Your Professor WILL Catch You Using AI (Unless You Do This)

Ethical AI use in academia hinges on one non-negotiable rule: researchers must take full accountability for every word they submit, while disclosing exactly how AI was used. The practical takeaway is simple—AI can be used, but only with transparent reporting of where it was applied, why it was applied, and how much it was involved. That disclosure expectation becomes stricter as AI shifts from routine assistance into higher-impact roles like automating parts of research or intervening with advice and prevention.

The transcript breaks AI use into four levels. “Routine use” is treated as low risk: tools such as ChatGPT for spell checking or Grammarly for writing assistance don’t require detailed disclosure in most cases because they don’t materially shape research outputs. The concern rises with “automate” use, where AI is used to run parts of the research process—such as contacting participants, selecting data, or generating recurring reports. In those cases, disclosure needs to start early because automation can quietly influence methods and outcomes.

“Generate” use is the most familiar academic pattern: AI producing a literature review, a research report, a first draft, or even figures and schematics. Here, the transcript emphasizes that disclosure must be specific—naming the tool and describing the purpose and extent of AI involvement. The most sensitive category is “intervention,” where AI provides advice or helps prevent outcomes. That’s where bias and model behavior complexity become especially dangerous, since small changes in model versions and underlying behavior can quickly alter what the system recommends.

Beyond categorizing usage, the transcript lays out what institutions and journals typically want. A common journal practice is to include an AI disclosure statement in the published work, often appearing just before the references. That statement should name the AI tool used and clarify the purpose (e.g., first draft) and the level of oversight (e.g., whether a human review process was in place).

University guidance is described as more detailed. One example requires listing the tool name(s) and version(s) for generative AI, the publisher, and a URL, along with a brief description of how the tool was used. Some universities also provide a form—sometimes an A4-page template—where students must report their AI usage in a structured way.

The transcript also flags a major ethical boundary: data handling. Researchers shouldn’t “willy-nilly” paste confidential or sensitive data into AI tools without matching privacy and security requirements to the data’s sensitivity. Many universities offer a “sandbox” environment that limits leakage risk, and the transcript advises using AI with confidential data only when the university provides assurances. Even when tools offer settings like “don’t use my data for modeling,” that may not be enough; researchers are urged to scrutinize terms and conditions and understand where data goes and how it might be disclosed. In short: use AI if it’s necessary, but document it thoroughly and protect data rigorously—especially when AI moves from drafting to automation or intervention.