Gut's AI Policy: BMJ Rules for Gastroenterology and Hepatology Authors

Gut publishes significant research on AI-assisted endoscopy, computer-aided detection (CADe) for polyp identification, computer-aided diagnosis (CADx) for characterization, and AI for capsule endoscopy reading. If your paper is about these tools, the AI is your research subject, described in standard Methods as methodology.

The manuscript preparation AI disclosure is separate. If you developed an AI endoscopy system and also used ChatGPT to edit your paper, you need two clearly distinct descriptions:

"The CADe system described in this study was developed using a ResNet-50 architecture trained on 50,000 annotated colonoscopy frames (see Methods: Model Development). Separately, during manuscript preparation, the authors used ChatGPT (GPT-4, OpenAI) to improve the language of the Discussion section. The authors take full responsibility for the published content."

Gut microbiome papers involve extensive bioinformatics: 16S rRNA gene sequencing, shotgun metagenomics, metabolomics integration, taxonomic classification, diversity analysis. If AI tools helped with writing analysis code, disclose it.

The analysis tools themselves, QIIME2, MetaPhlAn, HUMAnN, LEfSe, are research software, not AI writing tools. They belong in standard Methods. But if ChatGPT or Copilot helped you write scripts to run these tools or process their outputs, that's AI-assisted code generation.

Disclosure example for a microbiome paper:

"Microbiome analysis was performed using QIIME2 (v2023.9) and MetaPhlAn 4 as described in Methods. GitHub Copilot (Microsoft) was used to assist with writing custom Python scripts for alpha and beta diversity calculations and for generating the LEfSe input files. All scripts were validated against published tutorial datasets. ChatGPT (GPT-4, OpenAI) was used to improve the readability of the Results section. The authors take full responsibility for the published content."

Gut publishes GI clinical trials, drug efficacy studies, endoscopic intervention trials, dietary interventions. For these papers, the same rules that apply at NEJM or The Lancet apply here: keep AI away from clinical data interpretation and outcome reporting.

Don't use AI to draft sections describing primary endpoints, adverse events, or clinical significance. Gut's clinical reviewers will scrutinize these sections closely, and AI-generated clinical language often introduces subtle inaccuracies that human experts catch.

For a clinical GI paper:

"During preparation of this manuscript, the authors used Claude (Claude 3.5, Anthropic) to improve the language and clarity of the Introduction and Discussion sections. No AI tools were used for statistical analysis, clinical data interpretation, or reporting of trial outcomes. The statistical analysis was conducted by the study biostatistician (M.H.) using Stata 17. All AI-edited text was reviewed by the clinical investigators. The authors take full responsibility for the published content."

For a basic science GI paper:

"The authors used ChatGPT (GPT-4, OpenAI) to edit the Methods section for conciseness and to improve the language of the figure legends. All revisions were reviewed by the corresponding author (P.L.). The authors take full responsibility for the content."

For an AI-in-endoscopy research paper:

"The deep learning model for polyp detection described in this paper was developed using PyTorch and trained on the institutional colonoscopy dataset (see Methods: Model Architecture). Separately, during manuscript preparation, ChatGPT (GPT-4, OpenAI) was used to improve the readability of the Discussion. The research methodology and the manuscript editing tool are entirely separate systems. All AI text suggestions were reviewed by the senior author."

What BMJ Publishing Group's policy says about inadequate declaration:

During review:

• The editor considers whether the way AI was used and declared is reasonable and consistent with BMJ's publication policies

• The editor may ask for more information, or for information to be added to the content

Before or after publication:

• Content may be rejected or be subject to post-publication changes on the basis of inadequate declaration or on the particular circumstances of its use

• BMJ may use screening tools to help identify AI-generated content

For Gut specifically: The journal has a high proportion of clinical content that can influence treatment guidelines from organizations like the American Gastroenterological Association (AGA), the British Society of Gastroenterology (BSG), and the European Association for the Study of the Liver (EASL). If an AI-generated clinical claim in a Gut paper makes it into a guideline recommendation, the downstream consequences affect patient care.

Gastroenterology (AGA) follows Elsevier's AI policy, which is broadly similar. Hepatology (AASLD) follows Wolters Kluwer guidelines. The core rules are similar across all five journals, but details such as where to disclose and how images are treated differ, so read each journal's policy.

If you're choosing between Gut and Gastroenterology, both require disclosure and both prohibit AI authorship; read each policy for the rules on images. Your decision should be based on the science, the readership, and the fit with your manuscript.

For clinical papers:

• Don't use AI to draft clinical outcome descriptions, adverse event summaries, or treatment recommendations

• If your paper includes a CONSORT or STROBE checklist, complete it yourself, AI-generated checklist responses may not accurately reflect your study

• Keep AI away from the Abstract, Gut's desk-rejection process starts with the abstract, and AI-generated clinical language can trigger concerns

For microbiome research:

• Document which bioinformatics code was AI-assisted and which was written manually

• Deposit analysis code in a public repository with clear documentation

• If AI helped design your analysis pipeline, distinguish this from the pipeline's execution

For endoscopy and imaging research:

• Clearly separate your AI research method from your AI writing disclosure

• Include model training details, validation datasets, and performance metrics in standard Methods

• The ASGE guidelines on AI reporting in endoscopy research may provide additional framework for your Methods section

Before submission checklist:

• [ ] AI use acknowledged in the contributor section (fuller description in Methods for research use) with tool name, reason, and task

• [ ] Research AI tools described in the Methods, separate from your writing-assistance disclosure

• [ ] No patient/clinical data processed through cloud AI tools

• [ ] Clinical interpretations are human-generated

• [ ] Microbiome code deposited and validated

• [ ] Any AI question in the submission system answered honestly

• [ ] All co-authors aware of AI disclosure

A Gut submission readiness check can help verify your Gut submission meets the journal's requirements before you enter the competitive review process.

Comply proactively if:

• You used any AI tool (ChatGPT, Grammarly, Copilot) during manuscript preparation

• BMJ requires AI use to be declared in the contributor section (and in the Methods for research use)

• Your institution has its own AI use policy that may be stricter

Less concerned if:

• You used only basic grammar or spell checking (BMJ's policy publishes no list of exempt tools, so declare if in doubt)