Purpose and Context

The Documenting Generative AI Interactions in Research Interest Group takes stock of a fast-moving standards landscape, identifies where cross-domain documentation is missing, and extends the Three Ps framework to cover that gap. Working Groups then formalise the extended framework as a standard.

Core Problem and Approach

Current approaches centre on disclosure, which leaves a reviewer reading “ChatGPT was used for data analysis” unable to assess the use’s appropriateness or the judgement exercised.

We distinguish between disclosure (acknowledging GenAI use) and documentation (capturing methodological details sufficient for peer evaluation and reproducibility). Our central thesis is that GenAI interactions constitute research methodology and warrant documentation standards comparable to those for research software, laboratory protocols, and data management practices, and that no cross-domain standard yet specifies what that documentation must contain.

The inadequacy of disclosure-only approaches is empirically documented. The systematic review underpinning the CHART reporting guideline found that one of 137 chatbot health advice studies fully described the version of the model it evaluated; publisher instructions to authors vary widely in what disclosure they require (Ganjavi et al. 2024); and normative frameworks (the ICMJE Recommendations and the European Commission’s ERA Living Guidelines) already demand documentation that no standard yet specifies. This Interest Group exists to close the specification gap.

Our approach centres on the Three Ps framework, our own contribution and not yet published: Prompt (what input was given to the tool), Process (how the tool was used), and Provenance (tracking the interaction’s role in research). This framework provides high-level guidance while remaining adaptable across disciplines and GenAI modalities.

The relationship between documentation and disclosure is asymmetric: from a Three Ps record, a disclosure statement compliant with any current journal or funder policy can be generated automatically, while no disclosure statement, however detailed, can reconstruct the underlying interaction record. Documentation subsumes disclosure; disclosure cannot recover documentation.

Prior Work

This Interest Group builds on prior work at Macquarie University on AI research policy (Macquarie University Guidance Note, Framework Paper) and on preregistration and open data in the humanities. Two lineages ground the approach, one treating tacit research practice as requiring structured documentation, the other treating process records as research objects. We apply both to the researcher’s interaction with GenAI (full sources in the Statement of Work).

Activities and Outputs

The Interest Group will pursue four primary activities over 12 months. Landscape Analysis: inventory existing and emerging standards, guidelines, and initiatives; preliminary scanning (2026) identified more than twenty initiatives across three clusters, concentrated in disclosure, with no cross-domain specification of the record itself. Systematic analysis is the group’s first task. Framework Development: extend the Three Ps framework through iterative testing across GenAI modalities (text, image, code, audio/video generation) and research disciplines, and publish exemplar tooling that implements it, prototypes in hand. Stakeholder Engagement: structured consultations with related RDA groups, publishers, research infrastructure providers, and standards bodies. Working Group Specification: evaluate and prioritise potential Working Group outputs through community input (metadata schemas, minimum information guidelines, research grimoire frameworks, example tooling, publisher recommendations, and ethics guidelines), producing detailed Statements of Work ready for community review.

Distinctiveness and Positioning

Our work occupies one layer of a stack. Cross-domain policy from COPE, ICMJE, and the major publishers requires that GenAI use be declared; disclosure taxonomies such as the Global Reporting Standard establish what must be declared; machine-actionable encodings such as RO-Crate and W3C PROV carry the records; and domain reporting guidelines specialise requirements for particular fields. This Interest Group supplies the layer each of those presupposes, namely what researchers must capture about a GenAI interaction.

An Interest Group is the appropriate first step: the initiative has emerged from a Birds of a Feather session, and the landscape is moving too quickly to fix a Working Group’s deliverables now. Governance of GenAI use lies outside our scope. Ethics enters where documentation requires it, covering consent, sensitive prompts, and privacy in retained transcripts. Our technical work is limited to exemplar implementations that demonstrate the guidance, not shared infrastructure. We build deliberately on named precedents rather than proliferating standards: AI Usage Cards (Wahle et al. 2023) provide the closest existing artefact, which we will extend with the granularity of the Three Ps; domain reporting guidelines (TRIPOD-LLM, CHART, RAISE) demonstrate strong demand for process-level documentation but are each domain-bound. Our contribution is the cross-domain layer they can specialise. We extend FAIR principles, RO-Crate, and software citation standards, and will pursue EQUATOR Network registration for any successor reporting guideline.

The Global Reporting Standard for AI Disclosure in Research consortium is developing that disclosure taxonomy. We will align vocabularies and establish formal liaison so the two layers develop as interoperable standards rather than parallel ones.

Expected Impact

Documentation standards let researchers demonstrate rigour, reviewers assess whether GenAI use was appropriate, and reproduction attempts begin from a record.

Purpose and Context

The Documenting Generative AI Interactions in Research Interest Group takes stock of a fast-moving standards landscape, identifies where cross-domain documentation is missing, and extends the Three Ps framework to cover that gap. Working Groups then formalise the extended framework as a standard.

Core Problem and Approach

Current approaches centre on disclosure, which leaves a reviewer reading “ChatGPT was used for data analysis” unable to assess the use’s appropriateness or the judgement exercised.

We distinguish between disclosure (acknowledging GenAI use) and documentation (capturing methodological details sufficient for peer evaluation and reproducibility). Our central thesis is that GenAI interactions constitute research methodology and warrant documentation standards comparable to those for research software, laboratory protocols, and data management practices, and that no cross-domain standard yet specifies what that documentation must contain.

The inadequacy of disclosure-only approaches is empirically documented. The systematic review underpinning the CHART reporting guideline found that one of 137 chatbot health advice studies fully described the version of the model it evaluated; publisher instructions to authors vary widely in what disclosure they require (Ganjavi et al. 2024); and normative frameworks (the ICMJE Recommendations and the European Commission’s ERA Living Guidelines) already demand documentation that no standard yet specifies. This Interest Group exists to close the specification gap.

Our approach centres on the Three Ps framework, our own contribution and not yet published: Prompt (what input was given to the tool), Process (how the tool was used), and Provenance (tracking the interaction’s role in research). This framework provides high-level guidance while remaining adaptable across disciplines and GenAI modalities.

The relationship between documentation and disclosure is asymmetric: from a Three Ps record, a disclosure statement compliant with any current journal or funder policy can be generated automatically, while no disclosure statement, however detailed, can reconstruct the underlying interaction record. Documentation subsumes disclosure; disclosure cannot recover documentation.

Prior Work

This Interest Group builds on prior work at Macquarie University on AI research policy (Macquarie University Guidance Note, Framework Paper) and on preregistration and open data in the humanities. Two lineages ground the approach, one treating tacit research practice as requiring structured documentation, the other treating process records as research objects. We apply both to the researcher’s interaction with GenAI (full sources in the Statement of Work).

Activities and Outputs

The Interest Group will pursue four primary activities over 12 months. Landscape Analysis: inventory existing and emerging standards, guidelines, and initiatives; preliminary scanning (2026) identified more than twenty initiatives across three clusters, concentrated in disclosure, with no cross-domain specification of the record itself. Systematic analysis is the group’s first task. Framework Development: extend the Three Ps framework through iterative testing across GenAI modalities (text, image, code, audio/video generation) and research disciplines, and publish exemplar tooling that implements it, prototypes in hand. Stakeholder Engagement: structured consultations with related RDA groups, publishers, research infrastructure providers, and standards bodies. Working Group Specification: evaluate and prioritise potential Working Group outputs through community input (metadata schemas, minimum information guidelines, research grimoire frameworks, example tooling, publisher recommendations, and ethics guidelines), producing detailed Statements of Work ready for community review.

Distinctiveness and Positioning

Our work occupies one layer of a stack. Cross-domain policy from COPE, ICMJE, and the major publishers requires that GenAI use be declared; disclosure taxonomies such as the Global Reporting Standard establish what must be declared; machine-actionable encodings such as RO-Crate and W3C PROV carry the records; and domain reporting guidelines specialise requirements for particular fields. This Interest Group supplies the layer each of those presupposes, namely what researchers must capture about a GenAI interaction.

An Interest Group is the appropriate first step: the initiative has emerged from a Birds of a Feather session, and the landscape is moving too quickly to fix a Working Group’s deliverables now. Governance of GenAI use lies outside our scope. Ethics enters where documentation requires it, covering consent, sensitive prompts, and privacy in retained transcripts. Our technical work is limited to exemplar implementations that demonstrate the guidance, not shared infrastructure. We build deliberately on named precedents rather than proliferating standards: AI Usage Cards (Wahle et al. 2023) provide the closest existing artefact, which we will extend with the granularity of the Three Ps; domain reporting guidelines (TRIPOD-LLM, CHART, RAISE) demonstrate strong demand for process-level documentation but are each domain-bound. Our contribution is the cross-domain layer they can specialise. We extend FAIR principles, RO-Crate, and software citation standards, and will pursue EQUATOR Network registration for any successor reporting guideline.

The Global Reporting Standard for AI Disclosure in Research consortium is developing that disclosure taxonomy. We will align vocabularies and establish formal liaison so the two layers develop as interoperable standards rather than parallel ones.

Expected Impact

Documentation standards let researchers demonstrate rigour, reviewers assess whether GenAI use was appropriate, and reproduction attempts begin from a record.