Against “Against Personalised Learning” (Part 1)

Personalisation vs Belonging

In her essay Against Personalised Learning, education scholar Caroline Pelletier argues that much of the enthusiasm for “personalised” education is driven less by pedagogy and more by the logics of markets, metrics, and data flows. She is right to be suspicious of any solution that promises to “scale” care.

From inside schools and systems, though, the picture looks messier.

For teachers, the upsides of personalisation are not theoretical. It is the student who finally gets work pitched just above their current level instead of drowning or coasting. It is the class where you don’t have to choose between boring half the group and losing the other half. For leaders, the promise is equally concrete: a lighter differentiation and marking load for teachers, more precise data on who needs what, and (in theory) more time and energy for the social and emotional work no platform can do.

So I don’t find myself against personalisation. I find myself against pretending it comes without trade‑offs.

From a leadership perspective, AI‑driven personalisation sharpens at least three tensions in our systems:

• personalisation vs belonging
• early identification vs response capacity
• optimisation vs equity.​

This first piece focuses on the one we are least honest about: personalisation vs belonging.

The classroom win is real

For most teachers, the appeal of AI‑enabled personalisation is obvious.

In a typical mixed‑attainment class, you are always trading off. Push the pace and you lose the quieter or less‑prepared students; slow down and you lose the ones who were ready two lessons ago. Everyone knows that “teaching to the middle” is a compromise. We do it because we are finite.​​

Adaptive tools promise a different pattern. A student who has silently struggled with fractions can get targeted practice until something clicks. Another student can move on to algebra rather than sitting through yet another whole‑class recap. In the best cases, teachers report that this feels more humane than the batch‑processing we grew up with.

Personalisation can also free up some cognitive and emotional bandwidth. If a platform can handle some of the levelling, spacing, and feedback work with reasonable fidelity, a good teacher can use the saved energy on what no algorithm can do: noticing the child who came in flat, managing the dynamics of a tense class, or having the conversation that actually keeps a borderline student in school.

Those are not imaginary gains. It would be dishonest to pretend otherwise.

Pelletier is right to ask who else benefits when “personalised learning” becomes a policy mantra. And its not hard to see that platform providers, data brokers, accountability regimes all benefit. But if we stop the analysis there, we miss what teachers see: the very real relief of not having to carry all of this work alone.

Schools run on social architecture, not just content

The problem is that schools don’t just deliver content; they build and maintain social worlds.

When we wrote our Ecosystem of Care, we were trying to make that invisible work visible: teacher–student relationships, peer support and friendship networks, staff collegiality, and the sense of belonging between people and the institution. Around those, we organised enablers 9(prioritisation, customisation, affirmation) and social outcomes: “a second home” for students, “a home away from home” for staff.​

Academic learning sits inside that architecture; it doesn’t replace it.

AI personalisation starts from a different unit of analysis: an individual learner, a model of the curriculum, and a stream of tasks and feedback. When we drop that model into a real timetable without adjusting anything else, some predictable shifts follow:

• more time spent on individualised tasks, often on devices;
• more divergent pacing within the same class or cohort;
• fewer moments when everyone is working on the same thing, in the same way, at the same time.

For some young people, that is a relief. For others, especially those already on the edge of the sociogram, the visual maps of who is connected to whom in a class, it quietly narrows their contact with peers. The student who at least sat in the same lesson, at the same point, now spends longer on a personalised track with fewer natural excuses for classmates to lean over, ask a question, or pull them into a group.​

Teachers’ relational maps also take a hit. When groupings and tasks are constantly fragmenting around adaptive recommendations, it becomes harder to see who sits where, who talks to whom, and which fragile ties need reinforcing. None of this appears in platform analytics. Individual progress curves may look healthier; the binding energy of the community may be lower.

This is the first tension: the very logic that optimises individual learning can, if left unattended, erode the conditions that make learning sustainable.

That is not an argument against personalisation. It is an argument against handing the shape of our social architecture over to whatever is easiest to measure.

What must stay non‑personalised?

If we accept that both sides of this tension are real — the classroom win and the ecosystem risk — then the leadership question changes.

It is no longer “Should we personalise?” The more honest question is: What must remain deliberately non‑personalised in this school so that belonging survives?

In practice, that means making some conscious design choices before you scale AI.

Protect common experiences.

Camps, whole‑cohort projects, performances, assemblies: these are expensive in time and logistics, but they do social work that no adaptive pathway can do. They are where new friendships form, where identity and pride in the institution are built, and where students see themselves as part of something larger. Treat them as core infrastructure, not nice‑to‑haves to be squeezed out by more “efficient” learning time.​

Ring‑fence relational time.

Advisory or form time easily gets colonised by “useful” content. Decide explicitly that these slots are for relationships, reflection, and collective sense‑making, not for squeezing in more personalised tasks. That might mean holding the line when a new platform promises “high‑impact micro‑lessons” to drop into tutor time. The impact you care about there is not item‑level mastery; it is connection.​

Keep some lessons shared on purpose.

Design sequences where the point is that everyone wrestles with the same problem together, even if it is not optimally pitched for each individual. Those lessons create shared reference points, classroom stories, and moments of peer support. They are often where you see students notice each other as thinkers and contributors, not just as data points.​​

Watch your timetabling logic.

It is easy, in the name of flexibility, to keep regrouping students into ever finer sets based on current performance. Some flex is useful. Too much, and you lose the stability of classes as social units and the slow work of teachers building a community that can hold diversity. Before you make grouping more fluid, be clear what you are giving up.

In other words: if AI is going to help with personalisation, leaders need to help with de‑personalisation — the parts of school life we hold in common because they bind us together.

From tool adoption to ecosystem design

Framed this way, the question Pelletier poses — “why be careful about personalised learning?” — lands differently. The answer is not “because personalisation is bad”. In practice, personalisation can let teachers do a more humane job with limited time.

The answer is that every gain on the personalisation side sits inside an ecosystem with a finite budget of attention, time, and emotional energy. If we do not design for belonging at the same time, the ecosystem will pay for those gains in ways that do not show up on vendor dashboards.

And this is only one of the tensions AI will sharpen.

The second is early identification vs response capacity: AI can help us see risk earlier and more precisely, but unless we redesign casework pathways and staff time, we simply move from being late and surprised to early and overwhelmed.

The third is optimisation vs equity: adaptive systems can feel compassionate in the moment, but over time they can harden into soft tracking, offering very different futures to different students unless we put serious guardrails around how recommendations translate into opportunities.

Pelletier is right to warn that datafication and optimisation can hollow out education if we let their logic run unattended. From where I sit, the more practical question is this: will we do the institutional work needed to keep AI personalisation in service of human aims, or will we quietly re‑engineer our schools around whatever the dashboards make easiest to see?

In the next parts of this series, I’ll turn to those other two tensions — how we handle early warning without burning people out, and how we stop “support” from shrinking our most vulnerable students’ horizons.

For now, the first move for leaders is deceptively simple: before you personalise anything else, decide what you refuse to personalise.

Further reading

If you want to go deeper into the ideas behind this piece:

Pelletier, C. — “Against Personalised Learning” International Journal of Artificial Intelligence in Education (2023). A critical analysis of personalised learning and learning analytics, arguing that they often serve institutional and commercial logics more than educational ones. Link: https://discovery.ucl.ac.uk/id/eprint/10173863/

On critiques of datafication and learning analytics

• “What’s the Problem with Learning Analytics?” — critical review of assumptions and risks in learning analytics (ERIC). Link: https://files.eric.ed.gov/fulltext/EJ1237568.pdf
• Ifenthaler & colleagues, “Learning Analytics Considered Harmful” — on how dashboards can overwhelm rather than support action.​ Link: https://files.eric.ed.gov/fulltext/EJ982677.pdf

On AI, personalisation, and equity

• “Personalized Learning with AI: Adapting Education to Learners’ Needs” (IntechOpen).​ Link: https://www.intechopen.com/online-first/1230893
• Stanford Center for Racial Justice, “How will AI impact racial disparities in education?” Link (overview): https://law.stanford.edu/2024/06/29/how-will-ai-impact-racial-disparities-in-education/
• “Empowering learners with personalised learning approaches? Agency, equity and transparency in the context of learning analytics.”​ Link: https://research.monash.edu/en/publications/empowering-learners-with-personalised-learning-approaches-agency-