The essay subjects, and what a marker is reading for

The common structure under law, philosophy, politics, sociology

Essay subjects differ in content and agree almost entirely on what earns marks. A good essay makes a contestable claim, defends it with evidence somebody could check, deals with the strongest objection to it honestly, and does so in a structure where each paragraph moves the argument forward.

Generated essays fail on all four in a recognisable way, and it is worth knowing the failure precisely, because it tells you exactly which parts of your own work you must own.

The claim is not contestable. Generated theses tend towards the balanced and the safe: the causes were both economic and political, the answer depends on context. True, unarguable, and unmarkable. A marker cannot reward a position nobody would dispute.

The evidence is generic. Named sources without page numbers, examples chosen because they are the famous ones, no engagement with a specific passage or dataset.

The objection is a straw version. Some might argue X, however this overlooks Y: where X is stated so weakly that dismissing it costs nothing. Genuine objection handling states the other side at its strongest, which is uncomfortable, which is why it is rewarded.

The structure is smooth and static. Paragraphs of even length, each on a sub-topic, none of which depends on the one before. A real argument has paragraphs that could not be reordered.

Where it helps, precisely

• Sharpening a thesis. Here is my claim. Make it more contestable without making it indefensible. What is the strongest version somebody could disagree with?
• Stress-testing. The sceptic prompt, in a fresh chat. This is the single best use in essay subjects.
• Mapping the debate before you read, so you know which positions exist. Then read the actual sources; every name it gives you is unverified.
• Structural diagnosis of your draft. Which paragraph does not advance the argument? Where is my evidence thinnest? Which sentence is doing more work than it can support? Do not rewrite.
• Explaining a difficult primary text, grounded on the passage.

The legal-subject caution

Law students have a specific hazard. Fabricated case citations are the highest-profile AI failure in the whole professional world: lawyers in several countries have been sanctioned for filing briefs with invented cases, complete with plausible names, reporters and paragraph numbers.

And law is jurisdictional. A confident statement of the law is meaningless without knowing whether it is describing your jurisdiction, and models blend them constantly. A rule from one country’s statute presented as general principle will be wrong for you in ways that are not visible from the text.

Every case, statute and section gets checked against the primary source. Free sources exist: BAILII for UK and Irish material, CourtListener and Google Scholar’s case law for US, Indian Kanoon for India, AustLII for Australia, CanLII for Canada, and official legislation sites in most countries. There is no excuse for an unverified citation and no defence available if one is found.

Nothing here is legal advice, and the same rule applies to your own life: a model’s account of your rights is a starting point for a question to a real adviser, never an answer.

Philosophy, and the risk of plausible argument

Philosophy is unusual: the subject matter is argument itself, so generated text that resembles argument is maximally camouflaged. A generated paragraph will move confidently from premise to conclusion with a step that does not follow, and it will read like philosophy because philosophy reads like that.

The defence is the one from the proofs lesson, applied to prose: set out the argument as numbered premises and a conclusion, then ask what licenses each move. Also, philosophers are precise about what a named position actually claims, and models blur positions towards their popular versions. Check the primary text: most canonical philosophy is out of copyright and free at Project Gutenberg or the Stanford Encyclopedia of Philosophy, which is free, peer-reviewed and written by specialists, and is a better first stop than any assistant.

The thing that cannot be delegated

Every essay subject rewards the same rare thing: a specific, defensible claim that this particular student arrived at by engaging with particular evidence. There is no way to obtain that from a system that produces the average of what has been written. The average is exactly what a mid-band essay already is.

The practical consequence is that the outline is the essay. If the outline (the claim, the order of moves, the evidence chosen) is yours, everything downstream can be assisted and the work is still yours. If the outline came from a model, no amount of rewriting makes it yours, and it will read like the average, because it is.