The Essay Nobody Can Explain

A student hands in a flawless essay. Ask her to explain the main argument and she can’t. Ten years ago you’d have started a plagiarism hearing. These days you should probably start by looking at the assignment.

That’s the uncomfortable spot a lot of universities are in this year. The Higher Education Policy Institute surveyed 1,054 full-time UK undergraduates for its 2026 report with Kortext, and the headline is hard to argue with. Some 95% use AI in some form, and 94% use generative AI to help with assessed work. That’s not a habit among early adopters. That’s nearly everyone.

The Missing Rulebook

What’s missing is the rulebook. Plenty of students told the survey that their universities have been slow to offer guidance, tools or proper support, and fewer than half feel their teaching staff can help them with AI. Some worry about being accused of misconduct for doing something they weren’t sure was allowed. Fair enough. Ask a nineteen year old whether using AI to critique a draft counts as cheating, and if the honest answer from the institution is “it depends who marks it”, you’ve set a trap.

You can see why two reactions have taken over. Ban it, say some. That’s easy to write down and close to impossible to police, and students simply stop telling you. Embrace it, say others, because everyone’s doing it anyway. That sounds modern, but you can end up with shiny new tools bolted onto the same old coursework.

What Was the Assignment Actually For?

A third question tends to get skipped: what was the assignment actually for? If it was meant to show that a student can think through a problem, weigh the evidence and say something clear at the end, then the final document was only ever a stand-in for that. A decent one, for a long time. Less so now.

Teachers who’ve thought this through tend to collect different evidence. Drafts, so you can watch an argument grow. A short note on why the student chose one route over another. A five minute conversation, which often tells you more than a marking scheme does. It’s closer to how professionals get judged anyway. A surgeon explains a decision. An engineer defends a design. Nobody asks them to hand over a polished PDF and leave the room.

The Stakes of Critical Thinking

Students feel the stakes too. A RAND study, cited in Fortune this summer, found that most students who use AI for homework worry it’s hurting their ability to think for themselves. Gartner has predicted that through 2026, falling critical thinking skills will lead half of global organisations to ask candidates for AI free skills assessments. Forecasts are only forecasts, but employers clearly share the unease.

So what does better teaching look like? Educators working on this say it means moving from “here’s how to use the tool” to “here’s how to question it”. Give students messy problems where AI can help but can’t simply hand over the answer. Let them run into conflicting evidence and ethical grey areas. Let them find out, somewhere safe, that a system can sound completely sure and still be wrong. They can use AI to brainstorm, to poke holes in a draft, to explain a tricky idea. Then they have to judge what comes back and decide for themselves.

Engagement vs. Lasting Understanding

Games offer a useful parallel. They work well when the thing to be learned is a system or a decision, because you can plan, see what happens, adjust and try again without real world damage. But a beautifully designed game can also create excitement without lasting understanding, and sometimes a good discussion or a piece of writing teaches the idea better. Engagement and learning aren’t the same thing. AI tools in the classroom deserve the same scepticism: ask for evidence that students understood more, not that they enjoyed it.

It’s partly a governance question as well. Responsible AI isn’t something you sprinkle on after a system has been bought. It belongs in how a university designs courses, picks software and marks work from the beginning. And it’s a staffing question. If fewer than half of students think their lecturers can guide them, telling students to work it out alone isn’t a plan. Training, shared examples and time to redesign assessments cost far less than a degree nobody trusts.

Beyond the Computer Science Department

Looking a few years ahead, many educators expect AI to stop being a subject studied on its own and to become part of almost every profession, from medicine and law to design and engineering. If that’s right, the assessment problem won’t stay in computer science departments. Every discipline will need to decide what understanding looks like when a capable assistant sits in every student’s pocket.

None of this is simple. Oral assessments take time. Staff are stretched. Big cohorts don’t lend themselves to five minute chats. But the alternative is carrying on as if the essay still means what it used to mean.

The students have already walked through the door. The real decision for universities is what they want a degree to prove. If it’s understanding, the evidence has to change. If it’s judgment, they’ll have to build it on purpose, in classrooms where students are allowed to struggle, argue and explain themselves out loud.