Literature & Language Arts: How AI can best tutor this subject

What you'll be able to do: Use AI for the parts of this subject it's genuinely strong at, technique, structure, rival readings, while keeping the one thing it's fastest at, the interpretation, out of your hands until you've built it yourself.

The subject where the output and the deliverable are the same object

State this first, because everything else depends on it.

Ask an AI system to interpret the ending of The Great Gatsby, or to explicate Hopkins's "The Windhover", or to argue whether Frankenstein's creature is the novel's real monster, and it will produce, in seconds, a fluent, well-organised, technically literate piece of writing that could pass for a strong student essay.

In most subjects that's impressive but beside the point, because the assessed skill is something the AI's output doesn't resemble: solving the equation, not reading about equations; running the experiment, not reading about experiments. In literature, the AI's output IS a specimen of the exact skill being assessed. An essay interpreting a text is not evidence that you could write an essay interpreting a text, it is the thing itself, indistinguishable on the page from what you were supposed to produce.

That is a sharper trap than plagiarism, because you don't have to intend to cheat to fall into it. You can read the AI's reading, agree with it, absorb its vocabulary and its structure, and hand in something that is genuinely closer to your understanding than it was ten minutes earlier, and still not have done the thing the assignment exists to develop, which is the capacity to build a reading of a text yourself, under your own evidentiary pressure.

So the working rule for this subject is:

> Use it to interrogate, contest and refine. Never let it hand you the > reading.

What it is genuinely good at here

Naming technique. Meter, free indirect discourse, anaphora, caesura, enjambment, negative capability, unreliable narration, it has read enormously more criticism than you have and can name the device operating in a passage faster and more precisely than most students can. Used to spot patterns you then have to explain, this is real strength (article 06).

Building the reading you'd otherwise dismiss. Every strong interpretation has to survive its best rival. A tool with no investment in which reading wins, and that has absorbed a vast range of critical positions, is genuinely useful for constructing the version of the opposing case you were about to strawman (article 05).

Interrogating a text you have in front of you. Not summarising it, asking what it withholds, whose voice is absent, what it assumes you already know, what genre convention it is following or breaking. It generates the questions; you answer them against the actual words on the page (article 01).

Pressure-testing a generalisation. "This poet always uses X to mean Y" is a claim the text itself usually complicates somewhere. Finding the line that doesn't fit is close reading's actual discipline, and a tool that has the whole text available can hunt for it faster than you can reread (article 03).

Structural diagnosis of a draft. What claim is each paragraph actually making, on the page, not what you meant it to argue. The gap between the two is usually the essay's real problem, and it is the thing you cannot see in your own writing (article 09).

Turning "I liked it" into something you can defend. A gut response is real data, but it isn't yet an argument. Forcing it into a claim that specifies what evidence would support or contradict it is where the subject's actual work begins (article 02).

What it is bad at here, specifically

The exact words. Ask it to quote a passage and it will often produce something that is stylistically right and factually wrong, close paraphrase dressed as quotation, a line from the wrong chapter, a line from a different book by the same author. This gets worse, not better, the less famous the passage: it is fairly reliable on "Call me Ishmael" and considerably less reliable on a paragraph from Moby-Dick's chapter on cetology.

Plot detail on anything moderately obscure. Minor characters, the order of events in a subplot, which character says which line, it will answer confidently and can be wrong. Famous misattribution: "the lady doth protest too much, methinks" is Gertrude's line about the Player Queen, not Hamlet's about himself, and this is exactly the kind of confident, plausible, wrong attribution the tool reproduces without hesitation.

Which text it's actually describing. It conflates a novel with its film adaptation, an author's early work with their later work, one translation of a foreign-language text with another. If you haven't checked, you don't know which version you're getting.

Knowing a critic actually holds a position. It's reasonable on the broad shape of a critical debate and unreliable on who specifically argued what. Treat every "critics have argued that.." as something to verify before it goes in a bibliography (article 07).

Sitting with an ambiguous text. Plenty of literature is honestly irresolvable, the play doesn't tell you whether Nora was right to leave, the poem doesn't resolve which reading of the volta is correct. AI tends to produce a tidy, confident synthesis regardless, smoothing over exactly the tension a strong essay would sit inside.

The shape of a good session

1. Read the text first, before asking anything. There is no substitute step here; everything below depends on having actually been confused, moved or bored by the words themselves.
2. Interrogate the text with questions generated by the tool and answered by you against the passage (article 01).
3. Operationalise your first reaction into a claim specific enough to be wrong (article 02).
4. Hunt for the counterexample, the line in the text hardest to reconcile with your claim (article 03).
5. Surface what your reading assumes but never states (article 04).
6. Steelman the reading you're rejecting and answer its strongest point (article 05).
7. Write. Then reverse-outline to see what you actually argued, as opposed to what you meant to (article 09).
8. Verify every quotation and every critic attribution against a source you have personally checked before it enters the essay (article 07).

Notice where the tool is doing real work and where it is explicitly excluded. Step 1 is yours alone. Step 8 hands verification back to you even though the drafting used the tool throughout.

The instruction to set

For this session you are tutoring me in literature. Rules, never supply the interpretation; ask me questions that make me build it myself. Never invent or paraphrase a quotation as if it were exact, if you refer to a passage, tell me to find and paste the precise wording myself, and treat anything I haven't pasted as unverified. Don't tell me what a scene, poem or chapter "means": ask what evidence in the text would support or undermine a reading I propose. When you name a technique, ask me to find where else in the text it recurs, and what changes if I remove it. And if I tell you I "liked it" or it "felt sad", push me to say exactly what in the text produced that response before we go further.

The failure that looks like success

A fluent, structurally sound essay that identifies the right symbol, names the right technique, and states a defensible thesis, built by watching AI do it once and reproducing the moves, rather than by doing the moves yourself under the pressure of a text that didn't cooperate.

It often reads better than your unaided work. Every paragraph has a claim. Nothing is plot summary. And it proves nothing about whether you could produce that reading on a text you hadn't already seen interpreted, on your own, at speed, against a marker who asks a genuine follow-up question.

The tell: you can't point to the line in the text that gave you the idea. If the honest answer, for the essay's central claim, is "the AI said it and I agreed", the reading isn't yours yet, you've verified a conclusion, not reached one.

What it cannot replace

The actual encounter with the text: the confusion of a first read of Waiting for Godot, the slow accretion of noticing across three hundred pages that makes a pattern visible only because you were there for all of it, the specific discomfort of a passage that resists you. Close reading is a perceptual skill built by repeated contact with resistant material, and a tool that can produce the endpoint on demand teaches you to skip the contact that builds the perception.

Where to go next

01 Primary source interrogation is the direct fix for letting the tool supply the reading, it turns the text back into something you question rather than something you're told about. 05 Steelmanning rival readings is the single move most likely to raise a mark. 07 Verification drill is the one you cannot skip if any quotation from this session is going in front of a marker.