There is probably no phrase in educational technology that commands more effortless agreement than “personalised learning”. Ask anyone — a vendor, a policymaker, a well-meaning dean — what AI’s most exciting promise for education is, and personalisation will be near the top of the list, if not the first thing out of their mouth. It has become, in a way, the organising vision for AI in education: the idea that we can finally move beyond the one-size-fits-all classroom and give every student a learning experience that is genuinely theirs.

What is worth sitting with, though, is what we actually mean when we say that.

The question nobody is quite asking

A recent piece from UNESCO puts it bluntly: personalised learning, as typically implemented through AI systems, is less bespoke tailoring and more algorithmic triage. The system identifies what a learner cannot yet do, and routes them towards content designed to close that gap. It is remediation dressed in the language of individualisation. That is not without value, but it is a very specific thing, and calling it “personalised” papers over a set of questions we have not seriously answered. Personalised towards what end? Personalised according to whose model of the learner? And, perhaps most importantly, personalised in a way that reflects the learner’s own understanding of what they need — or merely what the algorithm infers from their click patterns and response times?

A systematic literature review of personalised learning terminology makes this confusion concrete: the field has been using the term for decades across wildly different contexts, sometimes meaning differentiated instruction, sometimes adaptive content delivery, sometimes student-directed inquiry. The word has expanded to cover so much that it has started to lose its grip on anything in particular. That would be a manageable problem in a theoretical debate. It becomes a more pressing one when institutions are making significant curriculum and procurement decisions on the basis of it.

What the system optimises for

There is a deeper issue worth naming here. Adaptive learning platforms — and increasingly, AI tutors — are built to optimise for measurable outcomes. That is not a flaw in their design; it is the design. They track progress towards defined competencies, adjust the pace and sequence of content, and surface interventions when a learner deviates from an expected trajectory. Within those parameters, they can be genuinely impressive.

But the parameters matter. A system that personalises the path to a predefined destination is not personalising the destination itself. It is, in that sense, a highly efficient form of convergence — shepherding learners more smoothly towards outcomes that someone else has decided are the right ones. That is a legitimate educational activity, but it is worth being honest that it is not quite what the word “personal” suggests to most people who use it.

The more interesting design question — one that adaptive systems have not yet answered, and may not be structured to answer — is what it would mean to personalise the goals of learning, not just the route. That would require a different kind of relationship between the system, the learner, and the educator. It would require space for uncertainty, for digression, for the kind of productive wandering that does not optimise well but sometimes produces the most durable understanding.

Where the educator comes back in

None of this is an argument against using adaptive tools. It is an argument against the comfortable assumption that deploying them is, by definition, a move towards more meaningful, learner-centred education. The technology can do some things very well. Whether those things amount to genuine personalisation depends on decisions that sit outside the technology: what competencies are being targeted, how much latitude learners have to shape their own trajectory, and whether the educator is in a position to use what the system reveals about a learner in ways that go beyond re-routing them through more content.

DigComp 3.0, the European Commission’s updated digital competence framework, points in a related direction. Digital competence is not just the ability to use tools effectively — it includes the capacity to critically evaluate them, to understand their limitations, and to make informed choices about when and how to engage with them. That applies to learners using AI tutors, but it applies equally to educators and institutions adopting adaptive platforms. Knowing what a system is actually doing when it claims to personalise is itself a form of digital literacy the sector has not quite caught up with.

A more honest framing

The point is not that personalised learning is a bad idea. It is that the term is doing a lot of work it has not earned. When “personalised” means “the system adjusts the difficulty of the next question based on your last answer”, that is a meaningful but narrow form of adaptation. When it means “the learning experience reflects who you are, what you care about, and where you want to go”, that is something else entirely — and it is not something any current AI system delivers on its own.

The distinction matters because it shapes what we ask of the technology, what we ask of educators, and what we ask of students. If we assume that personalisation is largely a solved problem once an adaptive platform is in place, we are likely to underinvest in the things that actually make learning personal: relationships, trust, purpose, and the space to be uncertain. Those are not features that scale automatically. They require deliberate design — and, usually, another human in the room.

References and further reading:

  • Shemshack, A., & Spector, J. M. (2020). A systematic literature review of personalized
    learning terms. Smart Learning Environments, 7, 17. https://doi.org/10.1186/s40561-020-00140-9
  • Farthing, B. (2025). Bespoke or prescribed? The myth of personalised learning. UNESCO.
    https://www.unesco.org/en/articles/bespoke-or-prescribed-myth-personalised-learning
  • European Commission, Joint Research Centre; Cosgrove, J.; Cachia, R. (2025).
    DigComp 3.0: European Digital Competence Framework. Publications Office of the European Union.