| Risk level | What this means | Typical use in research
practice | Key considerations | Declaration |
Green
(low risk) | AI supports
efficiency and
exploration without
influencing core
research decisions | Literature discovery, semantic
search, summarising public
literature; data cleaning,
deduplication, quality checks;
code scaffolding; project planning | Outputs should be reviewed,
verified and understood; avoid
over-reliance | Describe tool and
purpose and take
accountability |
Amber
(requires
caution and
care from
authors) | AI contributes to
analytical or
interpretive
processes | Identifying patterns in datasets;
supporting statistical modelling;
qualitative coding; generating
explanatory summaries or
comparisons | Risks include bias, lack of
transparency, reduced
reproducibility, loss of nuance;
human validation is essential | Provide clear
methodological
description and
validation steps and take
accountability |
Red
(not permitted) | AI replaces core
scholarly
judgement or
compromises
integrity | Generating hypotheses, analyses,
or conclusions without human
oversight; fabricating data or
references; using sensitive or
restricted data in public tools | Undermines accountability,
reliability, and ethical standards | Not permitted |
What this means in practice
AI can support many aspects of research, but its use should be managed with increasing levels of care proportionate to the level of risk involved.
In lower-risk contexts, such as literature discovery or data preparation, AI can improve efficiency without shaping the intellectual contribution of the work.
In higher-risk contexts, such as data analysis, modelling, or qualitative interpretation, AI may assist but must not replace critical reasoning. Authors should be able to clearly describe how AI was used, what inputs and outputs were involved and how results were validated.
Across disciplines, some common considerations apply:
- Quantitative research: ensure analytical methods remain transparent and reproducible. AI may be used to explore large datasets or support modelling. When doing so, authors should be able to clearly describe inputs, outputs and validation steps, ensuring results are reproducible and independently assessable.
- Qualitative research: preserve nuance and demonstrate human-led argument and interpretation. AI tools can assist with coding or theme identification, but argument and interpretation must remain human-led.
- Social or online data: reflect on bias, context, and ethical considerations. AI tools could be used to assist classification of data or sentiment analysis; however, this should be accompanied by critical human reflection and a strong adherence to privacy and consent expectations from participants and the wider research community.
- Evidence synthesis: ensure completeness, accuracy, and verifiability. AI may be used to manage large volumes of data or evidence synthesis but risks omitting key studies, oversimplification or fabricating content. Authors must clearly document their methodology and design.
Transparency and declaration
We recognise that authors may sometimes feel uncertain about declaring AI use, particularly where there are concerns about how this might affect editorial or peer review outcomes.
Our policy is designed to ensure the opposite.
- Declaration does not disadvantage your manuscript
- It supports fair, consistent, and efficient evaluation
- It helps keep attention focused on the quality and contribution of your research
Where AI use is unclear or undeclared, editors may need to seek clarification, which can delay assessment or shift focus away from the research itself.
Clear and transparent declaration supports trust, reproducibility, and the integrity of the research record.
Key principle
The acceptability of AI use in research depends not on the tool itself, but on whether its use supports transparency, rigour, and the integrity of the research process and outcomes.
