AI Policy

Policy on the Use of Generative AI and AI-Assisted Technologies in the GAIA Publication Process

Version 4, amended following the Editorial Board meeting of 25 September 2026

Based on the policy of May 2024, updated using discussions at the workshop at the GAIA-Jahrestreffen, Dübendorf, April 2026, and drawing on the EC Living Guidelines (2024, updated May 2026), DFG Guidelines (March 2026), and resources provided and regularly updated by COPE Guidelines (see further reading below).

Guiding Principles

Generative AI and AI-assisted technologies are tools. What determines their use is the human responsibility that accompanies it.

AI may assist. It must not replace human judgment, authorship, or responsibility.

For authors specifically: Permitted is what is based on one’s own research. Prohibited is what replaces it.

This applies across all stages of the publication process: manuscript preparation, peer review, and editorial decision-making. Regardless of how AI tools are used, responsibility for all outputs rests with the humans involved. GAIA operationalizes this commitment through the following four principles:

  • confidentiality: all parties treat unpublished manuscripts as strictly confidential
  • transparency: all parties disclose their AI use honestly and fully
  • quality: each person using AI tools is responsible for actively verifying and critically assessing the outputs
  • accountability: humans remain responsible for the final work, regardless of which tools were used

The sections below set out what this means for the three groups involved in the publication process: the editorial board and office, authors, and reviewers.

Editorial Board and Office

Editorial assessments and decisions, including manuscript evaluation, accept/reject recommendations, and communication with authors, are exclusively human responsibilities. These cannot be delegated to AI systems.

The editorial office may use AI tools in a supportive capacity for the following tasks:

  • Preliminary formal check: checking submissions against the GAIA guidelines (e.g., structure, length, abstract, reference style)
  • Fact-checking support: flagging possible errors and checking references for accuracy and existence
  • Copy-editing: fine adjustment of language and style (e.g., abstracts, lead texts, captions, acknowledgements, text sections) and formatting of reference lists
  • Discoverability: improving article discoverability (e.g., aligning titles with academic search engine optimization principles)

AI output is a suggestion only. The editorial office reviews every result, verifies flagged issues against the original sources, and decides what is adopted.

The editorial office publishes a general statement on its website and in the author guidelines describing which AI tools it uses and for which tasks, and updates it whenever this changes.

Authors and reviewers are themselves responsible for meeting their disclosure obligations under this policy; the editorial board and office do not verify or enforce individual compliance.

Confidentiality Conditions

Manuscript content (text, data, figures) must not be uploaded to any AI system unless the processing meets a documented data-protection standard, such as:

  • The AI runs locally on an own device (no cloud processing), or
  • The AI is hosted under institutional data governance that is demonstrably compliant with applicable data protection law (e.g., GDPR), or
  • A commercial AI service is used under a contractual data protection guarantee, with active privacy settings that prevent use of inputs for model training

What qualifies is the verifiable data-protection guarantee itself, not the type of provider — an institutionally hosted service must meet the same standard as a commercial one, and vice versa.

Authors

The central question is not “Did you use AI?” but “Do you take full responsibility for all content in this manuscript?” Accountability rests with the human author(s), regardless of which tools were used.

GAIA uses the STM Recommendations for a Classification of AI Use in Academic Manuscript Preparation (STM Association, September 2025) as its reference framework. The nine categories describe what AI was used for, not whether it was used.

# STM Category GAIA Position
1a Language refinement (grammar, style, clarity) Declaration required; permitted with correct disclosure
1b Condensing or shortening text* Declaration required; permitted with correct disclosure
2 Writing or drafting manuscript content Declaration required; extended declaration (see below)
3 Translation Declaration required; permitted with correct disclosure
4–8 Data processing, visualization, illustrations, coding, literature search and management Declaration required; permitted with correct disclosure
9 Presenting AI content as own research without own evidence base Prohibited

* GAIA deviates from the STM classification by listing condensing or shortening text separately from language refinement (STM category 1).

Declaration. Authors indicate in the author declaration form, by ticking the relevant categories, which AI uses apply to their manuscript, and state the name of the AI tool, its version, and its manufacturer. For categories 2 and 4–8, a corresponding statement must also be included in the methods section or acknowledgements. Categories 1a, 1b and 3 are recorded in the author declaration form only. AI tools must not be credited as authors or co-authors.

Declared AI use is not in itself a reason for rejection. The declaration serves transparency and traceability; the usual quality standards apply in full, and editors may use the information to look more closely at the relevant parts of a manuscript.

Note on STM category 2 “Writing or drafting manuscript content”

Category 2 is different from all other categories: it concerns AI generating the written expression of the research itself. This is permissible where the underlying research (the data, the questions, the findings) is genuinely the author’s own and is fully declared.

GAIA permits AI-assisted writing or drafting of manuscript content, subject to full declaration. Responsibility for all content rests with the human author; declaration makes this transparent to the GAIA community and its readers, enabling them to form their own judgment.

Declaration must go beyond naming the tool used. Authors must specify:

  • Any passages of AI-generated text used verbatim or near-verbatim, marked as such (e.g., in quotation marks), analogous to a direct quotation from a source
  • Any AI-generated ideas, arguments, or lines of reasoning adopted into the manuscript, cited indirectly, analogous to citing a secondary source
  • Whether the AI’s contribution influenced the manuscript’s key arguments or conclusions
  • How the author personally reviewed and verified the AI-generated content

Reviewers

Reviewers are entrusted with confidential material. Confidentiality is non-negotiable: it protects the authors’ intellectual property and the integrity of the review process. Any AI use that compromises this confidentiality is prohibited.

Reviewers’ substantive assessments, evaluations, and conclusions must be their own. Subject to the conditions below, reviewers may use AI tools in a supportive capacity, for example, to improve the language of their review text, to orient themselves in the relevant literature, or to structure their thoughts.

Reviewers disclose any use of AI tools in the review process, including use solely for language refinement or translation of their review text. The disclosure is made to the editor in the review form, indicating the tool and its purpose. Manual verification of all outputs remains mandatory, and responsibility for any AI-generated content remains with the reviewer.

Confidentiality Conditions

When any AI tool is used in connection with manuscript content, the confidentiality conditions set out under Editorial Board and Office above apply equally to reviewers, with “own device” and “institutional data governance” read as the reviewer’s own.

Manuscripts must not be uploaded to private AI accounts.

Reviewers should be aware that AI-generated references may be fabricated. Any reference cited in a review must be verified by the reviewer before submission; reviewers should not adopt AI-suggested references for their own statements without checking that they exist and say what is claimed. Tools such as CheckIfExist can support this verification.

This policy will be reviewed and updated regularly to reflect evolving technologies and emerging norms in scholarly publishing. GAIA’s editorial office welcomes ongoing dialogue with the editorial board and the community.

Further Reading

EC Living Guidelines (2024, updated May 2026), DFG Guidelines (March 2026), and resources provided and regularly updated by COPE Guidelines

COPE

DFG Guidelines (March 2026)

European Commission. EC Living Guidelines (2024, updated May 2026).

STM Association (September 2025)