Since generative writing tools became widely available, universities have been scrambling to work out what to do about them. Students, meanwhile, are getting anxious about work they wrote themselves being flagged as machine-written.
Both worries are reasonable. Here is what AI detection at university actually involves, what the tools can and cannot tell, and what to do if you are accused unfairly.
Detection and plagiarism checking are not the same thing
This trips up almost everyone, so it is worth separating clearly.
Plagiarism checking compares your text against a database — published papers, websites, and previously submitted student work. It finds matches. When it flags something, there is a specific source it matched against, and you can look at it.
AI detection does nothing of the kind. There is no database of AI-written text to compare against, because the same prompt produces different output every time. Instead the tool analyses the statistical texture of your writing and estimates how likely it is that a machine produced it.
That is a prediction, not a match. It is the single most important thing to understand about these tools, and it is why the two produce such different kinds of evidence.
What the detectors are measuring
Without getting too technical, they are looking at how predictable your word choices are.
Language models generate text by repeatedly picking a likely next word. That tends to produce writing that is smooth, evenly paced, and statistically unsurprising. Human writing is lumpier. We use odd phrasings, vary sentence length unevenly, repeat ourselves, and occasionally choose a strange word for no good reason.
Detectors quantify that difference. Text that is unusually consistent and unsurprising scores as more likely to be machine-generated.
Turnitin, which most UK and Australian universities use, added an AI writing indicator in 2023. It reports an estimated percentage of the document it believes was AI-generated, and that figure appears to the marker but not to you.
How reliable is it?
Reliable enough to be used, not reliable enough to convict you on its own — and Turnitin says so itself. Its own guidance states the score should not be the sole basis for an academic misconduct decision, and that a human being needs to review the work.
The known problems are worth knowing about:
False positives happen. Genuinely human writing does get flagged, particularly when it is plain, formulaic, or heavily structured. Technical writing, methodology sections and literature reviews are more prone to it because that style is naturally repetitive.
Non-native English writers are disproportionately affected. Several studies have found this pattern. Writers using a smaller, more careful vocabulary produce exactly the statistical evenness detectors treat as suspicious. If English is not your first language, you are more likely to be flagged for writing carefully.
Some institutions have switched it off. A number of universities, including some in the US, disabled Turnitin’s AI indicator over accuracy concerns. Others use it only as a prompt to look more closely, never as evidence in itself.
The tools are gameable, which is precisely why nobody should treat a low score as proof of innocence either.
What your university’s policy probably says
Policies vary considerably, and yours is the one that matters. But most now sit in one of three positions:
- Prohibited — no generative tools at any stage of assessed work.
- Permitted with declaration — you may use them for specified purposes, such as brainstorming or grammar checking, provided you say so.
- Permitted for defined tasks — explicitly allowed for some things, banned for producing content.
Find your institution’s academic integrity policy and read the section on generative tools. It has almost certainly been rewritten in the last two years, and the version your friend describes may be out of date.
One rule holds almost everywhere: submitting machine-generated text as your own work is misconduct, regardless of whether any detector catches it.
If you have been flagged and you did write it
This is genuinely distressing, and it does happen. Practical steps:
Produce your drafting history. This is the strongest evidence available to you. Google Docs and Microsoft Word both keep version history showing the document being built over time. A machine-generated document has no such history. If you have it, lead with it.
Gather your working materials. Notes, annotated PDFs, library loans, browser history, supervision emails, photos of handwritten planning. All of it demonstrates process.
Ask what the evidence is. You are entitled to know. If the only evidence is a percentage from a detector, point out — politely — that the vendor’s own guidance says the score should not stand alone.
Offer to discuss the content. Someone who wrote a piece can explain why they structured it that way, which sources they rejected, and what they would change. That conversation is usually more persuasive than any document.
Use your students’ union. Most have advisers who handle academic misconduct cases regularly and know the procedure better than you do. This service is free and confidential, and using it is not an admission of anything.
How to reduce the risk in the first place
Keep everything. Write in a tool with version history rather than starting fresh in a blank document. Save your outline and your rough notes.
Beyond that, the writing habits that lower detector scores are the same ones that raise marks: cite specific sources with page numbers, refer to your own module and lectures, use concrete examples from your reading, and let your sentence lengths vary. Machine text is smooth and unspecific. Human academic writing is specific.
Be careful with grammar tools too. Heavy paraphrasing or rewriting features can make your prose statistically flatter, which is exactly what detectors respond to. Spelling and grammar correction is fine; “rewrite this paragraph” is where it gets risky.
Where we stand on this
Everything we produce is written by a person. We do not generate text and pass it off as written work, because it would fail our clients at exactly the moment it mattered and because it is the thing universities are actively screening for.
We also offer plagiarism and AI removal for students who have drafted something themselves and want it properly rewritten into original prose, and proofreading and editing for work that only needs tightening. For pieces you would rather hand over entirely, we cover essay writing and assignment writing across most subject areas.
Whatever you do, read your own institution’s policy first. It is the only rulebook that applies to you, and it is the one you will be judged against.