On the Use of Artificial Intelligence in Medical Publishing

The rapid expansion and uptake of generative artificial intelligence (AI), including large language models (LLMs), in science and medicine hold important implications for scientific communication in the peer-reviewed literature. Most medical journals now include general statements in their “instructions for authors” about the permissible use of AI technology, focusing mainly on transparency and the need for authors to disclose any use of AI^[1] in their preparation of submitted manuscripts. At The Journal of Clinical Psychiatry, we believe that a more nuanced discussion is needed of both the potential benefits and potential hazards associated with AI use in medical publishing. We have framed this editorial around our guiding principle that the articles we publish aim to impact and improve patient care.

Use of AI in Manuscript Preparation

While an author’s disclosure of using AI in the preparation of a scientific manuscript is important for the sake of transparency, that acknowledgment alone fails to address more fundamental concerns about scholarly communication, intellectual property ownership, and the preservation of scientific integrity. At one end of the spectrum, authors might draw upon AI tools in ways analogous to seeking help from an expert librarian, editor, thesaurus, medical writer, or even a fastidious research assistant. AI platforms can perform literature searches (the accuracy of which remains the authors’ responsibility) and produce summaries to inform authors about a particular topic. AI platforms can also “polish” human-generated language, grammar, and overall prose; help reduce a manuscript’s word count; and translate text written by the authors from one language to another—a function especially valuable to medical scientists writing in a non-native language.

At the other extreme, authors might rely on AI to generate an entire or substantial part of a manuscript’s Introduction or Discussion section (ie, using AI to interpret existing literature or key findings) or even produce an entire review article de novo, with little more than an authorial prompt. Intermediate AI use might involve organizing or restructuring conceptual content to ensure the clarity of intended meanings and improve the coherence of presentation quality, as well as identifying overarching themes or unifying concepts that might help create summary tables or graphical representations. Discerning the extent to which AI performs critical thinking—a task that, in our view, rightly belongs to human authors—is increasingly difficult.

Scientific Integrity, Attribution, and Accountability

Using AI to deliberately fabricate data or automate the interpretation of study findings crosses ethical lines and undermines scientific integrity. It is equivalent to the generation of fraudulent communication in the pre-AI era. Similar ethical transgressions arise from using AI to essentially write a manuscript. The federal Office of Research Integrity identifies fabrication, falsification, or plagiarism as research misconduct (42 C.F.R. § 93.103). Motive might seem relevant: Did an investigator submit inaccurate AI-generated content through carelessness or sloppiness rather than deliberate intent to deceive? However, motive at best minimally mitigates the impact of de novo generation of data or the interpretation of research findings. Plagiarism is the fraudulent presentation of material created by another source as if it were one’s own, regardless of the intent.

Similarly, we (and probably most editors) would frown on authors incorporating verbatim swaths of AI-generated verbiage. However, barring the wholesale use of AI to generate content, questions linger about what constitutes “AI-assisted” versus “AI-generated” scholarly communication.

Consider the literal meaning of “authorship,” from the Latin auctor, meaning “creator, originator, or master.” (The Greek corollary is either ἀρχηγός, meaning “pioneer,” “founder,” or “leader,” or συγγραϕεύς, one who “writes together” or “compiles.”) How do those terms apply to manuscript credit, accountability, and attribution for actual written words, arguments, and phrasing or syntactic style? It is far different to use AI to ensure proper English terminology and grammar when translating a mature draft from another language than to have AI generate the draft itself. The World Association of Medical Editors has provided guidance regarding the use of “chatbots” such as ChatGPT, explicitly forbidding chatbot authorship.¹ The Committee on Publication Ethics also issued a policy statement banning LLM inclusion as authors on scientific reports (https://publicationethics.org/guidance/cope-position/authorship-and-ai-tools). AI should not have that level of “authorial” participation. Indeed, AI platforms cannot satisfy the International Committee of Medical Journal Editors guidelines requiring manuscript authors to give their approval of a final version of a manuscript and agree to be accountable for all aspects of the submitted work.

Detecting AI Use

A major challenge to journal editors is that AI detection is imperfect. It can be uncomfortable to question an author’s use of AI or imply misconduct, particularly because some publicly available AI detection tools have been shown to incorrectly classify human-written content as AI-generated.² One study of free AI detection software within the public domain flagged 27.2% of human-generated published academic text as AI-generated.³ Some proprietary commercial AI-scanning software packages may provide a more accurate quantitative estimated percentage of likely human-versus AI-generated content, but established thresholds or standards for judging AI “acceptability” are lacking—other than that 100% AI-generated content seems irreconcilably problematic. At the same time, despite some programs’ purported low false-positive rates for detecting AI-generated material,⁴ writing styles that are formulaic or mimic “humanized” AI phrasings (eg, highly structured sentences and clauses with flourish, as described below) can trigger false positives.

The “Language” of AI

Another important consideration involves what might be called “AI-speak,” or characteristic telltale AI phrasings. If, for illustrative purposes, we momentarily violate our own principles and ask Google Gemini to generate examples of this phenomenon, we learn that it involves “overly dramatic, polished, or predictable buzzwords generated by large language models.” Key examples include hollow descriptors such as “holistic” or “ecological,” overused filler verbs like “delve,” ostentatious adjectives like “paradigmatic” or “meticulous,” and vague transitional phrases like “evolving landscape.” AI-speak overuses sentence construction of rhythmic cadence-matching and suspense-building formulaic contrasts (eg, “The problem is not simply X. Nor is it Y. Rather, it is a more insidious and far-reaching Z”) or plays off abstract binary oppositions that may, to the trained psychiatric ear, ring of poverty of content (eg, “Rather than functioning as a static phenomenon, the system acts as a dynamic mechanism”).

Another AI source (www.stryng.io) identifies the “rule of balanced triads” (eg, “The proposed framework provides scalability, adaptability, and robustness”) and transitional and adverbial padding (eg, “Crucially, this observation underscores the pivotal role of X”). Although this eases detection of AI-generated text, future iterations of AI-generated text will likely more closely mimic the dispassionate text that is traditionally used in medical writing.⁵ Moreover, the scientific literature risks sounding homogenized when authors relinquish the uniqueness of their individual voices for AI-speak.

AI as a Threat to Human Creativity and Research Expertise

While some authors, journal editors, reviewers, and other stakeholders of scientific scholarship may concede that AI is “here to stay,” we wish to raise awareness of scientific and moral imperatives surrounding the responsible use of AI technologies. The standards by which investigators communicate their ideas, articulate their hypotheses, generate empirical observations to support or refute their speculations, and formulate cogent narratives cannot and must not change as a result of incorporating a new tool to facilitate scholarly writing. One might argue that the collective scientific and publishing communities share an ethical obligation to assure research honesty and the highest standards of scientific communication, irrespective of evolving technologies and consistent with the expectations of rigor, transparency, and integrity that have been hallmark features since the centuries-old dawn of the scientific method.

The Journal of Clinical Psychiatry aims to improve clinical care and advance clinical research. We believe that the nuance which human experience and reason bring to bear is essential, if imperfect. The Journal wants its authors to think critically and creatively. Authors should experience and present their original thought processes, and not relegate those tasks to a machine. Researchers who rely on AI for writing risk stunting the development of research ingenuity and good practices. This issue transcends authorial credit by raising broader questions about researcher literacy, creativity, and intellectual development in psychiatry and ensuring that their “natural” intelligence is reflected in deep knowledge and ownership of their science.

Acknowledging that AI will continue to improve and that the relationships between AI and the scientific and publishing communities will evolve, the current positions of our editorial team about the use of AI include the following:

1. The Journal of Clinical Psychiatry favors minimal AI use in scientific writing. We strongly believe that papers written by human researchers with minimal to no AI assistance are ideal. That said, we hope to develop parameters to meet the field where it is and where it may evolve. Transparency is a crucial but minimal expectation. We require that authors disclose any and all use of AI, detailing where and how it was used in the production of a manuscript, including the purpose and extent of AI utilized. Mandatory disclosure includes use that authors might consider minimal, such as editing language and translation services, and extending to its use in formatting and generating content—which we strongly discourage and insist be specifically reported.
2. AI is not permitted in any way in peer review. Manuscripts received and under our consideration have our promise of confidentiality. AI engines routinely capture all text provided and add it to their database. Hence, AI use in reviewing is prohibited. Moreover, we expect equally original, critical thought from reviewers as we do from authors.
3. As a medical, clinically focused journal, our Hippocratic lens is to “do no harm.” We will evaluate the use of AI not only to ensure scientific integrity but also because we appreciate the distinctly human quality of clinical nuance and judgment. AI bots are not clinicians.
4. While recognizing that AI has legitimate applications in research, we are concerned that it may also be used to generate potentially spurious or health-harming associations from publicly available datasets. Such misuse could undermine scientific integrity by circumventing established scientific principles and the discipline of hypothesis-driven research. The Journal seeks meaningful work rather than volume. As we consider how authors should best document precise use of AI, we are also screening for its use as part of the review process. We insist that authors disclose all such uses.
5. We expect all authors to stand behind the quality and integrity of their submitted works and that these works be original, not manufactured. This pertains to all elements of a paper, including its conception, appropriateness, strength of recommendations, clinical relevance, limitations, conclusions, and accuracy of references.
6. Our concerns about the use of AI in medical writing are broad and extend to its philosophical impact. In an era in which commitment to high-quality research and peer review remains central, AI potentially threatens to replace original thinking and research sophistication, stunting iterative advances and dulling our literature. We will be closely monitoring how AI is used in our field and how it impacts patient safety, and we will adapt these principles accordingly.

We are entrusted to bring vital, clinically relevant papers to the public. The Journal of Psychiatry is grateful for its place in our field and to the many authors who devote considerable time, energy, originality, and ingenuity to producing their best possible scholarly and scientific work. As we continue to evaluate the use of AI in scientific writing, we must reaffirm that the scientific process ultimately depends on human judgment, critical thinking, and intellectual effort. We expect that writers and peer reviewers will keep the Journal’s audience in mind as the ultimate beneficiaries of information we publish. Our responsibility transcends the publication process: We call upon writers and reviewers to share that responsibility as members of the scientific and clinical community.

Article Information

Published Online: September 30, 2026. https://doi.org/10.4088/JCP.26ed16673
© 2026 Physicians Postgraduate Press, Inc.
To Cite: Goldberg JF, Freeman MP, Markowitz JC, et al. On the use of artificial intelligence in medical publishing. J Clin Psychiatry 2026;87(4):26ed16673.
Author Affiliations: Department of Psychiatry, Icahn School of Medicine at Mount Sinai, New York, New York (Goldberg); Department of Psychiatry, Harvard Medical School, Ammon-Pinizzotto Center for Women’s Mental Health at Massachusetts General Hospital, Boston, Massachusetts (Freeman, Vanderkruik); Department of Psychiatry, Harvard Medical School, Boston, Massachusetts (Freeman, Shinn, Vanderkruik); Division of Clinical Therapeutics, New York State Psychiatric Institute, New York, New York (Markowitz); Department of Psychiatry, Vagelos College of Physicians and Surgeons, Columbia University, New York (Markowitz); Psychotic Disorders Division, McLean Hospital, Belmont, Massachusetts (Shinn); Department of Psychiatry, College of Medicine – Tucson, University of Arizona, Tucson, Arizona (Karp); Department of Psychiatry and Behavioral Sciences, New York Medical College, Valhalla, New York (Citrome); Department of Psychiatry and Behavioral Sciences, Northwestern Feinberg School of Medicine, Chicago, Illinois (Citrome); Department of Psychiatry and Behavioral Sciences, University of Texas Medical Branch, Galveston, Texas (Wagner); Department of Psychiatry, Yale University School of Medicine, New Haven, Connecticut (Wilkinson).
Corresponding Author: Joseph F. Goldberg, MD, 128 East Ave, Norwalk, CT 06851 ([email protected]).
Financial Disclosure: Dr Goldberg has been a consultant to Abbvie, Alkermes, Almatica, Alvogen, Compass Pathways, Genomind, Neurelis, Neuroma, Otsuka, Puretech, and Seaport Therapeutics; has received royalties from American Psychiatric Publishing, and Cambridge University Press; and has been on speakers bureaus for Alkermes, Axsome, Bristol Myers Squibb, Intracellular Therapies, Johnson & Johnson, and Vanda Pharmaceuticals. Dr Freeman reports the following financial relationships within the past 36 months (some are not current): As an employee of Massachusetts General Hospital (MGH), Dr Freeman works with the MGH CTNI and MGH Center for Women’s Mental Health, which has had research funding from multiple pharmaceutical companies and National Institute of Mental Health (NIMH). MGH National Pregnancy Registry: Current Sponsors: Alkermes, Inc. (2016-Present); Bristol-Myers Squibb Company (2025-Present); Eisai Inc. (2022-Present); Otsuka America Pharmaceutical, Inc. (2008-Present); Sage Therapeutics (2019-2023, 2024-Present); Supernus Pharmaceuticals (2021-Present). Past Sponsors: Forest/Actavis/Allergan (2016-2018, declined to sponsor: 2018-Present); Aurobindo Pharma (2020-2022, declined to sponsor: 2022-Present); AstraZeneca Pharmaceuticals (2009-2014, declined to sponsor: 2014-Present); AuroMedics Pharma LLC (2021-2022, declined to sponsor: 2022-Present); Johnson & Johnson/Janssen Pharmaceuticals, Inc (2019-2023, declined to sponsor: 2024-Present); Ortho-McNeil-Janssen Pharmaceuticals, Inc (2009-2014, declined to sponsor: 2015-Present); Pfizer, Inc. (2009-2011, declined to sponsor: 2012-Present); Sunovion Pharmaceuticals, Inc. (2011-2023, declined to sponsor: 2024-Present); Teva Pharmaceutical Industries Ltd. (2018-2025; declined to sponsor 2025-Present); Dr. Reddy’s Laboratories, Inc. (2023-2025, declined to sponsor 2025-Present). Updated sponsors can be found at https://womensmentalhealth.org/research/pregnancyregistry/. Research through MGH: Sage, National Institute on Aging, NIMH. Consulting through CTNI through MGH (included below). Advisory boards or consulting, Data Safety Committees/Independent Data Safety and Monitoring Committees: Janssen (Johnson & Johnson), Novartis, Neurocrine; Eliem, Sage; Brainify; Everly Health; Tibi Health; Relmada; Beckley Psytech; Brii Biotech; Reunion; Vistagen; Mycomedica. Educational activities (speaking, planning): MGH Psych Academy, WebMD, Medscape, Pri-Med, Postpartum Support International, PRIME, HMP Global, CME Institute. Dr Shinn participated in an invited presentation to educate patient advocacy staff at Bristol Myers Squibb (BMS) about schizophrenia (Aug-Sep 2025); has been a member of an independent Data Monitoring Committee (DMC) for a BMS phase 3 RCT of adjunctive Cobenfy in bipolar disorder (since Sep 2025); has been a psychiatry consultant/member, CVS Caremark national Pharmacy & Therapeutics (P&T) Committee (since Aug 2025); and was an invited speaker for Pri-Med CME Institute one-time speaking activity on June 21, 2024. Dr Karp reports the following activities within the past 3 years: sponsored research support from Johnson & Johnson Neuroscience (formerly Janssen), scientific advising for Johnson & Johnson Neuroscience, contributing to disease state educational activities (with no input from the company) about depression from Johnson & Johnson Neuroscience, investigator-initiated trial with financial support and medication supplies from Intracellular Pharmaceuticals (recently acquired by Johnson & Johnson Neuroscience), and one-time scientific advising for Otsuka. Dr Citrome has been a consultant for Abbvie, Acadia, Adheretech, Altus, Alumis, Axsome, Alkermes, Arc, Auritec, Autobahn, Avant Healthcare Solutions, Biogen, BioXcel, BMS/Karuna, Boehringer Ingelheim, Cadent, Cerevel, Clario/Medavante/Prophase, Clinilabs, Compass, Corcept, Definium, Delpor, Eisai, Enteris BioPharma, Health Wellness Partners, HLS, Idorsia, Inmune Bio, Intra-Cellular, Johnson & Johnson/Janssen, Little Bear, Lundbeck, Luye, Lyndra, Maplight, Marvin, Mindmed, Neurelis, Neushen, Neumora, Neurocrine, Noema, Novartis, Noven, Orexo, Otsuka, Ovid, Pontifax/Draig, Praxis, PSL, Real Chemistry, Relmada, Renew Research, Response, Reviva, Sage, Seaport, Sumitomo/Sunovion, Supernus, Teva, University of Arizona, Vanda, and Wells-Fargo and one-off ad hoc consulting for individuals/entities conducting marketing, commercial, or scientific scoping research; has received honoraria for lectures from Abbvie, Acadia, Alkermes, BMS, Eisai, Idorsia, Intra-Cellular, Johnson & Johnson/Janssen, Lundbeck, Luye, Neopharm, Neurocrine, Noven, Otsuka, Recordati, Takeda, Teva, Vanda, and CME activities organized by medical education companies such as Decera Clinical Education/Clinical Education Alliance/Clinical Care Options, CME Institute, CMEology, HMP/Psych Congress, Medscape/WebMD, MultiMedia Medical LLC, Neuroscience Education Institute, NEI, Paradigm, Real Psychiatry/Efficient, Real World; owns health-related stocks as follows: portfolio of small number of shares of common stock of multiple companies and managed externally, stock options: Reviva; and has received royalties/publishing income from Taylor & Francis (Editor-in-Chief, Current Medical Research and Opinion, 2022-date), UpToDate (reviewer), Springer Healthcare (book), Elsevier (Topic Editor, Psychiatry, Clinical Therapeutics, through Spring 2025). Dr Wagner has received research support from the State of Texas and the National Institute on Alcohol Abuse and Alcoholism. Dr Wilkinson has received contract research funding from Oui Therapeutics and Neurocrine Biosciences for the conduct of clinical trials (administered through Yale University) and has consulted or provided DSMB services for Neumora, Sooma Medical, Alumis, AtaiBeckley, Osmind, Indegene, Joyous, LivaNova, MindMed/Definium, Moonlake Immunotherapeutics, and Epic Health Systems. Dr Vanderkruik has received research support from The J. Willard and Alice S. Marriott Foundation and National Eating Disorders Association and has done advisory/consulting work for the World Health Organization. Dr Markowitz has nothing to disclose.
Funding/Support: None.