hese are explicitly forbidden by publishers. Violations can lead to retractions, damaged reputations, and institutional investigations.

1. Pasting Unpublished Manuscripts Into Public AI Tools

Why it’s forbidden: Your unpublished data, methods, and findings become part of the AI provider’s system. Elsevier explicitly states: “Authors may not upload previously-published or copyrighted content into external, open AI tools.”

What to do instead: Use local AI tools (GPT4All, Ollama) where data stays on your machine. If using cloud tools, only paste isolated paragraphs you’ve already written — never full manuscripts.

2. Using AI-Generated Text Verbatim

Why it’s forbidden: This is a form of plagiarism. You did not write it, and you cannot verify its provenance. AI models are trained on copyrighted material.

Elsevier: “AI output can only serve as inspiration… the manuscript needs to represent the authors’ authentic and original contribution and reflect the authors’ own analysis, interpretation, insights and ideas.”

3. Listing AI as a Co-Author

Why it’s forbidden: Every major publisher and the ICMJE agree: authorship requires accountability. AI cannot:

• Approve the final version
• Agree to be accountable for the work
• Respond to post-publication critiques

If you used AI, disclose it. Don’t list it as an author.

4. AI-Generated Figures or Images

Why it’s forbidden: Nature, Elsevier, and Science all prohibit generative AI images in manuscripts. Concerns include:

• Copyright (AI training data includes copyrighted images)
• Research integrity (AI can fabricate data-like visuals)
• The infamous “mouse penis” paper retraction started with AI-generated figures

Exceptions: Non-generative ML tools used for image analysis (e.g., cell segmentation, automated counting) are fine — just describe them in Methods. Nature allows generative AI images only if the paper is specifically about AI, and even then on a case-by-case basis.

5. AI-Generated References

Why it’s forbidden: LLMs hallucinate references. They invent plausible-looking authors, titles, journals, and DOIs that do not exist. Unlike Elicit or SciSpace, general-purpose chatbots are NOT connected to verified citation databases.

Distinction: Using AI to format a reference list (e.g., “reformat these references into Nature style”) is fine. Asking AI to generate references is not.

6. Uploading Manuscripts for Peer Review

Why it’s forbidden: All four major publishers explicitly forbid reviewers from uploading manuscripts to AI tools. This breaches confidentiality and may violate data protection laws.

7. Patient or Participant Data in AI Tools

Why it’s forbidden: GDPR and ethical research standards. Once data enters an AI system, you lose control over where it goes and how it’s used.

The Disclosure Template

When you use AI for anything beyond basic grammar checks, include a declaration. Elsevier’s template is widely accepted:

During the preparation of this work the author(s) used [NAME OF TOOL]
in order to [REASON]. After using this tool/service, the author(s)
reviewed and edited the content as needed and take(s) full
responsibility for the content of the publication.

Where to place it: End of the manuscript, immediately above the references, under the heading: “Declaration of Generative AI and AI-assisted Technologies in the Writing Process.”

Examples of specific disclosures:

• “During the preparation of this work the authors used Claude (Anthropic) to improve the clarity and conciseness of the Discussion section. After using this tool, the authors reviewed and edited the content as needed and take full responsibility for the content of the publication.”

• “During the preparation of this work the authors used ChatGPT (OpenAI) to generate alternative phrasings of the Methods section to avoid text recycling from a prior publication. After using this tool, the authors reviewed and edited the content and take full responsibility for the publication.”

Worked Example — Good vs. Bad Use

Scenario: A researcher at NBI has qPCR data showing gene RumC is upregulated 3.2-fold under mucin treatment in Ruminococcus gnavus (p = 0.003, n = 6 biological replicates).

Bad Use

Prompt: “Write a results paragraph about this qPCR data showing RumC is upregulated 3.2-fold in R. gnavus under mucin treatment.”

AI output (what could go wrong):

“Quantitative PCR analysis demonstrated a significant 3.2-fold upregulation of RumC in Ruminococcus gnavus ATCC 29149 when cultured on mucin as a sole carbon source (p < 0.01), consistent with previous reports of mucin-responsive gene regulation (Smith et al., 2019).”

Problems:

• AI changed the p-value from 0.003 to < 0.01 — fabrication
• AI invented the growth condition details (“sole carbon source”) — unsupported
• “Smith et al., 2019” may not exist — hallucinated reference

Good Use

Prompt: “Make this sentence more concise without changing any data or meaning: ‘Quantitative reverse-transcription PCR analysis demonstrated that the RumC gene exhibited a statistically significant upregulation of 3.2-fold in Ruminococcus gnavus cells grown in the presence of mucin compared to the glucose control condition (p = 0.003, Student’s t-test, n = 6 biological replicates).’”

AI output (after researcher wrote the original):

“The RumC gene was upregulated 3.2-fold in R. gnavus during growth on mucin compared to glucose (p = 0.003, Student’s t-test, n = 6).”

The researcher wrote the content. AI only trimmed it. All data intact.

Which Tool for Which Writing Task?

Different AI tools excel at different writing tasks. Here is a practical guide:

Read more about literature-specific tools and local AI for sensitive data.

Summary: The Ethical Scientific Writer’s Checklist

Before submitting any manuscript where you used AI:

• I wrote the core intellectual content myself
• I verified every fact, number, and reference AI helped with
• I checked the terms of service of every AI tool I used
• I never uploaded confidential, unpublished, or patient data
• I never asked AI to interpret my results or draw conclusions
• I included a disclosure statement where appropriate
• I can defend every sentence in this manuscript

Remember: The goal is not to avoid AI. The goal is to use it as a tool that amplifies your expertise — not replaces it. When in doubt, disclose more rather than less.

Previous submodule:

Caution