Assessment in the Age of AI

When Google arrived in classrooms, teachers panicked. Students could find any fact, copy any explanation, paste any answer into an assignment and submit it as their own work, and what became known at the time as "copy and paste syndrome" was so widespread and so well documented that it was being written about in education circles as early as 2007. Sound familiar? The anxiety we are living through now about artificial intelligence and academic integrity is not new, it is just wearing more sophisticated clothes, and the answer, as it turns out, is exactly the same as it was then: not better detection, but better design.

I was teaching during those years, and Google was the problem we were all trying to solve. My response was not to ban search engines, though those conversations were certainly happening in staffrooms at the time, but to ask myself a more useful question: what kind of assessment could I design that Google simply could not answer for a student? The answer led me toward a way of thinking about assessment that has stayed with me ever since, and that holds up just as well against AI as it did against a search engine.

The question was never really about technology. It was about what we are actually asking students to do.

Start with what you are trying to assess

When teachers ask me whether something should be a quiz or an assignment, my question back is always the same: what are you trying to assess? Because the format should follow the purpose, not the other way around, and yet in online teaching it is surprisingly easy to let the platform menu drive the decision rather than your own pedagogical thinking.

Quizzes work beautifully for foundational knowledge, terminology, dates, formulas, procedures, the kinds of things where a right answer exists and immediate feedback is genuinely useful to the learner. But if what you are trying to assess is whether a student can apply a concept to a real situation, analyse a complex problem, create something meaningful, or demonstrate a process of thinking, then a quiz is the wrong tool entirely, and an essay is often not much better.

The multiple choice test is particularly tempting in online teaching because it is easy to set up and easy to mark, and I understand that appeal completely when you are managing a large cohort without the support structures of a face-to-face institution. But multiple choice tests are also the easiest assessments in the world to game, with or without AI, because they require recognition rather than understanding, and recognition can be faked.

The constructivist argument

There is a particular approach to learning that I have always found compelling, and that shaped every assessment decision I made when I was teaching. Constructivism, in simple terms, holds that students learn best not by receiving information passively but by actively building knowledge through experience, and that the most meaningful learning happens when students are required to construct something real and useful rather than simply demonstrate that they have absorbed content.

When assessment is designed from this starting point, something interesting happens. The tasks that emerge naturally from constructivist thinking are also, almost by definition, the tasks that are hardest to fake. Not because they are designed to catch cheaters, but because they require something deeply personal: the student's own context, their own thinking process, their own genuine engagement with real material over time.

A student can ask AI to write an essay on constructivism. They cannot ask AI to document their own learning journey, apply a framework to their specific teaching context, record a screen demonstration of their own work in progress, or reflect authentically on how their thinking has shifted over the course of a semester. The assessment design makes the shortcut irrelevant, not because it is policed, but because the shortcut does not produce the thing being asked for.

Building assessment over time

One of the most effective strategies I have used is building a major assessment piece through smaller stages across the semester rather than setting one large task due at the end. This does several things at once. It forces students to begin engaging with the material early rather than cramming at the last minute. It allows you to provide feedback along the way so that the final product is genuinely better than anything produced in a single sitting. And it makes the assessment significantly harder to fake, because each stage builds visibly on the one before it and is anchored in work the student has already submitted and discussed.

When I was teaching pre-service teachers, I designed a semester-long project that was built progressively through a series of smaller assessments, each one contributing to a larger whole. Students were not completing disconnected tasks for points, they were building something meaningful over time, and that sense of progression and purpose changed the quality of engagement entirely. The work grew with the student, and because each stage was visible and connected to what came before, there was simply no seam where something borrowed or generated could be slipped in without it showing. The final piece of the puzzle was a presentation to the class where students had to talk through their project and answer questions from their peers and from me, and that moment of live conversation was the ultimate test of genuine engagement, because you cannot present work you do not truly own and answer unexpected questions about it convincingly.

Building in feedback

There is another advantage to this approach that is easy to overlook, and that is the opportunity it creates for meaningful feedback. When assessment is built in stages, feedback given at each checkpoint can be seen in action in the next submission, and that visibility is extraordinarily valuable, because a student who has genuinely engaged with your feedback will show it in their work, and a student who has not will show that too. A single final submission tells you what a student produced. A staged assessment tells you how they learned, and that is a far more interesting and useful thing to know.

And what about AI?

I want to be honest: no assessment design is completely immune to academic dishonesty, and it would be naive to pretend otherwise. A determined student with enough time and ingenuity can find ways around almost anything. But the question is not whether cheating is theoretically possible. The question is whether your assessment design makes cheating more effort than it is worth, and whether it requires something that no tool, whether Google or AI or anything that comes after, can genuinely provide on a student's behalf.

When assessment is built over time, anchored in personal context, requires demonstrated process rather than just a final product, and produces something the student themselves will use, the answer to that question is yes. Not because the design is policing behaviour, but because the design is asking for something authentic, and authenticity, by definition, cannot be outsourced.

The anxiety about AI in assessment is understandable, and I do not dismiss it. But the teachers I have seen respond most effectively to it are not the ones investing in detection software, though that has its place, they are the ones going back to first principles and asking: what do I actually want my students to learn, and how do I design an assessment that requires them to genuinely do that? It is also worth remembering that every technology that has arrived in education trailing anxiety behind it has eventually found its positive place in the learning environment, and AI will be no different. The question is not whether to engage with it but how, and educators who approach it with the same curiosity and critical thinking they bring to any new tool will be far better placed to use it well, and to help their students use it well too. That question about design led somewhere useful when Google was the problem, and it leads somewhere just as useful now.

Happy Teaching,
Grazia