Guidelines for reviewers of academic articles – Use of AI

1) Core principles

  • The reviewer bears full responsibility for the content of the review, regardless of whether artificial intelligence (AI) was used during its preparation.
  • The use of AI is permitted only as a supportive tool, not as a substitute for expert judgement.
  • The reviewer must ensure confidentiality, independence, and the protection of personal or non-public data contained in the manuscript.

2) Acceptable use of AI

AI may be used in the following cases:

  • Assistance with wording – minor stylistic revisions of the review text, summarising the reviewer’s own text, improving linguistic clarity.
  • Support in fact-checking or contextual understanding – general searches for information on methods, concepts, and theories; checking publicly available sources (with caution regarding potential AI “hallucinations”).
  • Analysis of article structure – recommendations on how to formulate comments, assistance in maintaining a neutral and constructive tone.
  • Translation of the reviewer’s own report.
  • In all cases: the review must reflect the reviewer’s expert opinion and must not be an automatically generated text without critical oversight.

3) Prohibited practices

The reviewer must not:

  • Upload parts or the full manuscript to public AI tools – this may breach confidentiality, licensing terms, or GDPR.
  • Generate review conclusions without expert assessment – AI must not decide on the acceptability of the article or replace the reviewer’s scientific judgement.
  • Rely on unsupported interpretations – AI models may “hallucinate” and produce false or inaccurate information; all outputs must be verified.
  • Attempt to identify authors or reviewers using AI – this is not permitted in double-blind review processes.
  • Use AI to verify citations without manual checking – AI may fabricate non-existent sources; the reviewer must verify the references manually.

4) Recommendations for safe use of AI

  • Always use institutionally approved AI tools (preferably paid versions with appropriate data protection, rather than freely accessible public chat tools).
  • Do not upload unpublished data, figures, tables, sensitive information, or the full manuscript to AI tools without clear data protection rules.
  • Verify all AI-generated content.
  • Let AI assist, not decide.