Find the Gaps, Then Make It Yours

TL;DR: At NU London, first-year students across disciplines prompt Claude first, then revise its work. The revision, they discover, takes longer than writing from scratch and teaches more.

What she’s doing: Alison teaches International Relations in Practice, a required first-year course at NU London that also draws students as an NU Path elective. Many of her students are not politics or IR specialists; engineers, physicists, and mathematicians routinely populate the course alongside declared International Affairs majors. That disciplinary diversity shapes her approach to AI integration.

For a portfolio assessment spanning the semester, students select a topic from each major course unit, then prompt Claude to generate a 750-word response grounded in the required readings for that week, complete with illustrative examples and citations. The generated draft is the starting point for the assignment. Students must read the sources themselves, critically evaluate what Claude produced, revise the draft to make it their own, and submit both the original AI output and their revised version alongside a final reflective piece evaluating Claude’s usefulness as a tool for academic writing, disciplinary understanding, and critical thinking.

Alison is quick to acknowledge that the pedagogy followed the instinct rather than preceded it. Her general approach, she says, is: this sounds interesting, let’s try it. What she found looking back was that the assignment had inadvertently set in motion something she recognized: students developing critical thinking, genre fluency, and research literacy all at once.

What’s working and what isn’t: Students discovered quickly that Claude is not a shortcut. Because the task requires reading sources closely enough to catch errors, many reported that the activity took longer than simply writing the essay themselves would have. That was partly the point, Alison suggests. Students identified consistent limitations in Claude’s output: a tendency toward narrative description over analysis, a reliance on a narrow set of illustrative cases (Ukraine-Russia, Israel-Hamas, and US-China tensions appeared with notable frequency regardless of the question), and unreliable citation practices. These limitations became productive friction. Rather than obscuring what AI cannot do, the assignment made those gaps legible and actionable.

Students who were not disciplinary specialists found a particular benefit: Claude gave them a scaffold for academic essay structure that their home disciplines do not always emphasize, and they brought their own lenses to IR content in return. One student offered that the exercise was essential to understanding what a literature review actually looks like.

What didn’t work as intended was the open-ended format for showing revisions. Alison left the method deliberately flexible, wanting to accommodate different learning preferences rather than impose a single approach. In practice, some students submitted only their final draft, missing the required comparison component and rendering their assessment incomplete. She takes responsibility for the gap and plans to offer two or three concrete formatting examples as options going forward.

What’s next: Alison is redesigning the course around a semester-long simulation scenario. Students will represent different delegations navigating a shared international crisis, drafting negotiating strategies with AI support and then using AI to predict possible outcomes, before reflecting on where the model fell short. Her explicit interest is in what AI misses: its tendency toward confirmation bias, its limits as a cross-culturally competent interlocutor. She is also developing a custom tutoring chatbot to support a fully flipped classroom model in which students process theoretical content before class and use class time for application.

Have you tried something new with AI? How did it go? Send an email to [email protected] to let us know!