Academic integrity, honestly
The conversation about AI and cheating is unhelpfully shrill in both directions, and students are caught between an institution that mostly forbids and a peer group that mostly does not care. This lesson is the practical middle: what detection actually does, what disclosure is for, and what the real risk is — which is not the one most students worry about.
What AI detectors actually do
They do not detect AI. They estimate how predictable a piece of text is, on the theory that model-generated writing is smoother and less surprising than human writing.
That theory is partly true and it produces two failure modes that matter enormously to you.
False positives on clear, plain writing. Text that is simple, well-structured, and uses common phrasing scores as machine-like — because it is predictable. So does writing by people composing in a second language, who reasonably favour standard constructions. This is the documented, serious problem with these tools: they systematically flag non-native English writers at higher rates. If you write in your second language, you are exposed to a risk that has nothing to do with your conduct.
False negatives on lightly edited AI text. Changing some words and varying sentence lengths reduces the score substantially. So the tool is weakest precisely against deliberate evasion and strongest against honest plain prose, which is close to the opposite of what you would want.
Institutional practice has been moving accordingly — most now treat a detector score as grounds to ask a question, not as evidence. Turnitin and others have published their own cautions about using scores as proof.
What this means for you. Do not try to defeat detectors; it is the act that turns a policy question into a conduct one. And if you are ever flagged wrongly, the defence is process, not protest: your drafts, your version history, your notes, and your ability to talk about your own argument. Which brings us to the thing actually worth doing.
Keep your working, always
Write in something with version history — Google Docs, Word with autosave, anything that records the document growing. Keep your reading notes. Keep the outline.
This costs nothing, because it is a by-product of working normally. And it is a complete answer to any question about authorship, in a way no assertion can be. A document that grew over three weeks, alongside twelve notes and an outline, is not something anyone can seriously claim was generated.
Ananya's version history shows a draft that got worse before it got better, a paragraph rewritten five times, and a conclusion added at the end. That is what real writing looks like, and it is unfakeable.
Disclosure, and why it helps you
Where a statement of AI use is invited, write a specific one. Where it is not, write one anyway at the end of your notes.
Specific means naming the tool, the use, and the boundary — as in lesson 8's example. Vague disclosure ("AI was used to assist") is worse than none, because it signals awareness without accountability.
The reason to do this beyond compliance: a use you would rather not disclose is a use worth reconsidering. That test is fast, it is private, and it is almost always right.
The risk nobody talks about
The disciplinary risk is real and it is not the main one.
The main risk is that you get a degree and cannot do the thing the degree says you can do. You sit in an interview and cannot construct an argument under questioning. You start a job that requires reading something difficult and forming a view, and you have never actually done it without help. Nobody catches you. You simply are not able, and it took three years to arrange.
This is not a moralising point — it is the whole logic of lesson 1, stated as a consequence. The person who used AI heavily on tasks and rigorously not on skills comes out faster and more capable. The person who used it on everything comes out with the same transcript and less than they went in with.
That trade is invisible while you are making it, which is why it is worth deciding deliberately, once, at the start of a term rather than at 2am the night before a deadline.
Do this today: check that whatever you are writing in has version history switched on. Then write your two-sentence disclosure now, before the essay is finished — it is much easier to write while your choices are still ahead of you.