Fabricated AI references in academic work

The core problem: you remain responsible for what you submit

When a generative AI tool produces false or fabricated references, sometimes called “hallucinated” citations, the fundamental point is that the university treats the submitted work as your work. The AI’s error becomes your problem. Universities do not accept “the AI made it up” as a defence, because you are responsible for verifying and standing behind everything in an assessment you submit under your own name.

Fabricated references typically show up in one or more recognisable ways: a source that does not exist at all, a real author paired with a title they never wrote, a genuine article given the wrong journal, volume, page numbers or DOI, or a quotation attributed to a source that does not contain it. Markers are increasingly alert to these patterns because they are a common signature of undisclosed AI use.

How universities usually categorise this

There is no single national rule. Each university in England and Wales sets its own academic misconduct regulations, so the exact label and penalty depend on your institution’s code of practice. That said, fabricated references generally fall under one or more established categories of academic misconduct:

Fabrication, meaning inventing data, sources or references that do not exist. Fabricated citations are a textbook example.

Unauthorised or undisclosed use of AI, where the regulations require you to declare AI use or prohibit it for a given assessment. Many universities have added specific generative AI provisions to their misconduct rules since 2023.

False citation or misrepresentation, meaning claiming to have read and relied on sources you did not consult, which is what happens when you cite something the AI produced without checking it exists.

The significance of this is that even where a university has not yet fully updated its rules to name “AI misuse” explicitly, fabricated references usually still breach existing rules on fabrication and false citation. As one institution’s guidance puts it, AI misuse often manifests in ways that mirror existing misconduct, such as made-up references and fabrication of results, so it can be dealt with under long-standing categories.

What the process typically looks like

1. A marker notices something wrong with the references, for example a source they cannot locate or that appears not to exist, and raises a concern.

2. The matter is referred into the academic integrity or misconduct process. This often begins with an academic conduct officer or panel reviewing the work and any supporting evidence, such as text-matching reports, the fabricated citations themselves, and sometimes metadata.

3. You are usually invited to respond, either in writing or at a meeting or hearing. You should normally be told the specific allegation and be given the chance to explain and to submit evidence, such as your notes, drafts, browser or library history and the genuine sources you actually used.

4. A decision is made on whether misconduct occurred and, if so, whether it is minor or major, often taking into account whether you have any previous offences.

5. A penalty is applied, and there is normally a right of appeal on defined grounds.

Likely penalties

Outcomes vary widely by institution and by the seriousness of the case. A realistic range, drawing on how universities describe their tariffs, includes:

For a first or minor offence, a requirement to resubmit with the mark capped at the pass mark, a reduced or capped mark, or a formal warning.

For a more serious or major offence, a mark of zero for the assessment or module, with the usual reassessment consequences, and the possibility that this causes you to fail the module or the year.

For the most serious cases, especially repeat offences or extensive fabrication, penalties can escalate to failure of the degree, suspension, or in the gravest cases termination of studies. Some universities set a very severe standard penalty for confirmed AI misuse, so it is important to read your own institution’s tariff rather than assume leniency.

Aggravating factors typically include the extent of the fabrication, whether it was deliberate, whether you denied it when confronted, and any prior findings. Mitigating factors can include genuine misunderstanding of the rules, early admission, a clean record, and personal circumstances.

Where innocent explanations matter

Not every incorrect reference is misconduct. It is worth distinguishing genuinely fabricated citations from ordinary referencing errors, because they can lead to very different outcomes.

A citation that is simply formatted wrongly, has a typo in a page number, or muddles a real source is usually a competence or accuracy problem, not misconduct. It might cost you marks for poor scholarship without triggering a misconduct finding.

A citation to a source that does not exist, or a quotation invented and attributed to a real author, is much harder to explain innocently and points towards fabrication, whether done by hand or generated by an AI tool and copied in unchecked.

The presence of several non-existent sources with plausible-looking but invented authors, titles and DOIs is a common indicator that AI-generated text was inserted without verification. This is exactly what many university panels now look for.

If your university actually permitted or required AI use for the assessment and you disclosed it, the framing changes: the issue may become whether you complied with the declared limits and whether you failed in your duty to verify sources, rather than a straightforward dishonesty allegation. Read the specific assessment brief carefully, because permitted AI use with a duty to check is very different from prohibited use.

If you are accused

Take it seriously and engage with the process rather than ignoring it.

1. Get and read the exact allegation and the specific regulation said to be breached, plus your institution’s academic misconduct procedure and penalty tariff.

2. Gather your evidence early: drafts showing how the work developed, your reading notes, library and database access records, browser history, and copies of the genuine sources you actually used. Evidence that you did real research and that the fabricated entries slipped in unchecked can significantly affect how the case is treated.

3. Prepare an honest account. Panels generally respond far better to candour than to implausible denials. If AI produced the references and you failed to check them, consider whether admitting the failure of verification, and showing you understand why it matters, is more credible than contesting the obvious.

4. Use your students’ union advice service. In almost every university this is free, independent and experienced in representing students at misconduct hearings, and they can attend with you.

5. Check the deadline and grounds for appeal if the finding goes against you. Appeals are usually limited to specific grounds such as procedural unfairness, new evidence that could not reasonably have been produced earlier, or a penalty that is disproportionate. You cannot normally appeal simply because you disagree with the outcome.

6. If you exhaust the internal process and remain dissatisfied, you can usually ask for a Completion of Procedures letter and take an unresolved complaint about the fairness of the process, though not the academic judgement itself, to the Office of the Independent Adjudicator for Higher Education, which covers most higher education providers in England and Wales.

Practical takeaway

The safest position is to treat any AI-generated reference as unverified until you have personally located the source, confirmed it exists, and checked that it says what you are citing it for. Fabricated references are one of the clearest triggers for a misconduct investigation precisely because they are easy to detect and hard to explain innocently. If you have already submitted work containing citations you did not verify, it is worth checking them now and taking advice from your students’ union about whether and how to raise it, because self-correction before detection is generally viewed far more favourably than being caught.

Because the labels, thresholds and penalties differ from one institution to another, the decisive document is your own university’s academic misconduct regulations and its specific policy on generative AI. Locate those two documents first, as they will tell you exactly which category your situation falls into and what range of outcomes you realistically face.