AI disclosures are statements that inform readers that the author(s) had assistance from Large Language Models (LLMs) or AI tools in preparing a document, course materials, or other educational content. AI disclosure statements are becoming common as individuals and businesses are increasingly AI tools to generate content. Many states require notices to inform consumers when they interact with AI. Educational institutions have developed policies that require students to disclose if AI was used to complete assignments. Academic journals and trade publications require authors to disclose AI assistance before publication. Going beyond frontier LLMs such as ChatGPT, Gemini, Claude, and CoPilot, authors also use other AI writing assistants such as Quillbot and Grammarly. However, many authors are not aware that these language, grammar, and style helper tools use LLMs to generate AI content.
The purpose of AI disclosure statements is to clarify and demonstrate transparency about how the authors used AI, which can help build trust that the content is AI-assisted rather than AI-generated. This type of disclosure will help users make informed decisions about the level of effort and research conducted by the authors in preparing the manuscript, report, course material, or assignment. By presenting an AI Disclosure, the authors show ethical and responsible use of AI.
Since there are no guidelines for writing AI disclosures, many such disclosures, unfortunately, fail to convey a clear understanding of how AI was used in the writing process. AI systems are very complex and have parameters such as bias and temperature that can be adjusted to produce output ranging from factual and accurate to creative and hallucinated (made up by AI).
Generic AI disclosure can overwhelm users and lead to “disclosure fatigue.” Let’s take the following unedited AI Disclosure found at the end of a published article:
“The author used a generative AI assistant (Claude) to help draft and refine this article. The argument, structure, revisions, and final wording are the author’s own, and the author takes full responsibility for its content”.
There are several issues with generic statements such as the one above, as it does not offer clarity, consistency, or accountability. It appears to be written more as a preemptive defense rather than a neutral disclosure that provides nuanced information on AI use. For example, “AI assistant” contradicts the “… author’s own” wording of the claim. AI assistants can easily draft and refine words, phrases, and sentences, which would imply meaningful involvement in generating AI language that potentially shaped authors’ ideas, so how could “argument, structure, revisions, and final wording” be the author’s original work? This makes authorship unclear and complicates attribution. Similarly, “Help draft and refine” is a skill that LLMs can easily perform, ranging from minor grammar edits to generating full paragraphs. Other questions come to mind, including the source of the generated content, the extent of editing, and whether the claims were suggested by AI or made by the author. One of the tenets of transparency is granularity, in which the degree of AI influence is made clear to readers.
To make the above disclosure more effective, it would be best to also include information on whether multiple drafts were AI-assisted, which prompt framework was used (if any), and whether there were any authorship guardrails when AI assistants were used. The author’s responsibility can be considered meaningful only when paired with evidence of verification processes (e.g., fact-checking and source validation). Further, how was the accuracy of AI-generated content validated, what biases were apparent when reviewing AI-generated output, and how were these reconciled?
The above policy is a good start but needs to be modified to specify the scope of AI use, clarify boundaries and sources of original arguments and claims, and describe how fact-checking was done. As AI becomes ubiquitous in academic and professional writing, the issue is not about the use of AI – it is about the vagueness of disclosures made. As educators and professionals, we need to be less focused on which AI tools are used and concentrate on how clearly and thoroughly their use is explained.
Recommendations
Several core principles, such as accountability, transparent disclosure, confidentiality of data, and verification of output, are essential to AI disclosure. Although specific expectations for AI disclosure can vary by publisher, the following guiding principles can help authors (and publishers) frame their AI Disclosure statements to demonstrate the ethical and acceptable use of AI tools in research and writing.
AI Disclosure statements are more than footnotes in a manuscript (or assignment, report, or instructional resource). Generative AI tools excel at summarizing, rewriting, classifying, translating, coding, and creating text, images, and videos. AI tools are being used during the entire lifecycle of any research and writing project. Therefore, the specific use of AI tools must be documented across all stages, from the research design through the final preparation of the manuscript. This log should include any prompts, prompt frameworks, codes, or AI agents used to generate the output. The log is especially important for multi-author work where each author may have used AI tools to accomplish different tasks. Publishers, institutions, or organizations may not formally require submission of the detailed log, but preserving these records will allow authors to answer any editorial integrity questions raised.
Many AI policies exempt ‘light editing’ such as grammar, spelling, and punctuation, where the AI tool does not introduce new ideas or modify the argument or interpretation. However, substantive use of AI, such as rewriting paragraphs, translating portions of the manuscript, synthesizing the literature, generating new arguments, or preparing figures and tables, should be clearly disclosed.
Helpful AI Disclosure statements should also include the name of the tool used and the version number (e.g., ChatGPT 5.5), as some tools can significantly affect the output generated based on their training date. Traceability of AI-generated output provide a clear record in case verification and auditing of results are needed later. In a recent development, Claude (a popular LLM used by scientists) has launched a version called Claude Science that “provides an auditable history behind every output so results can be validated and reproduced”.
Since there is currently no universal standard for AI Disclosures and institutions may have varying levels of policies for reporting AI use, the guiding principles above can help authors craft a flexible framework that demonstrates accountability, transparency, and ethical reporting of AI use, while allowing disclosures to be tailored to specific requirements when the manuscript (or assignment, report, or instructional resource) is ready to be submitted.
The following example is provided for illustrative purposes only and is not intended to represent a universal or preferred format for AI disclosure statements.
Sample AI Disclosure
Generative AI tools were used during the preparation of this manuscript to support selected editorial tasks, including brainstorming alternative wording, improving sentence clarity, reviewing organization, and identifying areas where additional explanation might strengthen the discussion. AI was not used to generate the article’s central thesis or substantive arguments.
The author independently conducted the literature review, developed the ideas presented, selected all examples, and made all decisions regarding the manuscript’s structure and content. AI-generated suggestions were evaluated critically, with all factual claims, interpretations, and references independently verified against original sources before inclusion. Any AI-generated text that was incorporated into the manuscript was substantially revised by the author to ensure accuracy, consistency, and alignment with the author’s intended meaning.
Throughout the writing process, the author maintained responsibility for all intellectual contributions, editorial decisions, and the final published content. No confidential or proprietary information was entered into AI systems during manuscript preparation.
Dr. Sunil Hazari is a Professor of Marketing in the Richards College of Business, University of West Georgia.
He teaches courses in Business Web Design, Marketing Research, Social Media, and Business Essentials of Artificial Intelligence. He is a reviewer for several Business and Marketing journals and is the Assistant Editor for Journal of Information Systems Education.
His current research involves investigating GenAI use for Business & Education, AI literacy, Ethics, Bias, Risks, and preparing students for responsible use of AI in the workplace.
For additional information, see www.sunilhazari.com/education
