The release of ChatGPT was enough to significantly disrupt the education system.
Giving students access to such powerful tools with almost no guardrails was always going to create challenges. Essays, coursework and projects are becoming increasingly difficult for teachers to assess with confidence. The question is no longer simply, “Did the student submit good work?” It is:
“Does the student genuinely understand the work they submitted?”
As AI becomes more capable, distinguishing between a student's own understanding and content generated or heavily supported by an LLM becomes increasingly difficult.
So what solutions actually exist today?
The most obvious solution is traditional, supervised examinations.
Most powerful AI tools still rely on software and internet-connected devices. Remove access to those tools, place students in an exam hall and ask them to demonstrate what they know independently.
This gives teachers a much stronger indication of genuine understanding.
However, there is an obvious limitation: you cannot run an examination every week.
Doing so would consume significant teaching time, increase administrative workload and take time away from other important learning activities.
Schools therefore need another method of verifying understanding between formal examinations.
This is where oral assessment becomes increasingly important.
Oral assessment, sometimes referred to as an oral defence, is not a new concept. It has existed in education for a very long time.
However, in the age of generative AI, it may become one of the most effective ways of determining whether a student genuinely understands the work they have submitted.
Instead of assessing only the finished piece of work, we can assess the student's understanding behind it.
Even if schools decide that oral assessment is the right approach, another problem quickly appears:
Capacity.
Do we ask teachers to sit down individually with every student, every week, and question them about their work?
For most schools, that simply isn't realistic.
So how can oral assessment be delivered at scale?
One approach is allowing students to record themselves answering questions using voice or video.
This is a useful starting point because students have to articulate their understanding in their own words.
However, it still has limitations.
Students can prepare answers beforehand, use AI to generate scripts or repeatedly record responses until they produce the answer they want.
The assessment still isn't truly dynamic.
This is where platforms such as HeyCleo could fundamentally change how schools verify student understanding.
Imagine a student submitting their project, essay, coursework or assignment.
The system ingests the work and understands the arguments, concepts, evidence and citations contained within it.
The student then enters a real-time oral defence.
Instead of answering a fixed list of questions, they have a conversation about their own work.
They may be asked:
Why did you make this argument?
What does this citation actually demonstrate?
Can you explain this concept in simpler terms?
What would happen if the scenario changed?
Can you apply the same idea to a different situation?
Why did you choose this method instead of another?
Every important concept, claim and piece of evidence can be explored.
The purpose isn't necessarily to stop students from using advanced technology to support their assignments.
The goal is to ensure that the student remains in control of the work and can understand, explain, apply and defend what they have submitted.
The key is that the assessment is dynamic and real-time.
A student could take the same oral assessment five times and encounter different questions each time.
The system can generate variations, follow-up questions and applied scenarios based on the student's responses.
If a student demonstrates strong understanding, the conversation can move deeper.
If their answer exposes a gap in understanding, the system can probe that area further.
This makes memorising a set of AI-generated answers significantly less useful.
Much like bringing students into an examination hall and assessing them individually, the system can capture signals such as:
Importantly, these signals should not automatically determine whether a student has cheated.
Instead, they give educators additional evidence.
The teacher remains the final decision-maker.
AI may have made generating high-quality work easier than ever before.
The answer isn't necessarily to ban the technology.
The challenge for education is to move beyond asking “Who wrote this?” and towards asking a much more valuable question:
“Can this student demonstrate that they genuinely understand it?”