Using AI tools in academic work

Generative AI tools are now part of the environment students work in. Institutions have not converged on a single rule. Some ban generative use in assessed work entirely; some allow it with disclosure; some allow it for specific tasks and not others. The only safe operating assumption is that your course handbook and assessment brief override every general article, including this one.

What follows is a framework for reading those rules and avoiding the failure modes that end degrees and applications.

Separate three different questions

Question	Why it matters
Is it allowed?	Policy and the assessment brief
Is it wise?	Learning value, interview risk, skill atrophy
Is it disclosed correctly?	Integrity processes treat undeclared use as the offence even when use itself would have been permitted

Students often collapse these into one instinct (“everyone uses it”). Integrity offices do not.

Common institutional positions

• Prohibited for assessed work — tool use treated like unauthorised collaboration or contract cheating
• Allowed for defined tasks — e.g. brainstorming, grammar help, code explanation — with a required disclosure statement
• Allowed unless the brief says otherwise — still usually requiring you not to submit AI text as your own analysis
• Required literacy — some courses teach and assess AI use deliberately; rules there are different again

Always check: module handbook, assessment cover sheet, department AI statement, and central academic integrity policy. Conflicts happen; the assessment brief usually wins for that piece of work.

Uses that are frequently restricted

• Generating prose that you submit as your own argument or analysis
• Generating code you submit as your own solution where coding skill is being assessed
• Producing reflective statements, personal statements or reference letters
• Fabricating citations, quotations or data (this is misconduct whether or not AI was involved)
• Uploading confidential, patient, commercial or unpublished data into a public tool

Uses that are more often tolerated — when policy allows

• Explaining a concept you then restate in your own words and verify against a textbook
• Suggesting search terms or outlining a revision plan
• Grammar and clarity editing of text you wrote (still disclose if required)
• Practice questions and worked examples for formative study
• Helping debug code when the course permits similar human help

Disclosure that actually works

When disclosure is required, vague lines like “AI was used” are often insufficient. Prefer:

1. Name the tool and version if known
2. State the purpose (e.g. “suggested structure for section 2”; “grammar pass”)
3. State what you did afterwards (rewrote, verified sources, discarded suggestions)
4. Confirm that all citations were checked against the original sources

If the institution provides a disclosure template, use it.

Detection, false confidence and evidence

AI detectors are probabilistic and contested. Institutions increasingly rely less on a single detector score and more on: mismatch with prior writing, inability to discuss the work in viva or interview, hallucinated sources, and process evidence (drafts, notes, version history).

Keep drafts. Keep notes. Be able to explain every claim. That discipline protects you whether or not a detector is involved.

A practical decision rule

1. Read the assessment brief and integrity policy for that piece of work.
2. If use is unclear, ask the module convenor in writing before you submit.
3. If allowed, use the tool for scaffolding — not for the assessed intellectual work itself — unless the brief explicitly assesses AI-assisted production.
4. Disclose exactly as required.
5. Verify every factual claim and citation yourself.

Related: citation managers compared and how selection committees read applications.