Desk Rejection in 2026: A 5-Gate Pre-Submission Audit for German Research Teams

Meyer and colleagues analysed 369 desk-rejected manuscripts at Academic Medicine [1]. An ineffective study question or design was cited in 92%, suboptimal data collection in 49% and a misaligned topic in 37%. Unclear writing appeared in only 24%, and most rejected papers had three or more problems. In a separate series of 898 rejection reports, the median time to a desk rejection was 3 days, against 42 days for rejection after peer review [2]. An editor can reach a design verdict quickly. Polished English does not change it.

Papers that pass the desk face a different filter. In Bordage’s analysis of reviewer comments, the leading problems were inappropriate or poorly described statistics (11.2% of negative comments) and overinterpretation of results (8.7%) [3]. Overinterpretation is common. Boutron and colleagues found spin in more than 68% of abstracts of randomised trials with non-significant primary outcomes [4]. A pre-submission check that ignores statistics and spin misses what reviewers look for first.

DEAL changes the cost of resubmission, not the risk of it. DEAL agreements with Elsevier, Springer Nature and Wiley cover hybrid open access for eligible corresponding authors. Publish-and-read fees are €2,550, €2,600 and €2,200–3,150 per article respectively, and gold APCs are discounted by 20% [5]. The APC is often covered. The real cost of a desk rejection is time.

The DFG rewards fewer, stronger papers. Since March 2023, DFG proposals list at most ten publications, and reviewers may not consider the journal impact factor or h-index [6]. The DFG Code (v1.2, January 2025) asks authors to choose the publication medium “with due regard for its quality and visibility” (Guideline 15) and sets authorship rules (Guideline 14) [7].

CONSORT 2025, published on 14 April 2025, replaces CONSORT 2010. It has a 30-item checklist with seven new items, three revised and one deleted, plus a new open-science section covering registration, protocol access and data sharing [8]. SPIRIT 2025 applies the same logic to protocols, with 34 items [9]. Editors now expect the report, protocol and registry entry to agree on the primary outcome and analysis plan. A mismatch is visible at triage.

Disclosure is rarely done voluntarily. In a manual check of 200 computer-science preprints from February 2024, only two disclosed LLM use [10]. The rules are now explicit:

• ICMJE (January 2026): describe AI use in the cover letter and in the manuscript. Put writing assistance in the acknowledgments and analysis use in the methods. Chatbots cannot be authors [11].
• DFG (statement of 21 September 2023): disclose which generative models were used, for what purpose and to what extent. Tools used only for grammar, spelling or translation need not be documented [12]. Journal rules may be stricter, so follow the stricter one.
• Publishers: Elsevier requires a separate declaration of generative AI use in manuscript preparation. Springer Nature asks for LLM use to be documented in the Methods, with AI-assisted copy-editing exempt [13, 14].

Estimated share of abstracts with LLM-modified content, by field, as reported by Liang et al. [10]. Estimates, not counts of individual papers.

Score each gate red, amber or green. Any red gate means do not submit.

Give each gate one owner and one evidence file:

• G1: the target journal’s contents from the last 24 months, with three comparable papers named.
• G2: one sentence stating what the study can and cannot show. If it cannot be written plainly, the paper is not ready.
• G3: the completed checklist with page numbers, with the registry entry and protocol side by side.
• G4: the statistical analysis plan beside the results, and the abstract’s conclusion checked against the primary outcome only.
• G5: the disclosure statement, data-sharing statement and CRediT table, checked against the journal’s instructions.

An amber gate means fix it before you submit; a red gate means rethink the claim or the journal. Score the gates again after the fixes. Run G2 and G4 with someone who did not write the paper, because authors are the least reliable judges of their own design and conclusions.

If G2 is red because the design cannot answer the question, no editing will fix it. The honest options are to narrow the claim or change the target journal. A case report or a pilot study needs a different checklist (CARE or the CONSORT pilot extension), so confirm the right one before starting.

Did any gate come back amber? Send us your manuscript and target journal. Pubrica’s PhD- and MD-qualified reviewers return a gate-by-gate audit, a completed reporting checklist and a reviewer-style report. See Pubrica’s pre-submission review.