Generative AI and the Research Record - The University of Utah

Generative AI can be used to help researchers in many activities, including reviewing literature, organizing ideas, summarizing complex material, and improving writing. U Research offers resources to help researchers use these tools responsibly to protect the integrity, accuracy, and origionality of their work.

The Office of the Research Integrity Officer (ORIO) provides education about AI-use policies and guidelines to help promote the ethical and rigorous use of AI in research.

Every researcher needs to understand the AI-use standards that apply to them because every researcher is accountable for what they produce through AI. It’s not just about ensuring the integrity of their work but protecting themselves from allegations of research misconduct. So, if you or the members of your team don’t understand the expectations for AI use that apply to you, we invite you to ask a question and have a conversation. We’re here to help. – Zach Mitchell, the U’s Research Integrity Officer (RIO)

The support and guidance offered by the ORIO is intended to help all research team members identify the standards that apply to them based on University guidelines ( available at ai.utah.edu) and the unique combination of their specific discipline, funding profile, and target journals. The ORIO can also highlight common concerns, adaptive strategies, and people to contact for support, including the One-U Responsible AI Initiative.

Common issues seen by research administrators

Researcher administrators are seeing recurring issues with AI-assisted work. Some AI-generated output can look polished while containing inaccurate or unsupported claims, hallucinated citations, real citations that do not support the claims they appear beside, content that has not been appropriately attributed, or even data that has been invented or modified to improve accuracy or reach statistical significance. Reviewing content carefully before it enters a grant proposal, IRB application, dissertation, or manuscript helps researchers avoid these problems and maintain confidence in their work.

Compounding these output challenges are gaps in training and assumptions about trust. Graduate students and postdoctoral researchers are not receiving enough mentoring on appropriate AI use and faculty at every career stage are discovering that materials supplied by colleagues cannot be blindly trusted. The gap between appearance and verification can weaken the trust that research teams need to do their work well.

When AI-assisted content needs closer review

Federal research misconduct standards focus on fabrication, falsification, and plagiarism. Fabrication involves making up data or results and reporting them. Falsification includes changing or omitting research materials, processes, data, or results so that the research record does not accurately represent the work. Plagiarism involves using another person’s ideas, processes, results, or words without appropriate credit.

AI-generated content may raise one or more of these concerns depending on the discipline, the research context, and accepted practice. A tool does not remove a researcher’s responsibility for a statement, a citation, or a result. Anyone who submits work into the research record should be able to explain where it came from and verify that it is accurate.

Practical steps before submission

Confirm the accuracy of any language, data, quotations, or references you did not create. Asking an AI tool to check scientific background or supporting references may help identify issues, but it does not replace review against the original evidence.

Research teams can also set clear, lab-specific expectations for AI use. Those expectations should address which uses are acceptable, how to disclosure appropriatly, and how the team reviews high-stakes materials before submission.

Requirements can differ across publishers, funders, sponsors, and individual research settings. For example, NIH Notice NOT-OD-25-132 states that NIH will not consider applications or sections of applications that have been substantially developed by AI to be the applicant’s original ideas. The notice also describes possible referral for research misconduct review and award consequences when AI use is detected after an award.

Before submitting AI-assisted content, ask:

- Can I verify this claim against the underlying evidence?

- Does the cited work support the statement?

- Have I preserved the source and context for this result?

- Does this use comply with the expectations that govern this work?

These are not new hurdles. They are the practices that protect reliable research: close review, clear attribution, and accountability for what enters the record. Used consistently, they allow researchers to benefit from new tools while preserving rigor, integrity, and confidence in the research record.

Have questions, concerns, or need support?

Contact Zachary.Mitchell@hsc.utah.edu or Caren.Frost@socwk.utah.edu.