Virginia Tech® home

Research

Using AI for Research

This page covers four common areas where guidance on the use of AI for research might be helpful: during literature review and synthesis, during coding and analysis, in developing manuscripts and proposals, and when working with living research subjects or protected data.

Getting Oriented

The use of AI in research is shaped by several sources, including the Responsible and Ethical AI Principles, the Office of Scholarly Integrity and Research Compliance (SIRC) Considerations for Researchers on the use of AI, your Institutional Review Board (IRB) protocol, and the terms dictated by sponsors and publishers. At times, the guidance provided may present ambiguities. Identifying which guidelines apply to a particular decision may be a significant element of the work.

Responsible Use

In research, it is particularly important to be conscientious about the use of AI. This is especially true during four phases of research where AI may be used: during literature review and synthesis, during coding and data analysis, when preparing manuscripts and proposals, and when working with live research subjects or protected data. Each recommendation below is followed by the related Principle for Responsible and Ethical AI in parentheses.

Literature review and synthesis

AI tools can help find adjacent literature, summarize dense content, and surface concepts or approaches you might have missed. They can also fabricate citations, misattribute claims, and hallucinate papers that don’t exist. Verification against the original source is strongly recommended for any citation.

  • Every citation that appears in your work must be one that you have read. AI-suggested references are leads, not sources. (Human Judgment)
    • Track down every reference through the library catalog, Digital Object Identifier (DOI) lookup, or the publisher's site.
    • Always remember that reading the abstract is not reading the paper. Especially when a citation was suggested by AI, reading the full paper is essential.
  • An AI-generated summary of someone else's work may not be accurate, or may flatten nuances that should have been preserved. You are accountable for the accuracy of all your sources, including those that are AI-generated. (Responsible & Ethical Use)
    • For any paper you cite or build on, take time to read the methods and limitations sections with care.
    • Compare the AI summary to the abstract; investigate any divergence.
  • The biggest legitimate win is increasing the breadth of your literature review and surfacing literature you wouldn't have found by keyword search alone. (Innovation for Good)
    • Use AI to ask “what fields outside mine work on this problem?”
    • Treat the answer as a reading list, not a synthesis.

Coding, data analysis, and statistical methods

AI assistants can help write and review code, analyze data, or suggest statistical methods for research. However, they are frequently wrong about statistics and data analysis in ways that are easy to miss. Reproducibility is key. Document the prompts used, the model version, and your workflow well enough that the analysis could be re-run in the future.

  • You are responsible for every data analysis decision and statistical claim. Verify AI suggestions independently before including them in your work. (Human Judgment)
    • Use statistical software or another human-based computational method to test any AI-proposed statistical method before applying the method.
  • Reproducibility requires that someone else, given your data and code, would be able to re-run your analysis. AI use that erases that path undermines the integrity of the research. (Responsible & Ethical Use)
    • Commit code, prompts used, and AI versions to your repository alongside the data.
  • Data you load into a coding assistant may leave your machine. Treat the AI assistant as an external collaborator under whatever data agreements bind your project. (Data Security & Privacy)
    • Don’t paste participant data, regulated data, or proprietary data into uncontrolled AI tools.
    • For sensitive data, use locally-hosted models or university-approved environments only.
  • Your published methods section must describe what you actually did, including AI’s role. (Fairness & Transparency)
    • Disclose AI tools, versions, and the specific tasks they performed.
    • If asked to provide prompts or AI-generated code, you should be able to comply.

Manuscripts, proposals, and reviews

Most journals and major sponsors have published their positions on AI use in the development of proposals, manuscripts, and reviews. While these positions vary, three key points are: AI cannot be granted authorship, AI use must be disclosed, and confidential drafts entrusted for peer review may not be uploaded to AI tools without permission.

  • Authorship requires intellectual contribution and accountability. AI cannot meet either bar. (Human Judgment)
    • The argument, the contribution to research, and the conclusions reached must originate with you.
    • If your document makes a claim, you must be able to defend it.
  • Read the journal or the sponsor’s specific AI policies for the manuscript or proposal you’re submitting. Ignorance of a policy is not a defense. (Fairness & Transparency)
    • Check the AI-disclosure rules prior to submission, not after.
    • If AI disclosure is required, write the disclosure carefully, keeping the publisher’s policies in mind.
  • Manuscripts under peer review and unfunded proposals are confidential. Pasting them into a public AI tool breaches that confidentiality. (Data Security & Privacy)
    • When serving as a peer reviewer, follow the journal’s stated AI policy. Many forbid AI use entirely.
    • For proposals, don’t paste unpublished aims or budget rationale statements into uncontrolled tools. 
  • The pressure to write and review proposals and manuscripts faster doesn’t dissolve the obligation to be honest about how the work was done. (Responsible & Ethical Use)
    • Discuss AI authorship norms with co-authors at the proposal stage, not at submission.

Working with participants and protected data

If you collect or hold data from human subjects, AI use is bound by your IRB protocol, your consent forms, and the federal regulations protecting your participants. IRB approval precedes AI use, and the tool must be approved for the relevant data classification.

  • Research participant data must only go into systems explicitly approved to safely handle that data risk classification, whether or not it is de-identified. Most consumer AI tools are not. (Data Security & Privacy)
    • Know the data classification for your project before piloting any AI workflow.
    • Use university-approved environments for identifiable data.
    • De-identify before AI involvement when feasible, and verify the de-identification.
  • Your IRB protocol defines the scope of the study. If AI use was not named in the protocol, consult with SIRC about whether an amendment is needed before proceeding. (Responsible & Ethical Use)
    • If AI will touch participant data in any form, name it in the protocol.
    • If consent forms didn’t disclose AI use, adding it may require re-consent or an amendment; check with the IRB office.
  • Participants who enrolled in your study based their decision on a description of how their data would be used. AI use that exceeds that description may raise concerns related to that consent. Consult SIRC before proceeding. (Human-Centered Benefit)
    • Match AI use to what participants were told they could expect.
    • For sensitive populations, default to more conservative choices than the AI use described in the protocol.
  • If AI was used to recruit, classify, or analyze participants, that fact is part of the methods, and may need to be part of the consent. (Fairness & Transparency)
    • Disclose the use of AI in publications.
    • Watch for differences in model performance across participant subgroups and take steps to disclose potential AI-related bias. 

When to ask for a consultation

Sponsors, IRBs, journals, and co-authors all have a stake in how you use AI. Some situations require a consultation with one or more of them before you proceed. Consult one of the contacts below when these come up.

  • When you’re considering loading research participant data into an AI tool that wasn’t named in your IRB protocol. Consult SIRC or the IRB office about whether an amendment to your protocol is required before proceeding.
  • When a sponsor’s terms or a journal’s policy is ambiguous about AI use. Write and ask the program officer or editorial office; keep the answer.
  • When an AI tool produces an analysis that contradicts your prior results, and you can’t reconstruct why. Pause publication, reach out to your co-authors, and reproduce manually before submitting.
  • When co-authors disagree about AI use or disclosure. Resolve before drafting, not after submission.
  • When you’re peer-reviewing a manuscript and considering AI assistance. Many journals forbid this; check with the editorial office before you do anything with the file.
  • When an AI workflow shows different performance for different participant subgroups. This is a finding to investigate, not an artifact to suppress.

Virginia Tech Resources