Leave to Reviewers
Good desk rejection criteria do two jobs at once: they stop the wrong papers reaching your reviewers, and they stop the right papers being held up by checks that belong later in the process. Most journals get the first half roughly right and the second half wrong, which is why screening feels slow and reviewers still receive manuscripts they should never have seen.
What follows is an ordered screen — cheapest checks first, most expensive last — with a clear line between what an editor decides at the desk and what only a reviewer can judge. The companion article covers how to write the rejection so it does not generate an appeal; this one is about what to look at, and in what sequence.
The principle: order by cost, not by importance
Every check has a time cost and a kill rate. Run the ones that are fast and frequently decisive before the ones that are slow and rarely decisive.
A scope check takes sixty seconds and ends perhaps a fifth of submissions. A statistical review takes an hour and ends very few. Running them in the wrong order means paying an hour to discover something a minute would have told you.
So: stop at the first fail. If a paper is out of scope, nothing downstream matters, and no further check should be run on it.
The ordered screen
1. Scope — 1 minute
Does the subject fall within your published aims? Read the title, abstract and conclusion, then re-read your scope statement. If they do not meet, stop.
The common failure is a scope statement too vague to exclude anything. “We publish high-quality research in the health sciences” cannot be failed, which means it cannot be used. If your scope cannot produce a rejection, rewrite the scope.
2. Article type and format — 2 minutes
Is this a type you publish? A narrative review sent to a journal that publishes only original research is a desk return regardless of quality. Check word count, structure, and whether the submission includes the sections your instructions require.
3. Completeness and declarations — 3 minutes
Title page, abstract, keywords, author list with affiliations, corresponding author, funding statement, conflict-of-interest declaration, data availability statement, author contributions. Missing items are a return to the author, not a rejection — but they should be caught now rather than after review.
4. Ethics and registration — 3 minutes
For human research: ethics committee approval with a reference number, and consent. For animal work: the equivalent approval. For clinical trials: prospective registration with a registry number you can actually look up. For systematic reviews: a protocol, usually PROSPERO.
Verify the registration number rather than noting its presence. An unverifiable registration is a serious flag and takes thirty seconds to check.
5. Reporting guideline — 3 minutes
Match the design to the guideline: CONSORT for trials, PRISMA for systematic reviews, STROBE for observational studies, ARRIVE for animal research, CARE for case reports. The EQUATOR Network maintains the full set. You are checking that the required items are reported, not that the study is good.
6. Similarity and AI disclosure — 5 minutes
Read the similarity report rather than the number. A 40 per cent score concentrated in the methods section of a lab protocol is usually fine; an 8 per cent score that is one continuous paragraph lifted from someone else’s discussion is not. Check that any AI-use disclosure your policy requires is present.
7. References — machine check
Do the cited works exist, do they match the record, has anything been retracted? This belongs here because it is now a common failure mode and because it is machine work rather than editor time. See how to check whether a submission’s references are fabricated for the method, or have it run for $18 a paper.
8. Fatal design flaw — 5 to 10 minutes, editor only
The last check and the only one requiring editorial judgement rather than verification. Is there a problem no revision can repair? A design that cannot answer the stated question. A sample that cannot support the conclusion drawn. An outcome measured in a way that does not measure the outcome. A control group that is not comparable.
This is the only domain where “reject” rather than “return” is the correct verdict on the paper’s merits. Everything above it is fixable by the author.
Total: about twenty minutes for a paper that passes everything, and a great deal less for one that fails early.
What to leave to reviewers
Equally important, and more often got wrong. Do not screen for these:
Novelty. Whether a finding is sufficiently new is a specialist judgement that requires knowing the literature better than an editor can across a whole journal’s scope. Editors who desk reject on novelty reject work in the areas they personally know least about.
Whether the statistics are correct. Whether a statistical method was reported is a desk check. Whether it was the right method, correctly applied, is a review question — and in many cases a specialist statistical review question.
Whether the conclusions are justified. Unless the gap is glaring enough to be a fatal design flaw, this needs someone who knows the field.
Quality of writing, on its own. Language that obscures the science justifies a return for improvement. Language that is merely non-native does not, and screening on it systematically disadvantages the authors with the least access to editing support.
Importance. Whether the work matters is precisely what peer review exists to establish.
The line is this: the desk screens whether the paper is reviewable. The reviewers judge whether it is good. Journals that blur this end up making specialist judgements at the desk with none of the specialist knowledge.
Writing your criteria down
Turn the list above into a page on your site, in your own terms, with three things stated: what is screened, what causes a return versus a rejection, and how long a desk decision takes.
Published criteria do three jobs. Authors self-select out, reducing volume. Rejections become applications of a stated rule rather than private judgements. And a new editor or managing editor can run the screen on their first day without absorbing years of tacit practice.
Run them as a scored instrument rather than a mental checklist, and you also get a record — which, as the appeals article sets out, is what makes a contested decision answerable.
Having the screen run for you
Every check above except the last is verification rather than judgement, which is why it can be delegated without touching editorial independence.
NeucitePress runs the ordered screen and returns a scored report out of 100 across six domains, a recommendation, and a rejection letter already drafted — two working days, $25 per paper. For journals wanting every submission screened, Submission Pre-Check runs monthly from $149 by volume. The specialist checks sit alongside: author and reviewer identity at $20, figure integrity at $55, and CONSORT/PRISMA/STROBE compliance at $60.
The fatal-flaw judgement stays with your editors, always. We verify; you decide.
Frequently asked questions
How long should a desk screen take?
About twenty minutes for a manuscript that passes every check, and far less for one that fails early — which is the point of ordering by cost. If your screening routinely takes an hour, the checks are running in the wrong sequence or judgement calls are being made at the desk that belong in review.
What is the difference between a return and a rejection?
A return is for something the author can fix: a missing ethics statement, the wrong format, an incomplete declaration. A rejection is for something they cannot: out of scope, wrong article type, or a design flaw no revision repairs. Conflating the two produces rejections that read as unfair and appeals that succeed.
Should references be checked before or after peer review?
Before, on everything that survives the first five checks. Sending a reviewer a manuscript whose evidence base has not been verified wastes the scarcest resource a small journal has.
Can we desk reject on novelty?
Better not to. Novelty needs specialist knowledge of a narrow literature, which is what reviewers have and editors, across a whole journal’s scope, usually do not. It is also the criterion most likely to be applied inconsistently and therefore appealed successfully.
Who should run the screen?
A managing editor, an editorial assistant, or an outside service can run checks one to seven, since they are verification against stated criteria. Check eight — the fatal-flaw judgement — and the decision itself belong to an editor.