Using AI when your department does not want you to

History, English, Law and Philosophy run the strictest rules, and the reason is structural: in those subjects the writing is the assessment. That does not mean you cannot use AI. It means you use it somewhere else entirely. One line separates the two: if it touches the words that get marked, do not. If it touches your understanding, do.

Almost always fine#

Explaining a difficult passage to you until you understand it. Testing whether you have understood it, by being questioned. Finding what to read, which you then go and read yourself. Arguing against your position so you can find the holes. Rehearsing for a seminar or a viva out loud. Organising your own notes and your own reading.

Every one of those leaves the marked words untouched and improves what stands behind them. A tutor shown that list has no complaint available.

Where people actually get caught#

Drafting a paragraph and then editing it. It is still its paragraph.

"Improving" your prose. The most common one on the list, and the one that flattens your voice.

Generating a structure you then fill in. The structure was the argument.

Paraphrasing something to avoid quoting it. Translating your own writing out and back again. Anything you would not say out loud to your tutor.

None of that list is about dishonesty in the way students expect. Three of the six are things people do believing they are on the right side of the line, and the third is the one worth staring at: in an argumentative essay, deciding the order of the argument is the intellectual work. Handing that over and filling in the prose yourself has the ownership exactly backwards.

Why the strictest departments are strict, and why they are not being unreasonable#

In a subject where the assessment is a laboratory result or a proof, prose carries the finding. In History, English, Law and Philosophy the prose is the finding: the ordering of the argument, the weight given to a counter-example, the sentence that concedes a point without conceding the case. There is no assessable residue underneath the writing, because the writing is the residue.

There is measured reason for those departments to worry about the prose specifically. Across five real undergraduate modules at Reading, 94 per cent of wholly AI-written submissions went undetected and outscored real students by just over half a classification boundary. And on homogenisation, Standard American English retained 77.9 per cent of its features in a model's reply against 2 to 3 per cent for five minoritised varieties. A department protecting a distinctive voice is protecting something a model measurably flattens.

Using it on the understanding side is not a concession, it is the better use#

The evidence on learning points the same way as the rule. Among nearly a thousand school students given unrestricted access, a hints-only tutor, or nothing, the unrestricted group scored 17 per cent below students who never had the tool once it was withdrawn, while the tutor group kept most of its gain. The arm that answered questions damaged the learning; the arm that made students work kept it.

Being questioned by a model until you can defend a passage is the guardrailed arm. Asking it to write the paragraph is the unrestricted one. Your department's rule and the learning evidence are, for once, pointing in the same direction.

Getting the answer in writing#

Rules differ by institution and often by module inside one institution, and they are being rewritten yearly. So the useful move is to email your module lead and ask, in writing, so the answer exists.

Ask about the specific use rather than in general. "Am I allowed to use AI" invites a cautious no. "May I use a model to question me on the reading before I write, if no AI-generated text appears in the essay" invites an answer, and usually a yes.

Your department's rule is the one that governs#

It cannot tell you your department's rule and it is not a defence if you break one. The marked-words line is a way of thinking, not an institutional policy, and where a policy says something stricter, the policy governs.

The line is also blurrier than it looks in two places. A structure you argue with and then rebuild is not obviously the same as a structure you fill in. And a model that questions you can put a phrase in your head that you later write down believing it was yours. Nobody has measured how often that happens, and anyone who tells you where exactly the boundary sits is guessing.

Key sources

• Scarfe, P. et al. (2024). A real-world test of artificial intelligence infiltration of a university examinations system.
• Bastani, H. et al. (2025). Generative AI Without Guardrails Can Harm Learning. PNAS, 122(26).
• Fleisig, E. et al. (2024). Linguistic Bias in ChatGPT: Language Models Reinforce Dialect Discrimination. EMNLP 2024.
• Higher Education Policy Institute (2026). Student Generative AI Survey 2026.

Related SuperSkills research#

For students, how to use AI at university, how to be honest about it and how to handle forty readings. On the rule underneath, keep, share, hand over. On what the institutions say, official guidance on AI in education. On voice, keeping your own voice and whether AI makes everyone think alike.

About this research#

The marked-words line and the two lists above are used by Rahim Hirji in teaching and appear in the Mastering AI deck, most recently in September 2026. No claim of first use is made. Findings are attributed to the studies that produced them and kept separate from the interpretation, which is that the strictest departments are strict for a structural reason rather than a technophobic one.