Your Journal's AI Policy Might Not Mean What You Think

On January 1, 2026, Science, the flagship journal of the American Association for the Advancement of Science, banned AI-generated text outright. The policy is unambiguous:

"Text generated by AI, large language models, or similar algorithmic tools may not be used in papers published in Science journals."

In marked contrast, Nature permits AI-generated text with disclosure. It makes a distinction that many researchers have missed: AI used for copy editing to improve grammar, readability, and language quality of human-written text does not require disclosure at all.

These are the two most important journals in the world, and their policies are fundamentally incompatible.

Here is what you need to know as of early 2026.

The spectrum of journal AI policies falls into four categories:

1. Full ban (Science/AAAS). No AI-generated text in the manuscript. AI tools for data analysis, coding, or figure generation are evaluated case by case but must be fully disclosed, including the prompts used.

2. Permitted with mandatory disclosure (Elsevier, Wiley, Taylor & Francis, SAGE, most major publishers). You may use AI to draft or revise text, but you must declare it. The ICMJE — whose guidelines most medical journals follow — requires that AI use be described in the Methods section. AI cannot be listed as an author.

3. Permitted with conditional disclosure (Nature/Springer Nature). This is where it gets interesting. Nature distinguishes between:

• Generative use (AI drafted or substantially generated the text) → disclosure required
• Editing use (AI was used to improve language, grammar, and readability of human-written text) → no disclosure required

This distinction is critical for non-native English-speaking researchers: if you write your manuscript yourself and then use AI to polish the English, Nature does not consider this a disclosure event. The logic is that this is functionally equivalent to hiring a human copy editor, and journals have never required disclosure of editing services.

4. No stated policy. Many smaller and regional journals have not yet addressed AI. The absence of a policy is not permission. It is a gap that will likely be filled, potentially retroactively.

What this means in practice:

The Nature distinction between generative and editing use is, in my view, the most important development in journal AI policy. It creates a safe harbor for the exact use case that matters most to our clients: a researcher who writes their own manuscript based on their own data and interpretation, and then uses tools — whether human editors, AI, or both — to ensure the English is correct and the prose is clear.

This is fundamentally different from asking ChatGPT to "write me an Introduction about the effect of X on Y." That is generative use, and every major publisher requires disclosure at minimum.

The challenge is that the line between editing and generation is not always sharp. If you paste your poorly structured paragraph into ChatGPT and ask it to "rewrite this to be clearer," is the output edited text or generated text? The honest answer is that it depends on how much the AI changed — and most researchers are not tracking this carefully.

Our advice to authors:

1. Check your target journal's specific policy before submission. Do not assume all journals follow the same rules. Science and Nature — journals that sit side by side on every university library shelf — have irreconcilable policies.

2. Distinguish between editing and generation in your own workflow. Writing a sentence yourself and asking AI to fix the grammar is editing. Asking AI to write the sentence is generation. Know which you are doing.

3. When in doubt, disclose. Overdisclosure carries no penalty. Underdisclosure carries severe consequences.

4. Document your AI interactions. The ICMJE recommends describing the AI tool used, the version, and the nature of the interaction. This costs you nothing and gives you a documented defense.

5. Remember: you are the author. No journal holds the AI accountable. Every claim, every data point, every citation in your manuscript is your responsibility, regardless of how it got there.

The policy landscape will continue to evolve. Some journals might tighten restrictions while others will formalize the editing/generation distinction that Nature has pioneered. What will not change is the fundamental principle: the author is responsible for the science in their paper. Tools change. That obligation does not.