In December 2022 a magazine article announced that the college essay was dead (Marche, 2022). The occasion was the public release of a conversational program able to produce, in seconds, several hundred words of orderly prose on almost any topic a teacher might set. The technology behind it, large language models built on the transformer architecture and trained to predict the next word across enormous bodies of text, had been developing for some years. What changed was access: a student no longer needed skill or money to use it.

Humanities teachers felt the shock most because their teaching leans hardest on the essay. A history, literature or philosophy course typically asks students to read, to form a view and to defend it in continuous prose written outside class. That exercise is how the disciplines have both taught and examined. This article argues that the two functions have now come apart, that the right response is to hold on to the first and rebuild the second, and that abandoning the essay would give up the thing most worth keeping.

The word comes from Montaigne, who called his writings attempts or trials, and who presented them as a record of a mind testing itself against a subject (Montaigne, 1958). Adorno (1984) defended the form on similar grounds: the essay proceeds without a fixed method, stays close to its object and resists the demand that thought arrive at a system. On this understanding an essay is not a container into which finished conclusions are poured. It is the place where they are reached.

Research on composition gave the idea an empirical footing. Emig (1977) argued that writing is a distinctive mode of learning because it is slow, leaves a visible record, and obliges the writer to connect and order ideas. Flower and Hayes (1981) described composing as a recursive process of planning, translating thought into language and reviewing, in which goals are revised as the text develops. Bereiter and Scardamalia (1987) distinguished novice writers, who tell what they know in the order it comes to mind, from experienced writers, whose struggle to fit content to a rhetorical purpose transforms what they know. The transforming kind of writing is what humanities teaching hopes to induce.

The essay set as coursework always did a second job. It gave the teacher an object to grade and the institution evidence that learning had occurred. The two jobs were held together by an assumption so obvious that it went unstated: a competent essay could only come from a student who had done the thinking.

That assumption has failed. A language model produces text by estimating which words are likely to follow which, and Bender et al. (2021) stressed that such a system manipulates linguistic form without access to meaning in the way a speaker has it. The output is often accurate and sometimes invented, and the system does not track the difference. Hicks et al. (2024) proposed that the right description is the one Frankfurt (2005) gave of speech produced without concern for truth, as distinct from lying, which requires knowing the truth and concealing it.

For teaching, the important fact is simpler than any account of how the tools work. Fluent, organised, grammatical prose on a set topic is no longer evidence of anything about the person who submits it. The product has been decoupled from the process. Everything else follows from how teachers respond to that decoupling.

The position taken here is that essay writing should remain at the centre of humanities education, because its educational value lay in the process and the process is untouched, and that the take-home essay should cease to be the main instrument of assessment, because its evidential value lay in the product and that is gone. Table 1 separates the functions the essay has served and indicates how each stands.

The practical consequence is a shift in what teachers look at. Figure 1 arranges kinds of written work by two properties: how far the work relies on a generative tool, and how visible the student's own reasoning is to the teacher. The conventional take-home essay sits in the lower left. It relies on no tool and shows no reasoning beyond the final text, and it is therefore indistinguishable, from where the marker sits, from the outsourced essay in the lower right. The upper row is where teaching can still see thinking.

The upper left of the figure describes a familiar practice: writing done in stages, with plans, drafts and revisions discussed in class and the finished piece defended in conversation. The upper right describes a newer one, in which students are given or generate a machine draft and are assessed on what they do with it: where they find it thin, wrong or evasive, and how they would argue differently. Both keep the essay. Neither treats the final document as sufficient evidence.

The argument meets objections from opposite directions.

Several implications follow. More writing should happen where it can be seen: in class, by hand or on screen, in short sustained periods, with the teacher present as reader and not as invigilator. Assessment should attach to the stages of work, to notes on reading, outlines, drafts and responses to criticism, so that what is graded is the development of a position. This is an application of the principle that assessment tasks should require the very activity the course intends students to learn (Biggs, 1996). Oral examination, long standard in some university traditions, deserves wider use beside written work.

Students also need to be told why the effort matters. Research on learning indicates that conditions which make practice harder and slower often produce more durable learning, a finding summarised under the name of desirable difficulties (Bjork and Bjork, 2011). A tool that removes the difficulty of composing removes the occasion for that learning, and students who understand this are better placed to decide when to use it. Finally, the tools themselves belong in the syllabus as objects of criticism. Reading a machine's account of a poem against the poem is a good exercise in close reading, and it trains the scrutiny of plausible prose that Nussbaum (2010) placed among the capacities a democracy needs from the humanities.

None of this is free. Seen writing takes class time from content. Staged assessment and oral defence are costly in large courses and for teachers on insecure contracts. In-class writing can disadvantage students who need longer, including those working in a second language, and adjustments must be designed in from the start. These are reasons to argue for resources, and they are not reasons to go on grading documents whose origin cannot be known.

Generative text tools have ended the essay as a self-certifying product. They have not ended the essay as a practice, and the argument of this article is that the practice was always the point. A classroom that brings writing into view, grades the path as well as the destination, and teaches students to read machine prose critically can keep what Montaigne meant by an essay. The future of the form depends less on the tools than on whether institutions are willing to pay for the slower kinds of teaching that make thinking visible.

To be declared by the author(s) before publication.

To be declared by the author(s) before publication.