The questions academics actually ask

The line depends on what the assignment is designed to measure. If it is designed to demonstrate the student's own reasoning, analysis, or capability — and AI has done the reasoning, analysis, and work instead — then the assessment is not measuring what it was designed to measure. That is the problem, regardless of whether we call it cheating.

The more useful question is not "is AI allowed?" but "what evidence of learning am I trying to capture, and does this submission provide it?" That question also tells you which assessments need redesigning more than it tells you what policy to set.

Your institution's policy sets the formal position. Where that policy is unclear, be explicit with students before they submit — not after. Ambiguity in the rules is unfair to students.

"AI-resistant" is the wrong frame. The right frame is "evidence of genuine understanding." Tasks that produce strong evidence of genuine understanding are harder to shortcut — with AI or any other tool.

The most effective moves are: add a process component (drafts, research journal, annotated bibliography); anchor the task to class-specific content the AI cannot know about; add a brief oral element even for written work; use staged submission so development is visible. The assessment design page covers each of these in detail, by discipline.

Increasingly, yes — though what that looks like varies by discipline. Students who graduate without understanding how AI tools work, where they fail, and what responsible use looks like are at a disadvantage in most professional contexts.

The most valuable AI literacy for undergraduates is not learning to prompt well — it is learning to evaluate AI output critically, understand its failure modes (especially hallucination and bias), and understand where AI use is and is not appropriate in their field. That is disciplinary AI literacy, and it is different in a law degree, a nursing programme, and a creative writing course.

The student guide on this site is designed to be shared directly with undergraduates as a starting point.

AI is genuinely useful here — but the workflow matters. The most effective approach is not to ask AI to generate feedback on individual student submissions (which would require entering their work into a third-party tool with data protection implications). Instead, use AI to generate a bank of high-quality, reusable feedback comments against your marking criteria — which you then select and personalise for individual students.

This gives you the speed benefit without the data protection risk, and the feedback quality is usually better because you are choosing from well-crafted options rather than editing generic AI output. The Prompt Lab has prompts designed for exactly this workflow.

No — not as evidence of misconduct. This is not a minority view. Researchers studying these tools consistently find significant false positive rates. AI detection tools have flagged legitimate student writing as AI-generated — including work by students whose first language is not English, students who write in a formal academic register, and students who closely follow structured conventions.

Using detection output as evidence in a misconduct process exposes your institution to serious fairness challenges and potential legal risk. Most institutional legal and academic integrity offices are advising against it.

The more reliable approach when you suspect a problem is a brief academic conversation with the student about their work. A student who has engaged with the material can usually explain it. One who has not, usually cannot. That conversation is also far less likely to lead to a wrongful allegation.

Follow your institution's academic integrity process — the same process you would use for any other suspected integrity breach. Document your concern clearly. Do not use AI detection output as the basis for the allegation.

Before initiating a formal process, have an academic conversation with the student. Ask them to explain their argument, describe their research process, or expand on a specific point. This conversation either confirms your concern or resolves it — and if it confirms it, you have something more useful than a detection tool output to bring to the process.

The principle is: treat this the way you would treat a suspicion that a student had plagiarised from a source. The misconduct is the same in kind. The process should be the same.

This depends on your institution's policy, your programme's regulations, and increasingly on the requirements of professional accreditation bodies for your discipline. Check all three before giving students guidance — and make sure your guidance is in writing.

The general principle is that a dissertation must represent the student's own original work and thinking. AI can support the research process — scoping literature, improving writing clarity, generating outlines — but the intellectual contribution, the argument, and the conclusions must be the student's. Any AI use should be disclosed as your institution requires.

For postgraduate research degrees, the original contribution to knowledge is the core requirement. This makes AI use in research design, analysis, and interpretation particularly sensitive — not because it is automatically prohibited, but because the distinction between "AI-assisted" and "AI-generated" is harder to draw and easier to blur.

Most major publishers now require it. Check the specific journal's author guidelines — requirements vary significantly between publishers and between journals within the same publisher group. As a baseline: if AI was used in the preparation of the manuscript in any substantive way — drafting, restructuring, literature scoping — disclose it in the methods section or acknowledgements.

The form of disclosure is evolving. A clear, specific statement is better than a vague one: say which tool, for what purpose, and what you then did with the output. "We used ChatGPT to improve the clarity of the discussion section, then substantially revised the draft" is useful. "AI tools were used in the preparation of this manuscript" is not.

No. No current major publisher or funder accepts AI as a named author, and this position is consistent across the field. Authorship requires accountability — the ability to stand behind the work, respond to correspondence, take responsibility for errors, and consent to publication. AI tools cannot do any of these things.

If AI has been used substantially in the preparation of the work, that use should be disclosed — in the acknowledgements, methods, or a dedicated AI use statement as the journal requires. Disclosure is not the same as authorship.

Generally yes, with caveats. AI can help you draft lay summaries, structure your methodology section, improve clarity, and anticipate reviewer questions. It cannot know your research context, your track record, or the specific priorities of your funder — all of which matter enormously in a competitive application.

Check your funder's terms before using AI in an application. Most funders have no specific policy and do not prohibit it. A small number do — particularly for applications that include a personal statement or where the applicant's own voice is part of the evaluation. When in doubt, ask your research office or the funder directly before submitting.

Whatever AI assistance you use, the intellectual content of the application — the research question, the methodology, the significance — must be yours. AI cannot make your research more fundable. It can make your application for fundable research clearer.

This is a data protection question as much as a pedagogical one. Student work is personal data — it identifies the student and may contain sensitive personal information. Entering it into a free AI tool without appropriate institutional cover is likely to be a GDPR compliance issue.

The practical workaround is to use AI to generate a bank of reusable feedback comments against your criteria — which you then select and personalise for individual students. This gives you the efficiency benefit without the data protection risk. The data protection guide covers this and the broader landscape.

If your institution has an enterprise agreement with one of these providers — Microsoft 365 Copilot with an education tenancy, for example — the data handling terms may be different. Check with your institution's data protection officer before assuming this.

Almost certainly not with free AI tools — and possibly not with paid tools either, depending on what the data is. Research participant data is personal data under GDPR. Sensitive categories — health data, data about beliefs, ethnicity, or sexual orientation — carry additional legal requirements. Processing this data through a third-party AI tool requires appropriate legal basis, a data processing agreement, and likely an update to your ethics approval.

Before using AI in any stage of participant-facing research, consult your institution's research ethics committee and data protection officer. This is not a formality — it is a genuine legal and ethical obligation, and getting it wrong exposes both you and your participants.