CONTENTS   /   JUMP TO A SECTION

1.Considering an invitation

Accept when your expertise and available time allow a useful assessment of the manuscript's main claims. Tell the editor which aspects fall outside your competence so that complementary expertise can be arranged.

• Expertise: consider the methods, application context and evaluation design.
• Availability: check both the response deadline and the report deadline. Reply promptly even if declining.
• Interests: disclose relevant relationships, collaboration, competition or financial connections before accepting. Ask the editor when uncertain.
• Alternatives: you may suggest other specialists when declining, but do not forward confidential files or invitations.

Related resources
Read the COPE ethical guidelines for peer reviewers (external resource).

2.Access & confidentiality

Open the invitation link or your Review workspace and complete acceptance before downloading review materials. Use the assigned review page for manuscript files and the review form.

Do not circulate the submission, contact the authors or involve a colleague without editorial permission. Report possible ethical problems privately to the editor with specific observations. Do not conduct a private investigation or make accusations in author-facing comments.

Related resources
Sign in to the journal system. The journal's Publishing Ethics Policy explains reviewer confidentiality.

3.Assessing the contribution

Identify the central claim, the supporting evidence and the assumptions connecting them. Explain the consequence of a weakness rather than simply listing it.

• Relevance: does the work address a meaningful AI question within the journal's remit?
• Originality: what is new relative to the closest relevant work, and is the distinction demonstrated?
• Soundness: are the definitions, derivations and algorithm descriptions sufficient to check the claims?
• Evidence: do experiments or proofs answer the research question, including important failure cases?
• Reach: are conclusions limited to the settings studied and are limitations clear?

4.Technical review of AI research

• Data and evaluation: examine provenance, preprocessing and train-validation-test separation. Look for leakage and unsuitable metrics.
• Comparisons: check relevant baselines and tuning. Consider whether data access, supervision or computing resources explain the advantage.
• Analysis: assess tests of major design choices and whether variation between runs changes the conclusion.
• Reproducibility: identify missing configuration or evaluation details. Examine supplied code within the agreed review scope.
• Applicability: question resource or deployment claims when they are central to the contribution.
• Presentation: check whether notation, captions and visualisations make the evidence understandable.

For theoretical papers, focus on assumptions, correctness and the significance of the result. For surveys, examine literature selection, competing perspectives and the synthesis. Distinguish experiments necessary to establish the claims from suggestions for future research.

Related resources
The Author Guidelines describe the reporting expected from submissions.

5.Writing a constructive report

Give the editor a reasoned assessment and authors a clear route to improve the work. The following structure is recommended.

• Summary and strengths: briefly describe the contribution in your own words.
• Major issues: number concerns affecting validity, interpretation or novelty. Cite locations and explain what evidence or clarification is needed.
• Minor issues: separate local corrections, identifying the section, figure, equation or page.
• Overall assessment: explain how the strengths and remaining issues support your recommendation.

Use professional language focused on the work. Request a citation only when it addresses a specific scholarly point; do not seek unnecessary citations to your own publications.

Related resources
For external training: Elsevier's guide to conducting a review. Submit this journal's report through JAIR, not Elsevier's submission system.

6.Submitting your recommendation

Select the recommendation available in the form that matches your report. The editor makes the publication decision; do not promise acceptance.

• Submit the completed review form and any permitted annotated files through the assigned review page.
• Remove your name and identifying attachment metadata where reviewer anonymity is required.
• Check that the review is shown as submitted. A saved draft is not a completed review.
• If a delay is likely, ask the editor early for an extension or release from the assignment.
• For a technical failure, record the manuscript ID and displayed error and contact the editorial office rather than the authors.

Related resources
Contact the editorial office if you cannot access or submit the assigned review.

7.AI assistance & responsibility

Scientific judgement must remain your own. Do not upload confidential manuscript material to public AI services. Check the journal's policy and privacy safeguards before using any assistant tool.

If language assistance for your own report is permitted, remove confidential details, use an authorised private tool, disclose substantive use as required and verify every change. Where permission is unclear, ask the editor first. You remain responsible for the assessment.

Related resources
External background reading: generative AI guidance for peer review.

8.Revisions & continuing responsibilities

When asked to assess a revision, compare the response and changed manuscript with your previous concerns. State which issues are resolved and which still affect the conclusions. Explain new substantive concerns and avoid changing requirements without scientific justification.

Maintain confidentiality after the assignment ends. Tell the editor if you discover an error in your report or a concern that could materially change the assessment.

Researchers interested in reviewing may contact the editorial office with their expertise and publication record. Selection for an assignment remains an editorial decision.

Need assistance?

Contact the editorial office with your manuscript ID and the details of your enquiry.

External links are supplementary learning resources. The journal's own policies and manuscript-specific instructions take precedence.