AI in Education: Customising the Learning Curve

Personalised tuition has long been associated with better performance in education at every level. The specific needs of learners differ vastly, with some traditional approaches to revision prioritising rote memorisation and recall over true conceptual understanding. Access to one-to-one tuition, however, has largely been limited to higher-income families, or those nearest to big cities. This creates a structural inequality in access to meaningful educational reinforcement. 

The conversation around AI in education has been focused on misuse and overreliance. But that is the wrong frame. The more important question is what kind of AI they are being given. With the clear benefit of personalised tuition, and the disparity in access to it, the debate should now shift to how responsibly built AI learning support platforms can democratise access to personalised learning. The Department for Education recently proposed that over 450,000 disadvantaged students could benefit from AI tutoring tools. 

Exam prep and AI use today 

To understand where AI can help, it is worth looking at how students prepared before these tools became widely available. If a student could not access a tutor, the methods of extra-curricular exam preparation were mainly limited to textbooks, past papers, study groups and flashcards. These methods, without proper reflection, often rely on rote memorisation: repetition to improve recall without true conceptual understanding. This may be fine for times tables, or when there is only one right answer, but when exams require critical thinking and subject expertise, they begin to fall short. 

Personal tutors, much like teachers in schools and universities, take a more pedagogical approach. They focus less on delivering answers and more on structuring how students arrive at them through questioning and feedback. That is a large part of why they are effective. It is also why unequal access to them matters. 

AI tools are now increasingly being integrated into students’ study arsenals, whether institutions are comfortable with that or not. However, AI use remains fragmented: some students use it to support their learning, others rely on it for shortcuts and answer generation, while some avoid it entirely. Widely varying institutional policies are contributing to this. Many have opted for flat-out bans, but these do not eliminate use. They simply push it into unstructured environments, where the chance of misuse is higher. 

The debate should therefore move on from whether students should use AI at all. The real issue is how students can be guided towards structured, pedagogical learning methods, and how AI can be used to encourage deep understanding rather than superficial output. 

From answer generation to systems of teaching 

That distinction matters because not all AI tools are designed in the same way. When AI is structured around teaching rather than output generation, it shifts students from passive consumption to active problem-solving. Trained on curricula and mark schemes, with outputs reviewed by teachers and examiners, these pedagogically led platforms can encourage the articulation of reasoning by querying incorrect answers and building gradual understanding of topics. 

By tracking this understanding, alongside engagement with the platform and progress through the curriculum, pedagogical AI platforms are able to adapt to each student’s style of exam preparation and study support. The right platforms can adjust how quickly and how deeply they move through material, provide more guidance when a student is struggling or disengaged, and help reduce stress by breaking tasks down. In other words, they can begin to replicate some of the benefits of one-to-one tuition, but without the prohibitive and unequal cost barrier. 

Generalist AI tools will continue to have a role in the world as the technology develops. But in education, it is imperative that students have access to that technology in its most constructive form. They need systems that are built to teach, not simply to generate answers. 

The misframing of AI in education 

This is where the public conversation often goes wrong. General-purpose models currently widely used by students are designed to give general answers. There are no built-in structures to encourage retention or understanding, despite students often looking for exactly that in the platforms they use. 

The idea that students mainly want AI to do the work for them is not supported by what happens when the tool is built differently. When students are provided with an AI platform that focuses on pedagogy-led systems of teaching, 84% of messages are explanation-seeking, while only 4% attempt full-answer generation, data from Medly’s platform shows. This suggests that students, when presented with the right tools, prefer to leverage AI to understand concepts and receive personalised feedback – rather than, as the public debate often suggests, to bypass effort or simply generate answers. 

Usage patterns further reinforce this: around 8% of all messages are sent between 10pm and 3am. This is when students are studying independently, teachers are unavailable and support is limited. 

These behaviours, in the context of the right tools being used, show structural academic pressures, not opportunistic misuse. The narrative that AI drives misconduct should be challenged. As with any new tool, behaviour and structure decide the outcome. 

Democratising personalised learning 

That is why this matters beyond the immediate debate about cheating or classroom policy. Responsibly designed, pedagogically led AI tools have the potential to make personalised tuition accessible at scale, across geographies and incomes, reducing longstanding inequities. They can also encourage habits of reasoning, reflection and critical thinking, rather than dependence on first answers. 

Beyond education, platforms that query answers and encourage critical thinking can help equip students with the judgement to work successfully alongside AI in the future. That will increasingly be expected in the workplace: the ability to look past the first answer given and exercise independent analysis. The way students access knowledge is changing, and education systems will have to adapt accordingly. 

AI in education should not be judged solely by the worst behaviour enabled by generic tools. It should be judged by whether it helps more students think, understand and learn well. If it does that, the real risk is not adopting it too quickly, but dismissing it too broadly.