Overview
Generative artificial intelligence (GenAI) now touches practically every stage of research, from literature searches to drafting manuscripts, analyzing data, and preparing grant applications. It’s no longer a question of whether researchers should disclose their use of AI, but rather to what extent disclosure of AI usage is required. Luckily, institutions can breathe a little easier thanks to the U.S. Office of Research Integrity (ORI). ORI issued guidance in August 2026 to aid institutions on how to proceed with research misconduct allegations when GenAI is used. ORI doesn’t require institutions to have a GenAI policy, but the central message is clear: identify the tools you’ve used and explain how you’ve used them. This article will provide researchers with practical steps they can apply today with their current and future research projects.
Highlights of Guidance
ORI doesn’t invent a new category of misconduct; instead, it applies its existing definitions of fabrication, falsification, and plagiarism to GenAI. According to ORI’s guidance for researchers, identifying the tools and explaining the steps taken to generate and analyze data support reproducibility and may help protect researchers against research misconduct allegations.
The guidance highlights the following points:
- Disclosure isn’t enough. Simply disclosing that GenAI was used doesn’t tell the institution how it was used or if the tool(s) were used in an acceptable manner. Undisclosed or inaccurate GenAI data generation may be evidence of fabrication or falsification.
- Verify, then trust. Verify results generated via GenAI as well as AI-assisted analyses before you trust them enough to include them in your research publication. GenAI tools can alter data without your knowledge and researchers are responsible for any reported output.
- Hallucinated citations can count as fabrication. This depends on whether the citation counts as data, such as when references are included in a literature review article.
- Institutions must conduct their own investigations. The same principle that applies to researchers also applies to institutions: ORI recommends that solely relying on AI plagiarism detectors as the primary evidence in a misconduct proceeding is inadequate.
Five Best Practices for AI Disclosure
Whether GenAI or any other AI method is used, agencies and journals may have different approaches to how AI is treated in research publications.
There isn’t a single, universal standard.
- The U.S. National Institutes of Health (NIH) considers generating data with AI while claiming it was generated by other means to constitute fabrication or falsification.
- The U.S. National Science Foundation (NSF) states that fabrication, falsification, or plagiarism committed through AI-based tools is classified as research misconduct.
- Journals following the International Committee of Medical Journal Editors (ICMJE) recommendations ask authors at the time of submission whether AI-assisted technologies were used.
- The Committee on Publication Ethics (COPE) explains that a description of when AI was used should be included in the materials and methods section.
Institutions can have their researchers follow these top five best practices:
Understand funding agencies’ and journals’ AI disclosure policies. Since there isn’t a single rule regarding GenAI use and disclosure, researchers should carefully review policies and procedures from these organizations. If you are unsure or unclear, it’s better to ask than to make assumptions about an organization’s policies.
Be specific when disclosing GenAI use and keep a record of what you’ve done throughout the research project. If you were reading an article and saw a statement that says, “AI was used to assist with this project,” would that tell you anything about which tools were used, how they were used, when they were used, or whether their output was verified? Probably not. If you have to guess, an agency or journal would likely have to guess, too. A well-documented AI-usage log should include prompts, outputs, specific tools used along with their versions, dates, edits made, and how usage of these tools adheres to institutional, funder, and journal policies.
Place the disclosure statement where readers will look for it. Researchers will want to defer to the specific funding agency’s and journal’s policies for where these statements should live. When in doubt, it’s recommended to write this information in the methods section of your article and include a summary of use in your cover letter.
Verify the result and own it once trusted. AI can’t be an author, nor does disclosing AI use transfer responsibility to the tool used. Researchers using these tools are held accountable for all outputs. Researchers are strongly encouraged to check every citation, confirm all claims made by AI, and review for text that may have been generated from existing work.
Protect the participants you’re collecting data from when AI is used during any part of the research project. Researchers must uphold standards of confidentiality for identifiable participant data, unpublished data, and other private information. Confirm best practices with your institution when using AI in human subjects research. This can include verifying with your institutional review board (IRB) if a specific tool is permitted for use, explaining in your protocol when and how AI could be used, and including plain language in the informed consent form so that participants can understand how AI and their data will interact with one another.
Institutions Can Create a Starting Template for Researchers
The following verbiage can serve as a starting point for submissions to journals or funding agencies:
“Generative AI was used in the preparation of this work. [Tool name, version, developer] was used on [dates] to [specific task, such as editing the language of the Discussion section]. The authors reviewed and edited all AI-generated content and verified all citations and factual statements. The authors take full responsibility for the content. No generative AI tool was used to generate, alter, or analyze research data or images.”
The following language can serve as a starting point for informed consent forms:
“[Tool name, version, developer] will be used to analyze (or transcribe) data collected from participants. Only de-identified information will be entered into the tool. Once the data is inputted, it can’t be withdrawn even if you withdraw yourself or otherwise no longer participate in the study.”
The bottom line: using AI with thoughtful intent is not research misconduct. The risk of research misconduct increases with undisclosed, unverified, or inaccurately described GenAI use.
References and Resources
- Committee on Publication Ethics (COPE). 2023. “COPE Position: Authorship and AI Tools.” Accessed October 16, 2026.
- International Committee of Medical Journal Editors (ICMJE). 2026. “Recommendations, Section V.A Use of AI by Authors.” Accessed October 16, 2026.
- National Institutes of Health (NIH). 2026. “Helpful Reminders to Ensure Integrity of NIH-Supported Research When Using Artificial Intelligence.” Accessed October 16, 2026.
- National Science Foundation (NSF). 2025. “Policy Notice: Implementation of Policy Changes to Proposal and Award Policies and Procedures Guide (PAPPG 24-1), Supplement 1. Provision 16. Research Misconduct.” Accessed October 16, 2026.
- The Office of Research Integrity (ORI). 2026. “ORI’s Guidance on Generative Artificial Intelligence Released.” Accessed October 16, 2026.
