CiteOwl

Blog/Using AI honestly

I accidentally plagiarized. What happens now?

Alexander Matuška· Updated 20 September 2026· 13 min read

A first small case usually lands in the low category your university keeps beneath misconduct. Your marker handles it, not a panel. That is a finding that misconduct did not occur. Manchester calls it poor academic practice, Melbourne a minor and unintentional departure from accepted scholarly conventions. What decides which side you land on is rarely the mistake itself. It is whether you raise it before somebody else does.

First, work out which of five things happened

These all get called plagiarism and the procedures do not treat them alike. Open your draft, find the passage, and decide which of the five it is.

• A missing citation for a source you did read. The source is in your bibliography and you left the marker out of the text. The University of Manchester's Academic Malpractice Procedure names this case under poor academic practice rather than malpractice: "A limited amount of material or copied text where the source is referenced in the bibliography but is not properly cited within the body of the assessment." The word doing the work there is limited.
• A paraphrase that stayed too close. Manchester puts a limited amount of text "subject to reasonable linguistic and phrasing changes but is without citation" in that same low category. Oxford is blunter about the general case: paraphrasing "by altering a few words and changing their order, or by closely following the structure of their argument, is plagiarism if you do not give due acknowledgement". Extent decides which of the two you are in.
• A quotation without quotation marks. Heavier than a missing citation even when the source is cited, because the reader cannot see where your words stop. Oxford again: "Quotations must always be identified as such by the use of either quotation marks or indentation, and with full referencing of the sources cited. It must always be apparent to the reader which parts are your own independent work."
• Work of your own, submitted twice. Every procedure we read names this, and permission is the pivot. The University of Toronto's Code of Behaviour makes it an offence to submit "without the knowledge and approval of the instructor to whom it is submitted, any academic work that has previously been submitted" elsewhere. UCL and Maryland word it similarly. The long answer is in can I reuse my own paper.
• A reference you never opened. Oxford's guidance on inaccurate citation says "you should not include anything in your references or bibliography that you have not actually consulted", and Toronto makes it a separate offence to submit work "containing a purported statement of fact or reference to a source which has been concocted". The University of Maryland's Code files invention of a citation under Fabrication, a different charge from plagiarism.

The first four are failures of attribution: the material is real and the credit is missing. The fifth is a statement about your own work that is not true, which is why institutions give it a different name and a different route. If the reference came out of a chatbot, read my professor said my sources do not exist next.

What the procedures actually say

Most universities run a formal category beneath misconduct for exactly the case this page is about. They call it different things, which matters, because you cannot look up your own procedure until you know its word for it.

The low category is not a discount on misconduct, it is a finding that misconduct did not occur, which is why Melbourne says identifying it "is not a finding of academic misconduct". The ceiling above it is real too: Oxford says penalties "may range from deduction of marks to expulsion from the University, depending on the seriousness of the occurrence". Melbourne's penalty schedule places plagiarism at Level 1, with a standard range running from a reprimand and caution up to failure in a subject, and that range does not include suspension, expulsion or revocation of an award. Which end you are near is decided by extent, by whether this is a first case, and by what you do next.

Intent changes the penalty, not the finding

The question everyone asks first is whether it counts if you did not mean it. In writing, at every institution we read, it counts.

Manchester states that academic malpractice "can result from a deliberate act of cheating or may be committed unintentionally" and that "whether intended or not, all incidents of academic malpractice will be treated seriously by the University". Oxford: "Even if plagiarism is inadvertent, it can result in a penalty." Toronto closes the gap directly: wherever its Code makes an offence depend on knowing, "the offence shall likewise be deemed to have been committed if the person ought reasonably to have known". Melbourne defines a breach as conduct with "the purpose or effect" of an unfair academic advantage, and an effect does not require you to have wanted it.

What intent is, everywhere, is a lever on the penalty. UCL lists among the mitigating factors an adjudicator must weigh: "It is a first offence", "The student admitted the offence at the earliest opportunity", and "There is evidence that the student did not intend their actions to lead to an unfair academic advantage". Harvard's Honor Council says its outcomes "generally depend on two criteria: 1) the seriousness of the infraction; and 2) extenuating circumstances, including the extent to which a student has had similar trouble before".

The same logic holds if the trigger was an AI tool rather than a slip in your notes. Melbourne files unauthorised or undisclosed use of generative AI as its own breach category alongside plagiarism, and where the line sits is the subject of is using AI to write essays cheating. If somebody has already asked you about it, accused of using AI on an essay is the closer page.

The one university that publishes its numbers

Nearly every university publishes its procedure and almost none publishes what happens at the end of one. UNSW Sydney is the exception we found. Its Conduct and Integrity office issues an annual report recording the outcome of every substantiated case. The 2024 edition, published in June 2025, covers 82,272 enrolled students and 2,274 cases. Of the 1,765 plagiarism cases logged, 1,332 were classed as poor scholarship or less serious plagiarism and 433 as serious.

Read the columns rather than the rows. Across the 1,321 outcomes recorded in the two lanes schools handle themselves, there is no fail in the course, no suspension and no exclusion. The heaviest thing at that level is a zero on the assessment, recorded 257 times. Suspensions and exclusions sit in the serious lane, and even there they are 40 of 889 outcomes. University-wide in 2024 there were 13 suspensions and 35 permanent exclusions, which the report puts at 0.02% and 0.04% of enrolled students, and it calls exclusion a last resort for admission fraud, large scale contract cheating and serious behavioural breaches.

The report supplies its own caveat: more than one outcome can be applied to a case, so these are outcomes rather than students. Column totals are 171, 1,150 and 889. The rows left out are small ones: remedial action only, a formal reprimand, one revocation of an award, ten not recorded.

One university is one university, and none of this transfers to yours as a promise. What it does is put a floor under the fear. The published distribution looks nothing like the thing you are imagining at two in the morning.

Why raising it first is the whole game

UNSW states the trade in one sentence: "Where students do not make a full and upfront admission, and an investigation subsequently substantiates the allegation, then a Fail (00FL) for the course is the standard penalty for a single instance of misconduct." That penalty was issued 551 times in 2024. The lighter outcome, a zero on the assessment, is what the office says it typically issues "as a reduced penalty when a student chooses to admit to the alleged misconduct".

Nobody else publishes the arithmetic, but the rule recurs.

• Maryland runs a self-referral route. If "no member of the University had a suspicion" of the misconduct before the student reported it, and the referral is accepted, "the Student will not be charged with Academic Misconduct or have a disciplinary record for the reported incident". It is available once per enrolment, an educational sanction still follows, and the instructor may still adjust the grade. Once somebody does suspect, the same document says a self-referral "may be considered a Mitigating Factor for purposes of sanctioning" instead. The window closes.
• Toronto's Provost tells deans that a student who admits "at the Divisional level, or earlier may be demonstrating a significant level of insight into their behaviour", which "may justify the imposition of a less significant sanction than would otherwise be appropriate".
• UCL runs an expedited procedure built on "the early admission of guilt", and lists admitting at the earliest opportunity among the mitigating factors.
• Melbourne says it in as many words: "Students may self-report to the University that they may have breached the University's academic integrity standards."

Only Maryland's promises anything close to a clean outcome, and it promises it under conditions. No procedure guarantees you anything for coming forward, and every procedure we read treats it as worth something.

The sequence, and the message you can send

1. Find your institution's procedure and learn its vocabulary. Search your university's name with "academic misconduct procedure" or "academic integrity policy". You want two facts: what it calls the low category, and who decides first, which at most institutions is your marker or a subject coordinator rather than a panel.
2. Work out what actually happened. Put the passage next to the source and write down which of the five kinds it is and what share of the assessment it covers. Extent is what the frameworks are keyed to: UCL's routes a case to a module leader, an exam board chair or a panel depending on whether the offending material is up to 10%, up to a third, or more.
3. Tell your marker or supervisor, in writing, today. Not the head of department. The person who set the assessment.
4. Bring the source and your drafts. Keeping drafts is ordinary practice, and having them in the room keeps the conversation on the actual passage.
5. Be exact rather than apologetic. Somebody has to write down what happened, so give them the sentence they need. A page of remorse with no facts in it makes their job harder and yours worse.
6. Ask what the procedure is and who decides. You are entitled to know. UCL's requires that a student suspected of misconduct be directed to advice and support, and that anyone invited to a panel be told they may bring a friend. Most institutions have an equivalent.

Something you can send, with the details swapped:

That is the whole message. It names the assessment, names what happened, offers the material and asks for the procedure. It does not argue and it does not ask to be let off.

What not to do

• Do not quietly resubmit a corrected version. A second file does not replace the first in anyone's records. Fix the draft, then tell somebody what you fixed and why.
• Do not deny something a report will show. Harvard's Honor Council procedures say the Council "may take into consideration a student's demonstrated commitment to honesty", and that "Dishonesty may result in increased sanctions or an additional disciplinary charge". A denial that collapses turns a marks problem into a character problem.
• Do not argue about the similarity percentage. The number is not a verdict, and what is being assessed is the passage rather than the score. What is a good Turnitin similarity score covers why.
• Do not wait to see whether anyone noticed. Every advantage on this page depends on you speaking first, and it disappears the moment somebody else raises it.
• Do not let it become the habit. Every procedure treats a repeat as a different case. UCL escalates a second offence to a higher adjudicator regardless of extent, and Toronto's Provost tells divisions to strongly consider a suspension where there is a prior offence. The prevention side is how to avoid plagiarism, and the habit that ends this is citing while you write.

That habit is what CiteOwl is built around. It searches OpenAlex, CrossRef, Unpaywall and the web, reads the papers it retrieves, and writes each claim from a passage it keeps beside the sentence, so the citation exists at the moment the sentence does. Every edit it makes arrives as a diff you accept or reject, so nothing reaches your draft that you have not read.