Academic Integrity

A commitment, even in the face of adversity, to five fundamental values: honesty, trust, fairness, respect, and responsibility." — International Centre for Academic Integrity

Academic integrity is not a rulebook to be memorised so much as a working relationship between you and the people who read, mark, and rely on your work. A grade only means something if it reflects what you actually understand and can do — the moment it stops reflecting that, it stops being useful to anyone, including you. The five values above give the idea some structure: honesty means representing your work truthfully — not claiming credit for what isn't yours, and not disguising what is. Trust is what lets an examiner take your word for it rather than re-deriving every claim from scratch, and it only survives if most people honour it most of the time. Fairness means the student who spent the weekend struggling through the reading is judged by the same yardstick as everyone else. Respect means acknowledging the people whose ideas you're building on, rather than letting a citation quietly disappear. And responsibility means owning the consequences of what you submit, even when a shortcut felt tempting under deadline pressure.

Common Violations

Most integrity breaches fall into a handful of recognisable patterns, and it helps to know the difference because institutions often treat them differently. Plagiarism is presenting someone else's work or ideas as your own — the clearest case is a passage copied without citation, but it covers paraphrase that's too close to the original as well. Cheating is receiving help you weren't authorised to receive, whether that's a glance at a neighbour's exam script or an unapproved collaboration on a supposedly individual assignment. Fabrication is inventing data, results, or citations that don't exist — a survey that was never run, a source that says something it doesn't. Facilitation is helping someone else commit one of these breaches, which matters because the person who shares their answers is usually held just as responsible as the person who copies them. And self-plagiarism — reusing your own earlier work without disclosing it — surprises a lot of students, because it seems harmless to reuse your own words, but submitting the same piece of work for credit twice misrepresents how much new work you actually did.

Citation Fundamentals

When to Cite

The rule that actually holds up under pressure is simpler than most citation guides make it sound: cite anything that isn't yours and isn't common knowledge. That covers direct quotes, paraphrased ideas, statistics and data you didn't generate, and images or code you didn't write. You don't need to cite genuinely common knowledge — that water boils at 100°C needs no reference — nor your own original analysis, your own experimental results, or content you created yourself. The grey area in practice is usually "is this common knowledge in my field?" — if you had to look it up, and a reader in your discipline might reasonably not know it, cite it.

Citation Styles

Different disciplines have converged on different citation conventions, and using the wrong one for your field is a common (if minor) error. Psychology and education typically use APA, rendering an in-text citation as (Smith, 2023). The humanities lean on MLA, which cites a page number directly — (Smith 45) — rather than a year. History favours Chicago, most often as a numbered footnote: ¹Smith, Title, 45. Engineering and computing frequently use IEEE, where citations appear as bracketed numbers like [1] referring to a numbered reference list. None of these is more "correct" than another — check what your department or the venue you're submitting to expects, and apply it consistently throughout a single piece of work.

In-Text vs. Bibliography

A citation style actually has two halves that have to agree with each other: the short in-text marker that appears at the point you use a source, and the full bibliography entry that lets a reader track it down. The in-text marker is deliberately brief — often just an author and a year, sometimes a page number, as in (Smith, 2023, p. 42) — because its job is only to point somewhere, not to describe the source in full. The bibliography entry carries the detail: author, year, title, and publisher, for example Smith, J. (2023). Academic Writing. Oxford Press. A citation missing either half is broken — a bibliography entry nobody's in-text marker points to is padding, and an in-text marker with no matching bibliography entry can't be checked by anyone.

Proper Paraphrasing

Bad: changing a few words here and there (sometimes called patchwriting).
Good: read the source, understand the idea well enough to close it, write the idea in your own words from memory, then cite it.

Take the sentence "The rapid advancement of AI has transformed healthcare delivery." Swapping a couple of words — "the quick progress of AI has changed healthcare delivery" — is not paraphrasing; it's the original sentence wearing a thin disguise, and most plagiarism-detection software catches it easily because the sentence structure hasn't actually changed. Genuine paraphrasing restructures the idea entirely, in your own voice, and still credits the source: "Recent advances in artificial intelligence are reshaping how medical care is delivered (Smith, 2023)." The test worth applying to your own writing: could you have written that sentence without having read the source open in front of you? If not, close the source, and try again from what you actually remember of the idea.

Collaboration Boundaries

Collaboration and cheating aren't opposites so much as neighbours, and the line between them is about what ends up in the work you submit rather than whether you talked to anyone. Discussing concepts with classmates, explaining an idea to someone who's stuck, helping debug a piece of code, or working through material in a study group are all normal, expected parts of learning — none of that puts words or code into your submission that aren't yours. Sharing code or answers outright, writing part of an assignment for someone else, copying a solution, or splitting up an individually-assessed piece of work between several people cross into territory that undermines what the assessment is measuring. The rule of thumb that covers most edge cases: your submitted work has to reflect your own understanding, even if you got there with someone else's help along the way.

GenAI and Academic Integrity

The Challenge

Generative AI tools have made old integrity questions harder to answer at a glance, because the boundary between "getting help" and "submitting someone else's output" has become far less visible. A tool like ChatGPT or Claude can produce an entire essay or a working piece of code in seconds, and unlike a human collaborator, it leaves no obvious trace of having been involved. GitHub Copilot raises the same question inside an IDE, suggesting whole functions as you type. Even tools that look innocuous can cross a line depending on what's being assessed: Grammarly is fine for catching typos, but if it's rewriting your sentences for style, it may be doing more of the writing than you are. Translation tools sit in a similar spot — useful for checking your own translation, but a way to bypass a language requirement entirely if used to produce the whole piece.

Institutional Policies Vary

There is no single, settled answer to what counts as acceptable GenAI use, and different institutions — sometimes different departments within the same institution — have landed on genuinely different policies. Some ban GenAI use outright for assessed work. Others take a cite approach, allowing its use provided it's acknowledged like any other source. A disclose policy asks you to describe how you used it, often in an appendix, without necessarily restricting the use itself. And an integrate approach treats the tools as something to be taught responsibly rather than policed — building the skill of using them well into the curriculum itself. None of these is the default; always check what applies to the specific module or assessment in front of you, because assuming the wrong policy is itself a common source of accidental breaches.

Responsible GenAI Use

Where GenAI use is permitted at all, the acceptable and unacceptable ends of the spectrum are fairly easy to tell apart once you look at what the tool actually contributed. Using it to brainstorm topics, explain a concept you're stuck on, talk through what a piece of code is doing, or check your grammar all leave the substance of the work as yours — the tool is acting as a sounding board or a tutor, not an author. Using it to generate the final text you submit, to produce code you don't understand well enough to explain, or to translate or rewrite a whole piece crosses into submitting someone else's output as your own — the same problem plagiarism was always about, just with a different source. A reasonable disclosure, where one is expected, looks something like: "I used ChatGPT to brainstorm outline ideas. All writing and analysis is my own." — specific about what the tool did and didn't do, rather than a blanket statement either way.

Detection & Consequences

Detection Methods

Institutions use several methods to check submitted work, though none is infallible and it's worth understanding their real limits rather than treating any of them as magic. Text-matching tools like Turnitin, SafeAssign, and iThenticate compare a submission against a large corpus of prior work and published sources, and are reliable at spotting copied text but say nothing about whether paraphrased ideas are properly attributed. AI-content detectors such as GPTZero attempt to flag machine-generated text statistically, but they are genuinely unreliable — they produce both false positives on human writing and false negatives on AI writing, and shouldn't be treated as conclusive evidence on their own. Stylometric analysis compares a piece of writing's style against a student's known writing, which can flag a sudden, unexplained shift in voice. And student interviews — simply asking someone to explain or defend their own submitted work — remain one of the most effective checks precisely because they're hard to fake if you didn't do the work.

Typical Penalties

Consequences generally scale with both the severity of the breach and whether it's a first offence or a repeat one. Minor plagiarism — a poorly cited paraphrase, say — typically costs a zero on the affected assignment the first time, escalating to failing the course on a repeat. Major plagiarism, exam cheating, and fabrication are treated more seriously from the outset, usually starting at course failure and escalating to suspension or expulsion on repetition. The exact thresholds vary by institution, but the shape is consistent: a first, isolated lapse is usually treated as a teaching moment with a real cost attached, while a pattern of repeated breaches is treated as incompatible with continued study.

Best Practices for Students

Most integrity breaches trace back to time pressure rather than dishonesty as such — starting early is the single biggest thing that removes the temptation to take a shortcut, because rushed work is where patchwriting, uncredited borrowing, and last-minute misuse of AI tools tend to creep in. Track your sources as you go, whether that's with a tool like Zotero or Mendeley or just a running spreadsheet — reconstructing where an idea came from after the fact is far harder than noting it at the time. Cite as you write rather than promising yourself you'll add references later; "later" is when citations get forgotten. Paraphrasing is a skill that needs practice like any other — don't expect to do it well the first time you try under deadline pressure. When you're unsure whether something is allowed, ask the person setting the assessment rather than guessing — a question costs nothing, and a wrong guess can cost a great deal. And read the actual policy for the specific module or assessment in front of you, since as covered above, institutional and departmental rules on things like GenAI use genuinely differ.

For Instructors

From the instructor's side, integrity is best supported before it's ever tested, rather than caught after the fact. Clear expectations set out at the start of a module — what's allowed, what isn't, and why — prevent more breaches than any detection tool does, especially where GenAI is concerned given how much policies vary. Scaffolding assignments into stages, and building in some in-class or process-based writing, makes it far harder for someone to submit work they didn't actually produce, because there's a visible trail of drafts and decisions behind the final piece. On the detection side, similarity tools and comparison against a student's past work are useful first signals, and stylometric flags or an oral defence can help resolve genuine ambiguity when something looks off. When a breach is confirmed, a consistent rubric applied evenly across students matters as much as the severity of the penalty itself — and where the case allows it, treating the response as an educational conversation rather than a purely punitive one tends to produce better long-term outcomes than punishment alone.

Resources