The last days of the take-home essay as we knew it
“Not to laugh, not to weep, but to understand”.
Leon Trotsky quoting Baruch Spinoza
A pupil during a math lesson in Finland (September 1956). Photo by Bror Brandt. Finnish Heritage Agency.
Metaphors are tricky, but sometimes they help. Let’s say you are at a restaurant with friends, you receive the bill, and you need to divide it by the number of people who ate. A few people might do the math in their head, but an overwhelming majority of us would take out our phone and use the calculator. I’m quite confident nobody would grab a piece of paper and do it by hand.
Now let’s imagine you are trying to teach a person how to divide. Drawing on their previous knowledge (adding, subtracting), you would explain the concept, give examples, guide their practice, help them work through problems. At a certain point, you’ll want to evaluate whether they have learned the skill. It would not occur to you to have them use a calculator. It also wouldn’t occur to you to give them a take-home evaluation, because then you wouldn’t be able to tell whether they used a calculator or not. The exam would be measuring their honesty rather than their math skills.
The take-home essay faces the same fundamental problem as that hypothetical take-home division test. For decades, the essay worked precisely because, in order to write it, students had to do the intellectual work that we wanted to assess: analysis, synthesis, argumentation. We wanted to improve their writing too, but for the most part, the written product was a way to test whether they had mastered these cognitive processes rather than their literary capacity. Now, with generative AI tools that can generate sophisticated-sounding prose, especially when given the right input, we can no longer assume that a polished essay represents the student’s achieving a certain intellectual and cognitive process. Just as we wouldn’t know if our math student used a calculator, we can’t tell if a history student used ChatGPT to analyse primary sources or if a literature student had Claude write their close reading of a poem. We simply can’t. Accepting this reality is the inescapable first step.
***
The second (and more intellectually challenging) step is to use this crisis to think more clearly about what we’re actually trying to accomplish in our teaching. I think it helps to frame this in terms of a crucial misalignment between learning objectives and evaluation.
Drawing on constructivist theory, Australian educator John B. Biggs developed the principle of “constructive alignment” to design teaching and learning activities in higher education. The core ideas are that learners construct meaning from what they do to learn, and that teachers must seek alignment between learning activities and learning outcomes. The principle of “backward design” is closely related. It proposes that when designing a course, first we need to be clear about our goals (what do we actually want students to learn?) and then devise evaluations (what counts as evidence that they’ve learned it?) and activities (how are they going to learn it?) that are aligned with those goals. People construct their own understanding, and for real learning to happen, our goals, our assessments, and our activities have to be pointing in the same direction.
Generative AI has not changed our objectives. We still want students to learn the crucial skills of a historian: how to analyse primary sources critically, to contextualize events within broader historical frameworks, to construct compelling arguments based on evidence, to synthesize information from multiple sources, to write clearly about complex ideas. We want them to develop the historian’s instinct for asking good questions, for spotting bias and perspective in sources, for understanding causation and change over time. We should not panic, because the crucial intellectual habits of our profession are alive and well, and will remain so. If anything, we might need to add new objectives (like how to use generative AI responsibly) but no fundamental skill is obsolete.
What generative AI does do is undermine how we’ve been measuring whether students have learned some of these skills. The take-home essay is no longer reliable proof that the learning happened. Generative AI tools strike directly at stage 2 of the backward design process (our “acceptable evidence”). We know what we want students to learn (stage 1), and we can still design activities to help them get there (stage 3), but we’ve lost confidence in our method of assessment.
***
To figure out what to do, we need to be more precise about when and how this crisis is hitting us. I think that the impact isn’t uniform across the curriculum. What’s needed is a closer look at constructive alignment, that is, being clear about our goals in the first place, and then thinking through which assessments actually measure them.
My impression is that in first-year courses at the bachelor level, and in survey courses in general, this is quite manageable. In most of these courses, we want students to grasp key historical concepts, understand major events and their chronology, learn important facts and figures, and begin to see how different periods and regions connect. An in-class written exam still works for measuring this kind of learning. An oral exam works too. There’s no need to reinvent the wheel here.
I’m also not so worried about advanced research courses, small-scale seminars, or thesis supervision. As teachers and supervisors, we can evaluate students’ development closely, working with them throughout the process, discussing drafts, helping reframing problems and questions, guiding their research choices. Our relationship and intellectual dialogue with them becomes part of the assessment.
The real challenge lies in between, in those intermediate courses for second- or third-year bachelor students, or early master's students, where the goals shift to process-heavy skills: analysis, synthesis, argumentation. These are the courses where students need to move toward thinking like historians, and in which the take-home essay or research paper was our preferred tool for building and assessing these skills. It’s precisely in these courses where the misalignment is more serious.
***
What to do? I don’t think we should abandon all take-home assignments and go back to 100% in-class, supervised exams. Rather, I think that oral defences of written work might need to become mandatory after any research paper or take-home essay. Students still benefit from the reflective thinking that comes with writing (and need to learn this skill as well) but then they have to stand behind their arguments, explain their reasoning, engage with unexpected remarks, respond to questions.
As a matter of fact, I believe we should rely more on oral components in general. They’ve been in the toolkit of universities for ages, and for good reason. Dialogue remains fundamental to any intellectual development. This doesn’t necessarily mean traditional oral exams, though those can work too. In seminars, flipped classroom activities (where students take the lead) should be promoted and can be weighted more heavily in our grading. Peer review sessions work as well: they create opportunities when students must engage substantively with classmates’ arguments. To participate meaningfully in these discussions, they have to demonstrate real comprehension. Having students record brief video reflections explaining their thinking at different stages of a project also works remarkably well. It’s like the take-home version of an oral presentation (of course, provided they don’t just read a script aloud). There's something about having to articulate ideas to a camera that reveals (or doesn’t reveal) authentic intellectual work.
These aren’t perfect solutions, to be sure, and I think we are all experimenting with what works best in different contexts. But what’s clear to me is that this “crisis” forces us back to fundamental questions about what we’re really trying to accomplish in the classroom. The take-home essay worked well for a long time, but the widespread availability of generative AI has brought that era to an end. Perhaps it can also create an opportunity to be more intentional about alignment between our goals, our methods, and our assessments. The only path forward is to be more creative and more willing to have honest conversations with our students about how learning actually happens.