Personalization without more isolated screen time

Personalization is a relationship, not a sequence

Personalized learning is often pictured as a student alone with software that continuously selects the next item. That can change pace or difficulty, but it is only one narrow form of adaptation. It may know whether an answer was accepted while knowing little about the learner's purpose, relationships, language, confidence, or contribution to a group.

Schools that have tried to personalize at scale show a more ordinary and more human picture. When RAND studied US schools funded to personalize learning, teachers and students reported more time for one-on-one support than a national comparison group, while flexible paths and extensive student choice were not common in either. Estimated achievement effects were small. Comparing about 5,500 students in 32 of these schools with matched peers in other schools, RAND found a statistically significant gain of about 0.09 standard deviations in maths and no significant effect in reading. The design matched students rather than assigning them at random, and RAND describes the evidence as early, recommending research with stronger experimental designs.¹

For project-based learning, a more useful definition begins with context. Personalization means shaping goals, choices, scaffolds, feedback, roles, and routes through meaningful work so that students can participate and grow without being separated from the people and situations that make the work matter.

The OECD's 2026 outlook recommends using generative AI selectively and purposefully for pedagogical reasons, without replacing cognitive effort or weakening the human relationships at the heart of education.² That boundary is more useful than asking whether a lesson uses more or less technology in the abstract.

Four levers educators can shape

Context

Students bring different prior knowledge, interests, cultural references, responsibilities, and ways of participating. Contextual personalization does not require a unique curriculum for every learner. It can begin by connecting a shared disciplinary goal to different local questions, examples, audiences, or entry points.

An educator might offer several community problems that call for the same scientific reasoning, or help students locate a role that lets them contribute while stretching a less-developed capability. The standard stays visible; the route into meaningful practice becomes more responsive.

Choice

Choice is meaningful when it changes a decision the learner can own. Choosing a font or clicking among three preselected topics is not the same as selecting a question, audience, method, role, or form of evidence and being accountable for the consequences.

The research on choice is less tidy than that distinction. A meta-analysis of 41 experimental studies found that offering choice raised intrinsic motivation, but its effect on subsequent learning was small and not statistically significant. Motivation effects were largest for instructionally irrelevant choices and for two to four successive choices, and the authors caution that too many or highly effortful choices may reduce the benefit.³ Most of those studies were brief, so they say little about choice across an extended project.

Useful choice also has boundaries. Too little leaves no agency; too much can make students spend their energy guessing what the assignment is. Teachers can name the non-negotiable knowledge and quality expectations, then create a small number of consequential decisions within them.

Scaffolding

A scaffold helps a learner perform thinking that is not yet independent. It might be a model, worked example, question sequence, checklist, vocabulary prompt, peer protocol, teacher conference, or temporary structure for planning. Its purpose is to fade as capability grows, not to complete the task on the student's behalf.

The Education Endowment Foundation's metacognition guidance includes explicit teaching, modelling, and scaffolding of strategies for planning, monitoring, and evaluating learning.⁴ These are teacher-supported practices. They should not be relabeled as an automated product outcome.

Group work

Personalization and collaboration are not opposites. A group can make individual strengths, needs, and perspectives visible in ways solitary practice cannot. The design challenge is to create meaningful interdependence while retaining evidence of each learner's understanding and contribution.

The challenge is real. In RAND's study, students in schools that offered some choice of path and content often reported working on different topics, and some said this made it harder to seek help from peers or collaborate with them.¹ Individual routes need deliberate points where the group's work comes back together.

Roles can rotate. Students can keep individual decision notes. A teacher can confer with one learner while preserving the shared project. Peer explanation can become both support and evidence—provided the classroom makes it safe to reveal uncertainty.

Technology should return attention to the work

Technology earns a place when it helps educators or learners act more thoughtfully in the real activity. It might make a process record easier to review, surface a question for a teacher's attention, provide a hint that protects productive struggle, or help a learner rehearse an explanation before sharing it with people.

The evidence on adaptive software is mixed, which is a reason to judge it by what it does rather than rule it out. A meta-analysis of intelligent tutoring systems in K–12 mathematics, drawing on studies published before 2011, found effects ranging from near zero to small, with smaller effects for low-achieving students.⁵ Those studies predate current AI tutors, so they do not settle how newer tools perform. In a randomized trial in Delhi, after-school sessions that combined adaptive software with small-group support from teaching assistants raised scores by about 0.37 standard deviations in maths and 0.23 in Hindi over four and a half months, among students whose levels varied widely.⁶ Software can earn its place, but even there it was not delivered alone, and the trial cannot fully separate its contribution from the group support.

It should not quietly redefine personalization as more time consuming recommendations. A hint is useful only if meaningful thinking remains. A suggested next step is useful only if the learner and educator can understand, change, or reject it. A dashboard is useful only if it returns attention to the student's work rather than presenting an unexplained proxy as the student.

Our resource on real-world learning in the AI era develops this distinction between producing an output and developing capability. Our responsible AI approach explains why support should remain hint-led and educator-controlled.

Teacher observation makes adaptation human

Some of the most responsive personalization happens in a short exchange: a teacher notices that a student is using evidence mechanically, asks one question, and changes the next challenge. That response draws on knowledge of the goal and the person. It can consider motivation, language, relationships, previous attempts, and what the student is ready to own.

This is why personalization cannot be reduced to a permanent profile. Learners change. A preference can become a constraint if it is treated as identity. A student who once needed a visual scaffold may be ready to explain without it. A learner who appears quiet in one group may lead in another context.

Design for revisable hypotheses rather than fixed labels:

• preserve recent evidence, not a permanent verdict;
• let students explain what support is and is not useful;
• make recommendations visible to educators;
• offer more than one path and allow a human override;
• revisit whether the scaffold still serves the learning purpose.

The educator pathway describes how continuity and support can reduce fragmented work while keeping educators in control. The companion article on visible growth shows why work-in-progress evidence matters more than a static learner category.

Privacy is part of the learning design

Personalization can invite excessive collection: more behavioral traces, inferred traits, or sensitive context retained because it might someday be useful. UNICEF's school data-protection guidance warns that digital education can involve personal and sensitive data processed by schools, governments, third parties, and commercial providers, and sets out practical data-protection obligations, compliance steps, and cybersecurity controls for schools.⁷

Before collecting another signal, ask:

• What learning decision will this information support?
• Could a less sensitive or shorter-lived record support the same decision?
• Who can see it, correct it, and decide when it is no longer needed?
• Does the student understand how it affects the support they receive?
• Can an educator act without turning a tentative inference into a permanent label?

Our current public commitments are described on the privacy page, while the learning philosophy states the human purpose those commitments protect. Those commitments should be reviewed as the product changes; an aspiration for future functionality is not evidence of a current practice.

A practical test

Personalization is working when a learner can enter worthwhile shared work, make meaningful choices, receive support without losing ownership, and become more capable of acting independently. The measure is not how continuously a system adapts. It is whether the learning relationship becomes more responsive while the student remains connected to thinking, people, and purpose.