English Is Not Neutral: Implications for Evaluating Artificial Intelligence in Medical Writing

Based on my decades-long experience in medical writing, I have come to believe that the use of artificial intelligence (AI) may unintentionally enlarge the imbalance between native and non-native English speakers in academic publishing. For many years, I have also written non-medical essays in Japanese, my native language, for a small-circulation journal. I always ask my wife to read them. She never gives me a ‘free pass’: she always points out linguistic issues. Her suggestions are almost always correct from a grammatical or stylistic standpoint. However, whether I adopt them depends on context, rhythm, and rhetorical intent. Sometimes I accept her advice; sometimes I deliberately do not. Looking back, the suggestions I chose to incorporate have consistently made the essays more appealing. This decision-making process is largely instinctive and reflects accumulated judgment. It does not require deliberate analysis. It may not even depend on intellectual ability or educational level. When working in one's native language, one can immediately sense whether an external suggestion improves the text while preserving one's own tone and touch. This ability seems to grow naturally within a linguistic environment. When it comes to English, however, the situation is altogether different. Long before the emergence of AI, I had already felt that commercial English editing often erased my tone. Sentences became smoother and more correct, yet something essential disappeared. This was not an AI-related phenomenon; it existed well before. What AI has done is not to create this problem, but to make it clearer, faster, and far more widespread. The core issue is that English is not a neutral tool in scientific writing. Medicine has long treated English as if it were equally accessible to everyone, but this assumption is evidently false. English favors certain people: those who have lived in an English-speaking environment and have developed a natural sense of rhythm and nuance in the language. Many native speakers, and a small number of exceptional non-native speakers, fulfil these conditions. These writers can use AI as much as I use my wife's advice when writing in Japanese: they can judge what to keep and what to ignore. They retain authorship judgment. Their tone survives. For most non-native writers, this is far more difficult. Many cannot confidently judge whether their original expression is better than the AI's suggestion. Some do not even suspect that their original phrasing might carry nuance worth preserving. They carefully check factual accuracy after AI editing, but they cannot easily determine which version better expresses their intended tone. As a result, AI-edited text is often accepted automatically. Thus, AI does not affect all writers equally. Native speakers tend to use AI to clarify their writing while preserving their voice. Non-native speakers often use AI in a fundamentally different way: their voice is replaced by a standardized, fluent, but impersonal tone. In many cases, they do not even notice that this has happened. This difference is unfair, yet largely invisible. Editors do not see it. Reviewers do not see it. Readers do not see it. Even the writers themselves often do not see it. Consequently, the distinctive voices of many experienced non-native researchers may disappear quietly from the literature. One of the strongest arguments in favor of AI-assisted writing is that it ‘reduces language barriers’ 1. This claim is partly true. AI can indeed help writers who are inexperienced in English and who might otherwise struggle to write at all 1. I am not referring to those cases. The problem lies elsewhere. AI may not benefit non-native authors who have been publishing in reputable journals for decades. These writers have developed their own style through prolonged effort and struggle. However, they have not necessarily acquired the instinctive ability—common among native speakers—to decide which expressions best preserve nuance and tone. With rare exceptions, this ability seems to arise more from linguistic immersion than from education or determination alone. To illustrate this, I recently asked ChatGPT-5 to edit one of my Japanese essays 2. The result was grammatically complete and logically sound. Yet it had lost the nuance and rhythm cultivated through years of writing. Any Japanese native writer would immediately choose to return to the original. In contrast, when non-native writers encounter similarly ‘peculiar’ or ‘foreign’ English expressions suggested by AI, they may accept them—and many of us have already submitted such manuscripts. This situation existed with commercial editing, but AI has expanded it rapidly and on a much larger scale. Despite affecting thousands of researchers, it is rarely discussed. At present, I do not have a clear solution. However, if ‘reducing language barriers’ is to remain one of the central justifications for AI use in academic writing, progress should not be evaluated solely by grammatical correctness or fluency. Greater attention should be paid to preserving individual tone and authorship judgment. More fundamentally, the medical community may need to acknowledge and critically reflect on the heterogeneity of English itself. There is Japanese English, French English, and many others. Linguistic heterogeneity, like other forms of diversity, may also contribute, quietly but meaningfully, to scientific progress. The author identified the significance and wrote the manuscript. ChatGPT-5 was used for grammar checking. The author has nothing to report. This editorial does not include research involving human participants, animals, or identifiable personal data and therefore did not require ethical approval. The author has nothing to report. The author declares no conflicts of interest.