Artificial intelligence-assisted manuscript text and visuals: Supporting scientific communication while preserving scientific integrity

The use of AI in manuscript text should therefore be considered according to the role it plays rather than simply whether an AI tool was used. Some applications primarily assist authors in communicating researcher-generated work. Examples include improving sentence structure, readability, grammar, and language expression. These applications may support scientific communication; however, as noted above, they do not replace the scientific judgement underlying the research itself.

A greater level of scrutiny is required when AI contributes to generating manuscript content. AI-assisted systems may produce text, summaries, factual statements, or references that require independent verification by authors before submission. Authors remain responsible for ensuring that information included in the manuscript is accurate, appropriately supported by evidence, and consistent with the underlying research. AI-generated references or citations require particular attention because inaccurate or unsupported references may compromise the reliability of the scientific record [1, 2].

Transparency is another central principle of responsible AI use. Authors should disclose the use of AI-assisted technologies according to the requirements of the target journal and should clearly understand the extent to which AI contributed to manuscript preparation. The use of AI does not transfer responsibility for the submitted work from authors to the technology. ICMJE emphasises that AI systems cannot be listed as authors because they cannot assume responsibility for the accuracy, integrity, and originality of published work [1]. COPE similarly highlights that accountability for published content remains with human authors and cannot be assigned to AI tools [3].

A useful way to interpret current editorial expectations is to consider whether AI is assisting the communication of existing scientific work or contributing material that becomes part of the scientific record. As AI involvement moves from language assistance toward content generation or modification of substantive meaning, the importance of verification, transparency, and author oversight increases. This role-based view, rather than the fact of AI use alone, should guide evaluation. These categories should be understood as points along a continuum rather than as mutually exclusive classifications, because even apparently assistive uses may alter meaning, emphasis, interpretation, or the reader’s perception of evidence Table 1.

A useful distinction is whether AI assists in communicating researcher-generated information or whether AI generates or modifies visual content that may influence scientific interpretation. AI-assisted preparation of explanatory materials, such as conceptual diagrams or schematic illustrations, may be treated differently from AI-generated or AI-altered images that represent scientific evidence. Researchers should therefore consider not only the use of AI itself but also the role AI played in creating the submitted visual material. In particular, AI enhancement or editing of real images (e.g., denoising, background removal, upscaling) occupies a grey zone and should be approached with extra caution and transparency.

Major publishers have developed guidance addressing the use of AI in manuscript figures and images, although the details vary across publishers [4-6]. For example, Elsevier’s guidance for authors addresses the use of generative AI and AI-assisted tools in relation to images and figures submitted with manuscripts. It restricts the use of AI tools to create or alter images that could affect the accuracy or integrity of the scientific record, while recognizing that AI may be appropriate when it forms part of the research methodology and is described transparently [4]. Similarly, Springer Nature emphasises that authors remain responsible for the accuracy and integrity of submitted material and provides guidance regarding the use of generative AI in figures and images submitted for publication [5].

The distinction between visual communication and visual modification is particularly important. AI-assisted tools may help researchers organize, format, or present information generated from their own research. However, the use of AI to add, remove, obscure, or modify features within scientific images raises different concerns because such changes may alter the interpretation of the underlying evidence. Generating or materially altering images that purport to represent actual research results—such as micrographs, histology or pathology images, Western blots, radiology or imaging scans, gel images, or photographs of specimens—is largely prohibited, because AI-generated data images can misrepresent findings that never actually occurred. By contrast, a number of journals and publishers allow AI-generated schematic diagrams, illustrations, flowcharts, and graphical abstracts that do not claim to be a record of an experimental result, provided their AI origin is clearly disclosed; however, some publishers prohibit AI-generated images without explicit editorial permission, and others apply case-by-case, risk-based scrutiny. Any AI involvement that affects the content, meaning, or reproducibility of a scientific figure requires careful evaluation and transparency. Some publisher policies distinguish between AI used to prepare manuscript visuals and AI that forms part of the research methodology. When AI is an integral component of the research method, its use should be described transparently and reproducibly as part of the study methodology rather than treated solely as manuscript preparation [4-6]. Current guidance therefore does not support a simple classification of all AI-assisted visual materials as acceptable or unacceptable. Instead, researchers should determine the function of AI within the visual workflow, distinguish between explanatory materials and research evidence, follow the requirements of the target journal or publisher, and, when in doubt, treat AI-generated data images as prohibited unless explicitly allowed.

AI will continue to influence how researchers prepare and communicate scientific manuscripts. As AI-assisted technologies become increasingly integrated into scholarly publishing, the central question is not simply whether AI was used, but what role AI played in creating the material submitted for publication.

Current editorial guidance generally emphasises that AI may support some aspects of manuscript preparation, but human authors remain responsible for the work’s accuracy and integrity. The extent of verification and disclosure should be proportionate to the role AI played in the workflow and should follow the requirements of the target journal or publisher. AI used primarily to assist the communication of researcher-generated information may present different considerations from AI used to generate or modify content that forms part of the scientific record; however, even communication-focused uses require human review for accuracy, attribution, bias, and unintended changes in meaning.

Researchers using AI in journal manuscripts should therefore understand how AI contributed to written and visual materials, verify all AI-assisted content, disclose AI use in accordance with—and, where journal requirements are unclear, consistently with—the policies of the target journal or publisher. As AI capabilities continue to change, maintaining scientific integrity will depend less on restating these principles and more on operationalizing them; through disclosure norms, verification habits, and editorial guidance that keep pace with the technology itself.

This is a single author study.