AI literacy · Using AI well · Case study
Two essays, one exam
Josh and Amara use the same tool on the same task and get similar marks. Three weeks later, only one of them has anything to write.
Josh and Amara are in the same year 11 English class, studying Jasper Jones. The take home essay asks: how does the novel explore the gap between how Corrigan sees itself and how it actually behaves? Two weeks, 900 words, due Friday.
Josh opens a chatbot on Tuesday night, pastes the question in, and adds one line: write about 900 words. Forty seconds later he has an essay. It is fluent, correctly structured, and mentions Eliza, the town pool and the shed. He spends an hour swapping words so it sounds less polished, changes furthermore to also in three places, adds one sentence of his own about the ending, and submits. Total time with the novel itself: zero minutes.
Amara starts differently. She rereads her notes, drafts a rough outline on paper, and then opens the same chatbot. Her first prompt: here is my outline arguing that Corrigan's racism is shown through what characters do at the pool and after the disappearance, not through what anyone says. What is the weakest point in this outline, and what would a sceptical marker push back on? The model flags that her second point repeats her first in different clothes. It is right. She merges them and adds a new point about Jeffrey's cricket match.
Then she asks it to argue against her thesis entirely, as hard as it can. The counterargument it produces, that the novel shows individuals changing even while the town does not, is good enough that she gives it a paragraph in her essay and answers it. She writes every sentence of every draft herself, then pastes drafts back in with the prompt: which sentences are doing no work? She cuts eleven of them. It takes her most of the two weeks, in pieces.
One evening the model, asked for feedback, praises a quote she has used from the pool scene and suggests she analyse its final image more closely. Something about the wording nags at her, so she checks the quote against her copy of the novel and finds the model has subtly misremembered it: close enough to sound right, wrong enough to lose marks in an essay graded on textual accuracy. She fixes it, and from then on she checks every quote against the book. The model never once flags that it might be misquoting.
The essays come back. Josh gets a B. Amara gets a B plus. Josh points out, reasonably, that his Tuesday night beat her two weeks. On the numbers, he is not wrong.
It is worth being precise about what each mark measured. Amara's B plus measured Amara: her argument, her evidence, her sentences, stress tested but hers. Josh's B measured a language model's ability to produce a competent essay about a widely studied Australian novel, which was never in doubt, lightly disguised by an hour of word swapping. Two grades that sit side by side on a report, measuring two different authors.
Three weeks later is the end of term exam. Devices in bags at the front. The essay question turns the take home task on its head: some characters in Jasper Jones escape the town's way of seeing and some do not. What makes the difference? It is not the same question. It is the same territory, approached from a new angle, which is exactly what exam questions do.
Amara turns the paper over and finds she has an opinion. She has already argued about which characters change, three weeks ago, against a machine that pushed back. She knows her pool scene and the cricket match in detail because she chose them, defended them and cut sentences about them. Her plan takes four minutes and the essay pours out of the argument she has already had.
Josh turns the paper over and finds he has a memory of an essay. He can recall its shape, and phrases from it, but the question in front of him is angled differently and the shape does not fit. He never chose the evidence, so he cannot rearrange it. He knows the town pool matters but not exactly what happens there. He writes two pages of general statements about prejudice that could apply to any book, pads the ending, and leaves early. The mark that comes back is the honest measure of the three weeks between the two papers: what was built, and what was borrowed.
Nothing about the gap is permanent, which is the other half of the story. Josh did not lack ability. He skipped the build, and the build is repeatable: the next take home task is a chance to run Amara's process instead, rough thinking first, the model as critic and opponent, every quote checked against the page. The two weeks feel slower than the Tuesday night, and three weeks later they are not.
Your tasks
Work through these in order, on paper or in a doc. They are the point of the story.
- 1List every step in Amara's process and label each one with the job the AI did: critic, opponent, editor or quiz master. Whose ideas was it working on at each step?
- 2Josh spent an hour rewording the generated essay. Explain why that hour produced almost no learning, using the first draft trap from lesson two.
- 3Apply the exam room test from lesson three to both students: for each, what did their AI use leave in their head that was available when devices went into bags?
- 4The exam question was a variation, not a repeat. Explain why variations punish borrowed work harder than repeats would, and what that means for how you prepare.
- 5Amara asked the model to argue against her thesis as hard as it could. Write the prompt you would use to do the same for an essay you are working on now, following the prompting lesson: context, precise ask, format.
- 6Amara caught the model misquoting the novel. Using the checking outputs lesson, explain why a model is more likely to misquote a text slightly than wildly, and why the slight version is the more dangerous of the two.
- 7Redesign Josh's Tuesday night so the tool builds his understanding instead of replacing it. Keep it realistic: he has limited time and low motivation. Specify at least four prompts he would send and what he must do himself between them.