Desk Rejection in Biomedical Journals: How to Diagnose the Problem, Repair Your Manuscript, and Use AI Responsibly

A desk rejection occurs when an editor declines a manuscript before sending it for external peer review. It is common, especially at selective biomedical journals. An editorial from Parasites & Vectors reported a pre-review rejection rate of 39% for that journal and cited a rate of 78% for JAMA Internal Medicine in 2017. These figures do not apply to every journal, but they show that editorial screening can be a more immediate barrier than peer review. (Dantas-Torres, 2022)

Authors often interpret a desk rejection as proof that their study is scientifically weak. That conclusion is not always justified. During initial screening, an editor usually considers a narrower set of questions:

• Does the manuscript fit the journal’s scope?
• Is the contribution relevant to the journal’s readers?
• Is the main result clear from the title and abstract?
• Are serious methodological problems already visible?
• Does the manuscript follow the journal’s reporting and submission requirements?
• Are the references, declarations, figures, and supplementary files reliable?
• Does the paper appear ready for reviewers to evaluate?

A study can contain valid data and still fail one of these checks. The correct response is not to submit the same file immediately to another journal. First determine whether the problem concerns the selected journal, the manuscript, the underlying study, or the submission package. Each problem requires a different correction.

What does a desk rejection mean?

A desk rejection may also be called an editorial rejection, pre-review rejection, initial-screening rejection, or rejection without external review.

The decision is usually made by an editor-in-chief, handling editor, section editor, or editorial team. Depending on the journal, the editor may examine only the title, abstract, cover letter, and submission files, or may perform a rapid review of the complete manuscript.

Common reasons reported by biomedical editors include:

• A mismatch with the journal’s scope
• Limited relevance to the journal’s audience
• Insufficient novelty or priority
• Inappropriate or incomplete study design
• Missing methodological information
• Weak data reporting
• Missing ethics approval or trial registration
• Poorly prepared figures or tables
• Language that prevents accurate evaluation
• Plagiarism, excessive text similarity, or redundant publication
• Failure to follow the journal’s submission instructions

The Pakistan Journal of Medical Sciences, for example, lists scope mismatch, old references, missing ethics approval, unregistered clinical trials, weak language, and failure to follow author instructions among the reasons manuscripts may be declined during initial screening. (Jawaid and Jawaid, 2019)

A desk rejection does not prove that the analysis is incorrect or that the study has no value. It shows that the manuscript did not provide enough evidence of relevance, clarity, methodological credibility, or compliance to justify external review.

Diagnose the rejection before revising the manuscript

Start with the decision letter. Ignore standard wording such as:

Thank you for considering our journal. We receive more manuscripts than we can publish.

That sentence does not identify the problem. Look for the part that refers specifically to your submission.

Do not assume that “outside our scope” means the paper should be sent elsewhere without revision. The title and abstract may have represented the study poorly, or the next journal may require a different emphasis, article structure, reporting checklist, and cover letter.

1. Determine what type of problem you have

Before editing individual sentences, classify the failure.

A journal-selection problem

The manuscript does not match the journal’s scope, audience, selectivity, or accepted study types.

Examples include:

• A single-center prevalence study submitted to a journal that prioritizes multicenter clinical research
• A case report submitted to a journal that rarely publishes case reports
• A technically valid diagnostic model submitted to a general medical journal without evidence of clinical utility
• A locally focused public health survey submitted to an international journal that requires broader applicability
• A narrative review submitted to a journal that accepts only systematic reviews

Improving the grammar will not solve a journal-selection problem. The primary correction is to choose a journal that publishes the type of evidence you produced.

A manuscript problem

The study may be suitable, but the manuscript does not explain it clearly.

Typical signs include:

• The title identifies the disease but not the contribution.
• The abstract describes the procedure but does not report the main result.
• The introduction presents a broad clinical problem but no specific knowledge gap.
• The contribution is described using unsupported words such as “novel,” “robust,” or “effective.”
• The results section does not report effect sizes or uncertainty.
• The discussion exaggerates clinical implications.
• The terminology changes between the abstract, methods, and results.
• The figures are difficult to interpret.
• The conclusions do not match the primary outcome.

These problems require manuscript revision rather than a new experiment.

A study problem

The design, data, analysis, or evidence cannot support the main conclusion.

Examples include:

• A diagnostic study has no appropriate reference standard.
• A prediction model is evaluated only on the data used to develop it.
• A case-control design is presented as if it estimates population prevalence.
• A retrospective association is described as a causal effect.
• A clinical trial was not prospectively registered.
• The primary outcome was changed without explanation.
• The sample is too small for the proposed subgroup analyses.
• Missing data are ignored without justification.
• A diagnostic model reports only accuracy despite substantial class imbalance.
• A treatment is described as clinically beneficial based only on a laboratory biomarker.
• A model developed at one hospital is described as generalizable without external validation.

A new cover letter cannot correct these problems. The study may require additional analysis, further data collection, external validation, reduced claims, or a different research question.

A submission-package problem

The manuscript may be suitable, but the uploaded submission violates a requirement.

Examples include:

• Missing ethics approval
• Missing informed-consent statement
• Missing clinical trial registration number
• Incorrect article category
• Missing reporting checklist
• Missing data availability statement
• Incomplete authorship declarations
• Figures uploaded in the wrong format
• Identifying information left in a blinded manuscript
• References that do not follow the required style
• Missing supplementary methods
• Failure to disclose related manuscripts or preprints
• Missing declaration of AI use when required

These failures are usually preventable.

2. Check whether the journal is a genuine match

Reading the journal’s aims and scope is necessary, but it is not enough. Scope statements are often broad. Recent publications show which topics, study designs, and contribution levels the editors actually select.

Open the journal’s two most recent issues and examine at least 15 research articles. Record:

• Medical specialty
• Clinical or biological question
• Study design
• Population
• Sample size
• Primary outcome
• Type of contribution
• Geographic scope
• Article category
• Level of clinical or scientific significance

Then compare those characteristics with your manuscript.

A practical journal-fit score

Assign 0, 1, or 2 points in each category.

A low score does not mean that the study lacks scientific value. It means the journal is unlikely to recognize that value.

Biomedical example

Suppose you completed a cross-sectional study of medication adherence among 180 patients attending one outpatient diabetes clinic.

The study may be useful for local service improvement. However, it may be a weak match for a general medical journal that primarily publishes multicenter trials, nationally representative cohorts, and studies that change clinical practice.

A diabetes education, pharmacy practice, regional public health, or healthcare quality journal may be a better match.

Do not manufacture broader relevance by claiming that one clinic represents all patients with diabetes. Select a journal whose readership needs the evidence you actually collected.

3. Make the research gap and contribution visible

Editors cannot infer the contribution from the time spent collecting data. They evaluate the scientific claim presented in the manuscript.

Before revising the introduction, answer four questions:

1. What is already known?
2. What remains uncertain?
3. What did this study examine or change?
4. Why does the answer matter?

A useful contribution statement follows this structure:

Previous studies have shown [established knowledge], but [specific uncertainty or limitation] remains unresolved. We therefore [precise study action] in [population or setting]. We found [main result], which indicates [bounded scientific or clinical meaning].

Weak example

Sepsis is an important global health problem. Several machine-learning models have been developed to predict sepsis. This study proposes a novel and robust artificial intelligence model for early sepsis prediction.

Problems:

• “Important” adds no specific information.
• “Several models” does not identify the limitation in prior work.
• “Novel” is asserted rather than demonstrated.
• “Robust” is undefined.
• The population and validation setting are absent.
• The clinical purpose is unclear.

Stronger biomedical example

Existing sepsis prediction models have reported high discrimination during internal validation, but their performance often decreases when they are evaluated in hospitals that were not represented during model development. We externally validated an electronic health record model in adult intensive-care patients from three independent hospitals. Discrimination decreased from an area under the receiver operating characteristic curve of 0.86 in the development cohort to between 0.71 and 0.75 across the external cohorts, indicating limited transportability without local recalibration.

This version states:

• What previous studies established
• What remained uncertain
• What the study did
• What was found
• What the result means

It does not claim that the model improves patient outcomes because the study did not test clinical implementation.

4. Match the strength of the claim to the evidence

Many biomedical manuscripts are rejected because the conclusion is stronger than the design permits.

Weak conclusion

The biomarker can be used to improve early cancer diagnosis.

More accurate conclusion

The biomarker distinguished patients with confirmed cancer from controls in this retrospective case-control sample. Prospective evaluation in a clinically representative population is required before its diagnostic utility can be established.

Precision does not weaken a paper. It shows that the authors understand what their evidence can and cannot support.

5. Test the title and abstract

Editors often form their initial judgment from the title, abstract, and cover letter. These sections must report the contribution and result, not only the subject.

The one-sentence result test

Give the title and abstract to a colleague who did not conduct the study. Ask:

What did this study find?

A weak answer sounds like this:

The authors studied vitamin D and respiratory infections.

That identifies only the topic.

A useful answer sounds like this:

In a prospective cohort of older adults, lower baseline vitamin D concentration was associated with more respiratory infections, but the association weakened after adjustment for frailty and chronic disease.

That identifies the finding and an important qualification.

When the reader can describe only what you studied, revise the abstract.

Use a five-part abstract structure

A clear biomedical abstract should normally contain:

1. Clinical or scientific problem: What specific problem motivated the study?
2. Knowledge gap: What was not known?
3. Methods: What design, population, exposure, intervention, or test was used?
4. Results: What was found?
5. Interpretation: What does the result support?

The results should include quantitative evidence when appropriate:

• Number of participants
• Effect size
• Confidence interval
• Sensitivity and specificity
• Absolute risk difference
• Hazard ratio or risk ratio
• Calibration measure
• Difference from the comparison group
• Adverse event rate
• Missing-data rate
• External validation performance

Weak result sentence

The intervention significantly improved patient recovery.

Stronger result sentence

Median time to hospital discharge was 5.2 days in the intervention group and 6.1 days in the control group, an adjusted difference of −0.8 days (95% confidence interval, −1.3 to −0.2).

The stronger sentence tells the editor:

• Which outcome changed
• How much it changed
• In which direction
• How uncertain the estimate is

Do not use “significant” as a substitute for reporting the result.

6. Inspect methodological problems that an editor can see quickly

An editor may not reproduce your analysis during screening, but several problems are visible from the abstract or a rapid methods review.

Before submission, confirm that the manuscript answers the following questions.

Research question and design

• Is the primary research question explicit?
• Does the selected design answer that question?
• Is the primary outcome defined?
• Was the primary analysis prespecified?
• Is the study exploratory, confirmatory, diagnostic, prognostic, or causal?
• Does the language reflect that purpose?

Population and sampling

• Is the source population described?
• Are inclusion and exclusion criteria justified?
• Is the recruitment period reported?
• Is the sample clinically representative?
• Is the sample-size calculation or justification provided?
• Is participant flow reported?

Clinical trials

• Was the trial registered before the first participant was enrolled?
• Is the registration number reported?
• Do registered and published outcomes match?
• Are allocation, blinding, and randomization procedures described?
• Are harms reported?

The International Committee of Medical Journal Editors recommends registration of clinical trials in a public registry at or before the first participant provides consent for enrollment. Ethics approval does not replace prospective trial registration. (ICMJE clinical trial registration guidance)

Diagnostic and prediction studies

• Is the reference standard appropriate?
• Were index-test assessors blinded to the reference standard?
• Is the clinical setting representative of intended use?
• Are discrimination and calibration both reported?
• Was external validation performed?
• Were missing values handled appropriately?
• Is the decision threshold justified?
• Was performance compared with current clinical practice?

For clinical prediction models, the TRIPOD+AI statement provides updated reporting guidance for regression and machine-learning methods.

Statistical analysis

• Are effect sizes reported with uncertainty?
• Are confidence intervals included?
• Are assumptions of statistical tests addressed?
• Is multiple testing handled appropriately?
• Are subgroup analyses prespecified?
• Are missing data reported and managed?
• Are sensitivity analyses explained?
• Do tables contain the values required to verify the claims?

Do not hide an important limitation behind general language. State it and reduce the claim accordingly.

7. Use the correct biomedical reporting guideline

A reporting guideline does not repair a weak study design after data collection. It helps authors report what was done clearly and completely.

The EQUATOR Network maintains a searchable collection of health research reporting guidelines. Common examples include:

• CONSORT: Randomized controlled trials
• STROBE: Observational studies
• PRISMA: Systematic reviews and meta-analyses
• STARD: Diagnostic accuracy studies
• TRIPOD+AI: Clinical prediction models using regression or machine learning
• CARE: Case reports
• COREQ: Qualitative interview and focus-group studies
• ARRIVE: Animal research
• SPIRIT: Clinical trial protocols
• PRISMA-P: Systematic review protocols

The EQUATOR Network organizes these resources by study type and specialty.

Complete the relevant checklist before submission. For every checklist item:

1. Identify where the information appears in the manuscript.
2. Add missing information when it is available.
3. State when an item does not apply.
4. Do not claim compliance when the required information is absent.
5. Upload the completed checklist when requested.

8. Run a literal submission-compliance check

Open the journal’s author instructions and create a list of every requirement containing words such as:

• “Must”
• “Required”
• “Mandatory”
• “Should include”
• “Authors are expected to”

Check the actual files that will be uploaded, not your memory of the manuscript.

Manuscript requirements

• Correct article type
• Word limit
• Structured abstract
• Required abstract headings
• Required keywords
• Correct reference style
• Maximum number of figures and tables
• Line numbering
• Blinded or non-blinded version
• Required title-page information

Ethics and research reporting

• Ethics committee or institutional review board approval
• Approval number
• Informed consent
• Consent for publication of identifiable information
• Trial registration
• Data availability statement
• Code availability statement
• Protocol availability
• Funding disclosure

The ICMJE recommends independent ethics review for research involving human participants and requires written consent for publication when identifiable patient information is essential to the report. (ICMJE protection of research participants)

Author and publication declarations

• Author contributions
• Conflicts of interest
• Funding sources
• Acknowledgments
• Previous presentation
• Related manuscripts
• Preprint disclosure
• AI-use declaration when required

Submission files

• Cover letter
• Title page
• Main manuscript
• Figures
• Supplementary methods
• Reporting checklist
• Graphical abstract
• Highlights
• Suggested reviewers
• Previous decision letter, when required

A missing ethics statement or trial registration number is not a minor formatting issue. It can prevent the manuscript from proceeding.

9. How AI use can contribute to desk rejection

Using AI does not automatically make a biomedical manuscript unacceptable. AI can support language editing, organization, and error checking. However, problems introduced by AI, failure to verify its output, or violation of a journal’s policy can contribute to rejection.

The current ICMJE recommendations state that authors remain responsible for all submitted material, should review AI-generated output for inaccuracies, should not list an AI system as an author, and should disclose how AI-assisted technologies were used. (ICMJE guidance on AI use by authors)

AI-related problems include:

• Fabricated references
• Incorrect reference metadata
• Real articles attached to unsupported claims
• Invented clinical evidence
• Changed numerical results
• Altered causal language
• Missing methodological conditions
• Generic or exaggerated novelty claims
• Inconsistent terminology
• Undisclosed generative use
• Uploading confidential patient or manuscript information
• AI-generated or altered scientific images that violate journal policy
• Text that the authors cannot explain or defend

The risk is not that a sentence has a particular “AI style.” The risk is that the submitted manuscript contains unreliable, unverified, inappropriate, or undisclosed material.

AI can fabricate references

Never trust a reference because it contains plausible authors, a journal title, a year, and a DOI-shaped string.

A study published in Scientific Reports examined 636 references generated by earlier versions of ChatGPT. In that evaluation, 55% of GPT-3.5 references and 18% of GPT-4 references were fabricated. Among references that corresponded to real publications, substantive metadata errors were also common. These percentages describe the models and test conditions used in 2023, not every current AI system, but they demonstrate why independent reference verification is necessary. (Walters and Wilder, 2023)

Elsevier’s current journal policy explicitly warns that AI-generated references can be incorrect or fabricated and states that including fabricated references may lead to rejection. (Elsevier generative AI policies for journals)

Treat every reference suggested by AI as an unverified search lead, not as a source ready to be cited.

A real paper may not support the connected sentence

Finding the paper does not complete the verification.

AI may identify a real publication but attach it to a statement the article does not support. It may:

• Confuse an association with a causal effect
• Apply an animal-study result to humans
• Generalize an adult study to children
• Attribute a secondary outcome to the primary analysis
• Remove an important confidence interval
• Cite a study protocol as if it contained results
• Cite a systematic review for a numerical value reported only in one included study
• Describe a surrogate outcome as a clinical benefit

Biomedical example

Suppose the manuscript states:

Metformin reduces cardiovascular mortality in patients with prediabetes.

An AI system may suggest a real article about metformin and metabolic outcomes. That article may report glucose concentration, progression to diabetes, or cardiovascular risk factors, but not cardiovascular mortality.

The reference exists, but it does not support the sentence.

For every citation, ask:

1. Does the source examine the same population?
2. Does it examine the same exposure, intervention, test, or disease?
3. Does it report the outcome stated in the manuscript?
4. Does it support the direction and strength of the claim?
5. Is the evidence direct?
6. Would a primary source be more appropriate?

Open the article and locate the exact result, table, figure, or discussion passage that supports the sentence.

10. A reliable AI-assisted reference workflow

Do not copy AI-formatted references directly into the manuscript.

Use the following procedure for every reference suggested by AI.

Step 1: Locate the article independently

Search for the exact title, author, or DOI using:

• The publisher’s website
• PubMed
• Crossref
• A trusted institutional database

Step 2: Open the publisher’s version of record

Use the final article page maintained by the publisher whenever it is available.

Confirm:

• Complete title
• Author names
• Journal title
• Publication year
• Volume and issue
• Page range or article number
• DOI
• Correction or retraction status

A DOI should resolve to a registered landing page. Crossref explains that DOI resolution can be checked by opening the DOI link and confirming that it reaches the correct article record. (Crossref DOI verification guidance)

Step 3: Read the relevant source content

Do not cite an article based only on:

• Its title
• An AI summary
• A search-result snippet
• The abstract of a complex study
• Another article’s description of it

Read enough of the source to understand the methods, population, result, and limitations.

Step 4: Verify the connection between claim and citation

Create a temporary audit table:

This process identifies citation drift, in which a source gradually becomes attached to a broader claim than it originally supported.

Step 5: Import the reference from a reliable record

Prefer citation metadata from:

1. The publisher’s version of record
2. PubMed or another authoritative biomedical index
3. Crossref
4. A verified reference-manager import

Do not manually reproduce an AI-generated reference when an authoritative record is available.

Step 6: Perform a final reference audit

Confirm that:

• Every reference exists.
• Every DOI resolves to the correct article.
• Every cited paper supports the connected statement.
• Every in-text citation appears in the reference list.
• Every reference-list item is cited in the manuscript.
• No retracted article is used without an explicit reason.
• Corrections and updated versions have been considered.
• Primary studies are used when a specific original result is discussed.

11. How non-native English researchers can use AI safely

English-language editing can be a real barrier for researchers whose first language is not English. The solution is not to avoid AI completely, nor to let AI generate the scientific reasoning.

ICMJE guidance recognizes that AI tools may be useful when authors are writing outside their primary language, but authors must still verify the accuracy and validity of the final text. (ICMJE guidance for editors)

Use AI as an editor of your reasoning, not as its source.

Step 1: Write the scientific version yourself

Prepare the first version using:

• Your research question
• Study protocol
• Verified methods
• Results
• Tables and figures
• Literature you have read
• Your interpretation
• Your limitations

The first draft does not need perfect English. It must contain the correct scientific meaning.

Step 2: Ask AI to perform a narrow task

Avoid open instructions such as:

Rewrite this to sound professional and academic.

That instruction allows the system to change the argument, certainty, terminology, and evidence.

Use a controlled instruction:

Edit this paragraph only for grammar, sentence structure, and clarity. Preserve the scientific meaning, technical terminology, numerical values, citations, and degree of certainty. Do not add evidence, claims, explanations, or references. Mark any sentence whose meaning is unclear instead of guessing.

For a results section, use:

Correct grammar and readability only. Do not change numbers, units, statistical values, group names, sample sizes, or the direction of any result.

For a discussion section, use:

Improve clarity while preserving the distinction between association and causation. Do not strengthen the claims or remove limitations.

Step 3: Compare the original and edited versions

Review changes sentence by sentence.

Check:

• Clinical meaning
• Technical terminology
• Participant numbers
• Units
• Effect sizes
• Confidence intervals
• Statistical significance
• Causal language
• Negations
• Limitations
• Citations

A grammatically smoother sentence is not an improvement when it changes the scientific meaning.

Step 4: Replace vague language with evidence

AI-assisted academic text often contains polished but empty statements.

Vague

The findings provide valuable insights into the management of cardiovascular disease.

Specific

The association was strongest among participants younger than 60 years, but the confidence interval included no effect in older participants.

Vague

The proposed diagnostic method demonstrated robust performance.

Specific

Sensitivity was 84% in the development cohort but decreased to 67% in the external hospital cohort.

Vague

Further research is needed.

Specific

A prospective study is needed to determine whether use of the score changes treatment decisions or patient outcomes.

Words such as “important,” “robust,” “effective,” “promising,” and “significant” should not replace a result.

Step 5: Ask a human reader to examine high-risk sections

Prioritize human review of:

• Title
• Abstract
• Research gap
• Main contribution
• Results
• Limitations
• Conclusion
• Cover letter
• AI-use declaration

The final manuscript should contain language that every author understands and can defend.

12. Do not write for AI-detection software

Do not intentionally add grammar mistakes, unusual synonyms, or awkward sentences to make a manuscript appear human-written. Do not use “humanizer” tools designed to evade detection.

These practices can:

• Reduce clarity
• Change scientific meaning
• Introduce factual errors
• Create inconsistent terminology
• Conceal rather than disclose AI use
• Produce text that the authors cannot defend

One 2023 study tested seven public AI detectors on 91 essays written by non-native English speakers and reported an average false-positive rate of 61.3%. The study concerned student essays and specific detectors available at that time, not biomedical manuscripts or every current detection system. It nevertheless shows that a detector score should not be treated as conclusive evidence of authorship. (Liang et al., 2023)

Protect yourself through a transparent and traceable writing process:

• Keep the original outline.
• Preserve early drafts.
• Use tracked changes.
• Save analysis code.
• Retain laboratory or clinical research notes.
• Maintain a verified reference library.
• Record which AI tool was used.
• Save the prompts and relevant outputs when appropriate.
• Document how the output was reviewed.
• Disclose AI use according to the journal’s policy.

The goal is not to make the manuscript appear less polished. The goal is to ensure that the reasoning, evidence, and final responsibility remain human.

13. Understand the journal’s AI policy

AI policies differ between publishers and may change. Check the exact journal policy before every submission.

ICMJE

The current ICMJE recommendations state that authors should disclose whether AI-assisted technologies were used, describe how they were used, review generated content for accuracy and bias, and remain responsible for originality and proper attribution. AI tools should not be listed as authors. (ICMJE AI recommendations)

Elsevier

Elsevier permits supportive use of generative AI during manuscript preparation but requires human verification and oversight. Its journal policy requires disclosure of generative AI use, while basic spelling, grammar, and punctuation checks do not require declaration. It also warns that fabricated references may lead to rejection. (Elsevier journal AI policy)

Springer Nature and Nature Portfolio

Springer Nature states that authors remain responsible for accuracy, originality, and integrity. AI-assisted copy editing of human-written text generally does not require declaration when it is limited to readability, grammar, spelling, punctuation, tone, wording, and formatting. Generative content creation must be documented according to the applicable journal policy. (Springer Nature AI guidance)

Do not assume that a policy used by one publisher applies to another.

14. Protect biomedical data and confidential material

Before uploading any manuscript content to an AI system, examine the tool’s privacy, retention, and data-use terms.

Do not upload:

• Patient names
• Medical record numbers
• Identifiable clinical narratives
• Unredacted imaging data
• Facial images
• Genomic information linked to individuals
• Restricted clinical datasets
• Confidential trial information
• Unpublished collaborator data without permission
• Peer-review reports
• Manuscripts you are reviewing
• Copyrighted material you are not permitted to process

ICMJE states that manuscripts under editorial consideration are privileged communications and warns that uploading confidential manuscripts to AI systems may violate confidentiality.

For your own manuscript, remove identifiable information before using an external tool and confirm that your institution, ethics approval, data-use agreement, and journal policy permit the intended processing.

15. Do not use AI to alter biomedical evidence

AI tools must not be used to invent or modify research results.

High-risk uses include:

• Creating synthetic patient observations and presenting them as collected data
• Changing values in a results table
• Generating missing experimental outcomes
• Altering microscopy images
• Adding or removing bands from western blots
• Changing features in radiological or pathological images
• Modifying patient photographs
• Generating a figure that is not derived from the reported data
• Creating a false clinical trial registration
• Inventing ethics approval information

Elsevier identifies fabrication or alteration of data, references, scientific images, microscopy images, western blots, and patient images as inappropriate AI use.

When AI or machine learning is part of the research method, describe it reproducibly in the methods section. Include the tool or model, version, task, inputs, validation process, and human oversight when relevant.

16. Write a useful AI-use declaration

Use the wording required by the target journal. Do not copy a declaration from another publisher without checking its policy.

A declaration should normally identify:

• Tool or service
• Purpose
• Sections or tasks affected
• Extent of use
• Human review performed
• Author responsibility

Example for language editing

During manuscript preparation, the authors used [tool and version] to identify grammatical errors and improve sentence clarity in author-written text. The tool was not used to generate scientific claims, results, interpretations, or references. All suggested changes were reviewed by the authors, who take full responsibility for the final manuscript.

Example for structural assistance

The authors used [tool and version] to suggest alternative organization for selected paragraphs in the introduction and discussion. All scientific content, claims, interpretations, and references were written or selected by the authors. The authors independently verified the revised text and take full responsibility for the final content.

Use these only as templates. The journal’s required wording takes priority.

17. Run a final editorial screen before submission

Complete the following checks in order. Stop when a check fails and correct it before proceeding.

Check 1: Journal fit

Confirm that the journal publishes:

• Your topic
• Your study design
• Your article type
• Your contribution level
• Evidence relevant to your intended audience

Check 2: Title and abstract

A reader should be able to identify:

• The clinical or scientific problem
• The knowledge gap
• The study design
• The population
• The main result
• The supported interpretation

Check 3: Contribution

Complete this sentence:

After reading this study, researchers or clinicians will know [specific new knowledge] that was not established before.

When the answer depends only on “novel,” “effective,” “promising,” or “robust,” the contribution remains unclear.

Check 4: Methodological credibility

Confirm that:

• The design answers the question.
• The sample is described.
• The primary outcome is defined.
• The analysis matches the outcome.
• Effect sizes and uncertainty are reported.
• Validation is appropriate.
• Limitations are explicit.
• Conclusions remain within the evidence.

Check 5: Reporting requirements

Confirm that:

• The correct reporting guideline was used.
• The checklist is complete.
• Ethics information is present.
• Trial registration is reported when applicable.
• Data and code statements meet the journal’s requirements.

Check 6: References

Confirm that:

• Every source exists.
• Metadata came from an authoritative record.
• Each source supports the connected sentence.
• No AI-generated reference remains unverified.
• Retracted or corrected articles have been identified.
• References are current enough for the topic.

Check 7: AI use

Confirm that:

• AI did not generate unverified scientific content.
• All numerical values were checked.
• Technical meaning was preserved.
• No confidential information was improperly uploaded.
• The journal’s current policy was read.
• Required disclosure was included.
• Every author can explain and defend the final text.

Check 8: Cover letter

The cover letter should state:

1. What the study found
2. Why the finding matters to the journal’s readers
3. What differs from the closest previous work
4. Whether the manuscript is original and not under review elsewhere
5. Any required ethics, registration, or AI-use information

Do not use journal prestige as the argument for fit.

Weak

We believe that our innovative study is suitable for your prestigious journal and will be of great interest to its broad readership.

Stronger biomedical example

This prospective multicenter study evaluates whether a preoperative frailty score predicts 30-day complications after colorectal surgery. The journal regularly publishes research on perioperative risk assessment and surgical outcomes. Unlike previous single-center studies, our analysis includes external validation across four hospitals and reports calibration as well as discrimination.

18. What should you do after a desk rejection?

Do not revise every section automatically. Revise according to the diagnosed problem.

When the problem is journal fit

• Select a journal that publishes the study type.
• Examine its recent articles.
• Rewrite the title, abstract, introduction, and cover letter for its readers.
• Recheck every submission requirement.

When the problem is priority or general interest

• Determine whether the contribution is too narrow for the journal.
• Do not exaggerate generalizability.
• Target a specialist journal when the evidence addresses a specialist audience.
• Strengthen the significance argument only when supported by the results.

When the problem is presentation

• Rewrite the abstract around the main result.
• Define the research gap.
• Remove empty evaluative language.
• Standardize terminology.
• Improve the figures and tables.
• Review the manuscript for language that changes clinical meaning.

When the problem is compliance

• Correct every missing declaration and file.
• Confirm whether the same journal permits a new submission.
• Do not assume that correcting one missing statement resolves other editorial concerns.

When the problem is methods or evidence

• Pause before submitting again.
• Reassess the design, statistics, validation, and claims.
• Conduct additional analyses when justified.
• Add data or experiments when feasible.
• Reduce the conclusion when the existing evidence supports a narrower claim.

When should you appeal?

Check the journal’s appeal policy before writing.

An appeal is most defensible when:

• The editor made a specific factual error.
• A methodological point was misunderstood.
• A required document was overlooked despite being present.
• A conflict of interest may have affected the decision.
• The journal did not follow its stated procedure.

An appeal is usually weak when it argues only that the study is important or that the authors disagree with a judgment about priority.

Elsevier, for example, does not consider appeals for manuscripts rejected outright by its editorial team and excludes disagreements based only on interest, novelty, or suitability. Its appeals policy applies to peer-reviewed manuscripts and requires supporting evidence. Other publishers may follow different procedures. (Elsevier editorial decision appeals policy)

Where automated manuscript-review tools can help

A manuscript-review system such as Fukurō AI Peer Review can assist by identifying material that requires author inspection, including:

• Missing statements
• Inconsistent terminology
• Differences between the abstract and results
• Unsupported conclusions
• Unclear contribution statements
• Undefined abbreviations
• Reference inconsistencies
• Sections that do not follow the expected structure
• Language that may obscure the scientific meaning

Automated analysis should not make the final decision about:

• Whether the study design is valid
• Whether the interpretation is clinically correct
• Whether novelty is sufficient for a specific journal
• Whether a reference truly supports a claim
• Whether the manuscript should be submitted
• Whether the evidence justifies a clinical recommendation

Use the output as a list of items to verify, not as a replacement for scientific judgment.

Final checklist

Before submission, confirm that you can answer yes to every question.

Journal selection

• Does the journal publish this subject?
• Does it publish this study design?
• Is the article category correct?
• Is the contribution suitable for its readership?

Scientific argument

• Is the research gap specific?
• Is the difference from previous work explicit?
• Does the title identify the central contribution?
• Does the abstract report the principal result?
• Does the conclusion match the evidence?

Methods and reporting

• Does the design answer the research question?
• Are the population and sampling procedures described?
• Are the primary outcome and analysis defined?
• Are effect sizes and uncertainty reported?
• Was the relevant EQUATOR guideline used?
• Are ethics approval and consent reported?
• Was a clinical trial prospectively registered when required?

References

• Does every cited publication exist?
• Was the citation copied from the publisher’s version of record or another authoritative database?
• Does every source support the connected sentence?
• Were all AI-suggested references independently verified?
• Were corrections and retractions checked?

AI use

• Did the authors write and understand the scientific reasoning?
• Was every AI-edited sentence reviewed?
• Were numbers, units, and statistical results preserved?
• Were no claims or references accepted without verification?
• Was confidential information protected?
• Was AI use disclosed according to the journal’s policy?
• Can every author defend every part of the manuscript?

Desk rejection cannot always be prevented. Editors still make judgments about scope, priority, novelty, and readership. However, many avoidable rejections can be prevented by separating four questions:

1. Did we select the correct journal?
2. Does the manuscript communicate the contribution clearly?
3. Can the study support its claims?
4. Does the submission satisfy every requirement?

AI can help authors improve grammar, organization, and consistency, particularly when English is not their first language. It can also introduce fabricated references, unsupported claims, altered meanings, privacy violations, and undisclosed content. The safest workflow is to write the scientific argument first, give AI narrow editing tasks, inspect every change, verify every reference through its authoritative source, follow the target journal’s policy, and retain full human responsibility for the manuscript.

References and useful policies

1. Dantas-Torres F. Top 10 reasons your manuscript may be rejected without review. Parasites & Vectors. 2022;15:418.
2. Jawaid SA, Jawaid M. Common reasons for not accepting manuscripts for further processing after editor’s triage and initial screening. Pakistan Journal of Medical Sciences. 2019;35(1):1–3.
3. ICMJE. Use of artificial intelligence in publishing.
4. ICMJE. Use of AI by authors.
5. ICMJE. Clinical trial registration.
6. ICMJE. Protection of research participants.
7. EQUATOR Network. Reporting guidelines for health research.
8. Collins GS, et al. TRIPOD+AI statement: updated guidance for reporting clinical prediction models. BMJ. 2024;385:q902.
9. Walters WH, Wilder EI. Fabrication and errors in the bibliographic citations generated by ChatGPT. Scientific Reports. 2023;13:14045.
10. Liang W, et al. GPT detectors are biased against non-native English writers. Patterns. 2023;4(7):100779.
11. Elsevier. Generative AI policies for journals.
12. Springer Nature. AI guidance for researchers and publishing communities.
13. Crossref. Verify a DOI registration.
14. Elsevier. Editorial decision appeals policy.