Using ChatGPT to check a dissertation for typos is almost certainly fine. Asking it whether your argument holds up is a different question, and at least one major university forbids that outright. Most advice online collapses the two into "using AI for editing" and then answers only the easy half.
Typos, Grammar, and Reasoning Are Three Different Questions
It helps to picture AI assistance as a ladder rather than an on/off switch. The bottom rung is spelling and punctuation. Next is grammar and syntax inside sentences you already wrote. Above that sits reordering your paragraphs, then feedback on whether your reasoning actually works, and at the top, sentences the model wrote that end up in your submission.
Nobody is anxious about the bottom rung, and no policy reviewed for this article treats it as misconduct. Oxford's guidance for postgraduate research students permits "local editing tools" including "grammar assistants" and "spell-checkers", and says explicitly that they "need not be declared". The stated reason is that such tools "only make small, local changes", usually affecting just a few words at a time.
The worry lives further up the ladder. Anyone searching for whether a reasoning check counts as cheating has usually worked out that it might be answered differently from the typo question. They are right.
What Dissertation Rules Say That Coursework Rules Don't
Dissertation and thesis regulations are consistently stricter than ordinary coursework rules, and that gap is where students get caught out. Oxford now requires research students to include a statement on their use of generative AI in the submitted thesis, placed immediately after the abstract, for submissions from Trinity Term 2026.
Toronto goes further. Its School of Graduate Studies states that "the use of generative AI tools in any aspect of researching or writing of the thesis must be done with the prior approval of the supervisor(s) and supervisory committee". Students must also cite and describe any use in the thesis, so that "it must be clear to the reader which generative AI tools were used, as well as how and why they were used", and must be able to defend that use at the final oral examination.
Read the order of operations there. Approval comes first, disclosure second. A student who ran a dissertation draft through ChatGPT intending to mention it in the acknowledgements has already missed a step.
Using ChatGPT to Check a Dissertation: Where the Line Actually Falls
Universities disagree about the middle rungs, and pretending they agree would be dishonest. UCL's Academic Manual permits "Language and Writing review", defined as having "a third-party or software check areas of academic writing such as structure, fluency, presentation, grammar, spelling, punctuation, and language translation". Structure is on the permitted list there. The ceiling is that it becomes misconduct "if substantive changes to content have been made by the reviewer or software or at their recommendation".
Edinburgh draws the line lower. Its rules forbid a proofreader from touching "the sequence of information or line of argument", giving "the order of sentences, paragraphs and sections" as the examples. What UCL allows, Edinburgh prohibits. Two well-regarded universities, opposite answers, which is why asking your own institution beats trusting a general rule.
The Proofreading Rule That Existed Before AI
Edinburgh's proofreading guidance is the most useful document in this debate, largely because it was written before anyone was panicking about chatbots. It applies to "a friend, family member, paid professional or company" and to "a variety of online proofreading tools" alike.
Direct editing is allowed "only to indicate and correct minor issues", where minor covers "spelling, punctuation, grammar and syntax" and is "localised, typically single words or punctuation items within existing sentences". Larger problems "can all be commented on, but not corrected by direct editing". A proofreader must never "rewrite any of the text", never "add new content", and never "check, correct or comment on the truth or accuracy of any factual content".
That gives you a test with nothing to do with AI. Is the tool pointing at a problem or solving it for you? A comment saying this paragraph does not follow from the last one leaves the thinking where it belongs. A rewritten, reordered section does not.
ChatGPT is awkward here for a mechanical reason. Ask whether your argument is sound and it will frequently hand back a rewritten argument rather than a comment on one. You asked for rung four and received rung five. Structural changes are safer made deliberately in your own words, and where language work genuinely needs a tool, an AI to human rewriter at least keeps that step visible and separate from your thinking.
If You've Already Done It
Many people reading this have already pasted something in. That is not misconduct on its own, and panic is not a useful response to it.
Work out which rung you actually used. Spelling and grammar on sentences you wrote sits inside what every policy examined here permits. If you took structural or argumentative suggestions, read your graduate school's dissertation regulations for a disclosure requirement, and disclose if there is one. Edinburgh advises students to "retain copies of their work directly before and immediately after third-party or online proofreader input", which is worth reconstructing from your file history if you can.
If the worry is phrasing rather than argument, because English is not your first language, that work sits on the lower rungs, and a human text rewriter applied only to your own sentences keeps it there.
Asking a supervisor is also less dangerous than students assume. They are fielding this question constantly at the moment, and a conversation you start is a very different one from a conversation they start. For the broader question of where AI use crosses into misconduct outside a thesis, the general case is covered separately.