Personalized Learning

Personalized learning — tailoring educational experiences to individual learner profiles, including prior knowledge, learning pace, preferences, and affective states. AI enables personalization at scale, though the gap between system personalization and learner-perceived personalization remains an open measurement challenge. Alongside adaptive learning and intelligent tutoring, it is one of the application-side members of the learner modeling and adaptive instruction family — consuming learner models to adapt instruction.

Questions to Consider

• When you think of 'personalized learning,' do you imagine content tailored to a learner's pace, or to their chosen goals? The page says these are deeply different (uniform outcomes via varied paths vs. diverse outcomes). Which do you value more, and why?
• The page distinguishes personalized learning (the goal) from adaptive learning (one mechanism). Can you think of personalization that doesn't involve real-time adaptation—and does it still count?
• A system can adapt without the learner ever feeling recognized. When have you experienced being 'personalized to' without feeling genuinely known? What's the difference?
• The page flags that over-personalization can strand learners in low-expectation tracks. How might well-intentioned AI tailoring accidentally lower the ceiling for a learner?
• Personalization needs detailed learner data; privacy needs data minimization. Where do you draw the line between 'enough data to adapt' and 'so much that the learner is exposed'?
• What would an AI need to remember about you across sessions to genuinely personalize your learning—and what are the risks of it remembering those things?

Introduction

Tailoring educational experiences to individual learner profiles, including prior knowledge, learning pace, preferences, and affective states. AI enables personalization at scale, though the gap between system personalization and learner-perceived personalization remains an open measurement challenge.

• Mishra et al. distinguish two forms of personalization with deep historical roots — uniform outcomes reached via varied paths (Skinner's teaching machines to Khan Academy-style mastery tutoring) vs. diverse, learner-chosen outcomes — mapping onto the field's control-vs-agency tension.
• Personalization can be met by generating the material rather than finding it. DagFlow-AI generates synthetic statistics datasets matched to a student's stated interest, the course learning objectives and the intended skill level, a three-way match the authors argue real datasets rarely satisfy (Arthur et al., 2026). The limit is built in: synthetic data cannot teach data collection, measurement or study design, so the tool complements real data rather than replacing it.

Architectures for AI-Driven Personalization

Longitudinal Memory (PersonaVLM → Education)

Nie et al. (2026) developed a Multimodal AI long-term memory architecture (PersonaVLM) that maintains persona consistency across interactions. Mapped to education, this enables tutoring systems that remember a learner's Misconceptions about AI, preferred explanations, and progress history across sessions—addressing a critical deficit in stateless chatbot tutors.

Agent-Native Personalization Substrate (DeepTutor)

Zhao et al. (2026) design every feature of DeepTutor to share a common personalization substrate, rather than bolting personalization onto reactive tools. This architecture ensures cross-modality coherence: the same learner profile drives problem solving, question generation, and collaborative writing.

Multi-Agent Social Personalization (MAIC)

Yu et al. (2024) personalize not only content but social context. Classmate archetypes (Class Clown, Deep Thinker, Note Taker, Inquisitive Mind) create varied peer-learning dynamics matched to individual learner needs.

AutoML for Learner Portraits

Personalization is a central objective for improving educational quality, yet processing multi-source heterogeneous learning-behavior data remains a challenge. A personalized neural cognitive architecture search framework, driven by automated machine learning, builds learner portraits and generates diagnostic models for heterogeneous learner profiles, integrating multi-modal data to move beyond static examination outcomes.

A static knowledge base cannot personalize: ontologies evolve slowly and handle uncertainty poorly, so the architecture matches representation to knowledge type — declarative to ontologies, procedural to rules, uncertain to fuzzy or probabilistic ontologies, implicit to analytics and machine learning — and prefers a system of small mapped ontologies to one monolithic model (Ivanova, 2026).

Relationship to adaptive learning and intelligent tutoring

Personalized learning is often conflated with adaptive learning, but they are not the same. Adaptive learning refers to the mechanism — a system adjusting content, pacing, and difficulty in real time based on a learner model. Personalized learning is the broader goal — tailoring the full learning experience (content, pathways, pacing, preferences, goals) to an individual, of which real-time adaptation is one implementation. Adaptive systems are a means toward personalization, but personalization can also be achieved through static learner profiles, choice-based pathways, or human-tutor tailoring that does not adapt in real time.

A PRISMA 2020 review of 22 higher-education interventions places personalized-learning support and adaptive pathways among the dominant AI use cases, yet most implementations improved existing practice rather than transforming it (AlSheikh et al. (2026)).

Intelligent tutoring sits in between: ITS are the canonical adaptive platforms that deliver personalized instruction through structured student modeling, while Large Language Models (LLMs)-based tutors personalize conversationally. All three are the application-side members of the learner modeling and adaptive instruction family — they consume the learner representations produced by student modeling, Knowledge Tracing, and Cognitive Diagnosis to decide what to teach next. The distinction matters for evaluation: studies that label a system "adaptive," "personalized," or "individualized" interchangeably (see below) can obscure whether the claimed benefit comes from real-time adaptation, learner choice, or content tailoring.

A 2026 systematic review of 61 higher-education studies gives that distinction an empirical edge. Learner-directed control, meaning students decided when, what and how to ask, was coded in 46 studies (75.4%). Purpose-built or course-customized systems appeared in 10 (16.4%) and documented system-driven adaptation in five (8.2%); the review's sensitivity analysis left one (Aybar-Bellido et al., 2026). Most of what the literature calls personalized learning is flexible access to a general assistant, not a model of the learner.

Measurement Challenges

• System vs. perceived personalization — A system can adapt without the learner feeling recognized

• Longitudinal validity — Personalization benefits may decay if profiles become stale or overfit

• Equity risks — Over-personalization can strand learners in low-expectation tracks

• Bias risks — Conditioning on student attributes can encode stereotype: with essays held constant, feedback for students marked by race, language, or disability became more praising and less critical (Tan, Phalen & Demszky (2026)).

Personalization and assessment

Personalization and Assessment are tightly coupled in AI-driven learning. Adaptive personalization depends on ongoing formative measurement of what a learner knows (via Knowledge Tracing, Learner Modeling and Adaptive Instruction, and Cognitive Diagnosis) to decide what to adapt next — so the reliability of the Assessment signal directly constrains the quality of personalization. Conversely, when summative assessment is personalized per-learner, fairness and comparability become harder to establish. The knowledge base's research warns against over-adapting to shallow or noisy signals: adaptive systems that mis-measure a learner can personalize in ways that reduce learning rather than support it, and AI-native students whose Self-Assessment is unreliable (an "absent cognitive baseline") are harder to model accurately.

Personalization in the AI era

The strongest evidence that this concern is not hypothetical comes from a three-wave longitudinal study of 486 Chinese undergraduates (Li, Lin & Qiu, 2026). It found that the more personalized students perceived their AI-adaptive environment to be, the lower their self-regulated learning. The study names this the "personalization paradox." Shifts in academic emotions carried most of the effect: encountering the adaptive environment predicted less enjoyment and more anxiety and boredom, and those emotional changes together accounted for roughly half of the association between personalization and reduced self-regulation. AI literacy buffered the damage, weakening the negative emotional association to non-significance at high literacy. Personalization therefore appears to buy adaptive fit at a cost to the learner's own regulatory activity, and the study points to emotional experience — not only cognitive load — as the channel through which that cost is paid.

Where the diagnosis behind a path is validated, personalization pays off by lowering load rather than by covering more: shortest-path remediation averaged 3.82 steps and cut study time 22.0% (57.6 versus 73.8 minutes), with cognitive load carrying 53.7% of the post-test effect (Feng & Huang, 2026).

Evidence for the tool can itself be negative at small scale: in a five-day elementary fraction trial (final n = 22), the business-as-usual group showed significantly greater comprehension gains than the AI-adaptive Mathbot group, and the authors flag grade-level confounds, guessing and licensing cost as limits — an "adaptive" label carries no effect (Holman (2024)).

Reinforcement learning is a distinct mechanism for personalization, and Riedmann, Schaper & Lugrin (2025) map its empirical track record: their PRISMA review of 89 RL-in-education studies finds RL personalization concentrated in Higher Education and Math Education, with adaptation implemented mainly as content scheduling (n = 53) or guidance-related personalization such as hints and feedback (n = 36). They report that RL policies beat non-adaptive baselines most often on guidance-related adaptation and on affective variables (63% of tested studies), and that learning gain — especially normalized learning gain — was the most effective reward source — practical guidance for designing reward signals that personalize toward genuine learning rather than engagement.

Wang et al. (2026) show the reward objective is itself a personalization choice: an RL coach trained on the learner's independent competence cut lap time 27.9% (p = 0.005) where rule-based fading produced no reliable change, and the authors argue coding agents optimized for task performance carry no incentive for what the human retains.

Bernstein and Sibia (2026) sharpen a distinction between interest personalization and expertise personalization: interest-matched GenAI analogies were reported as more engaging and memorable but not uniformly more trusted, and some students preferred the generic technical explanation even when the analogy matched their stated interest, for self-sufficiency and completeness (Flawed but Memorable: Student Critical Reception of Interest-Personalized GenAI Analogies in Computing Education). Their design recommendation is to personalize through source-domain structure and to ask students what they already know, not only what interests them, since familiarity with a source domain is what lets a learner inspect the analogy. They also recommend giving learners control over personalization through a menu of analogies, opt-in, or offering generic and personalized versions together. Sidorkin (2026) documents a further pairing at the level of course materials rather than individual explanations: weekly readings generated on demand for a graduate educational leadership course were tailored at once along interest (sector, professional role, local examples) and comprehension level (pacing, definitions, depth). The resulting logs shared a common backbone (TF-IDF cosine similarity of 0.50 to 0.61), which he reads as a template with adjustable dials rather than a wholesale rewrite per learner. The same corpus shows that tailoring was structural but uneven in intensity: artifact-level tailoring markers averaged 52.24 per 10,000 words and ranged from 38.74 to 74.29 across logs, while comprehension-oriented prompts produced 3.4x to 8.7x more definitional Scaffolding than baseline explanatory text.

A third axis of personalization is the goal, and it is the input AI planners handle worst. Liu et al. (2026) paired 2,000 synthetic learner personas with a 347-textbook, 4,092-concept prerequisite graph and asked ten LLMs to plan, step by step, which knowledge a learner should study to reach a stated target unit. The models produced structurally sound curricula — DeepSeek-V3.1 reached 90.9% on prerequisite-and-hallucination validity — while failing to adapt them to the learner. Adaptivity topped out at 44.7%, DeepSeek-V3.1's final pass rate was 29.5% in Basic Education and 14.6% in Higher Education, and removing the mastery field from the persona cost up to 26.1 percentage points of adaptivity while leaving validity almost unchanged. Generating the whole path in one pass instead of interactively raised validity by as much as 30.8 points while cutting adaptivity by 28.8. The claim "personalized" is a claim about responding to a learner's state, and the state variable is the part these planners can most easily do without — a computational counterpart to the measurement concern above.

A fourth axis is the audience rather than the individual learner: Bespoke regenerates an existing lecture for a named professional group (healthcare, finance, or energy), and its expert raters scored industry-framed versions 0.32 points higher on personalization depth (3.97 vs. 3.65) while audience calibration lagged (3.52). Tailoring to a cohort rather than to a learner is a cheaper and more tractable form of personalization, but the rubric that measured it assessed judged fit, not learner outcomes.

Personalization can outweigh a human presenter: in a large online course (493 respondents), students ranked AI-generated personalized videos above non-personalized human-recorded ones (mean rank 2.26 versus 2.69) and 88.4% ranked some personalized video first, against 73.8% for human-recorded (Tomlinson et al. (2026)).

Personalization can also assemble all three learner variables into one pipeline on a real second-language platform: Gharahighehi et al. (2026) combine preference, proficiency, and engagement models over NedBox logs from 7,346 learners. They rank materials with the lightweight EASE recommender (AUROC 0.955), re-rank the top-10 by DeepIRT proficiency estimates (best MCC 0.34), and filter items a survival forest flags as high dropout risk (C-index 0.662). Each learner's entry difficulty is set to the level whose predicted probability of a correct answer sits closest to 0.75.

How strong is the evidence

The only pooled estimate available here comes from a meta-analysis of four controlled studies covering 420 students, all in computing and engineering higher education (Govea Sotomayor & García Ortiz, 2026). The random-effects pooled effect is large, g = 1.01 (95% CI 0.33–1.68), and that number should be read as a caution rather than a headline. With four studies, the Hartung-Knapp adjustment widens the interval to −0.10 to 2.11 (p = .063), which includes no effect.

Heterogeneity is severe: Q(3) = 25.20, I² = 88.8%, τ² = 0.41. The 95% prediction interval, −0.42 to 2.43, means a new implementation could plausibly yield a null or negative result. Two of the four studies were at serious risk of bias and carried 53% of the weight. Publication bias was not assessable, because the usual tests need at least ten studies.

GRADE certainty was rated very low. The honest reading is that the point estimate is large and the interval is too uncertain to act on. The pooled outcome was academic performance in computing and engineering courses, not learning broadly. Claims that AI personalization works should carry that qualifier, since this is the thinnest base on which the page's evidence rests.

Prompt-conditioned micro-personalization

Basu, Kakar & Goel (2026) show that the gap between system and perceived personalization can be addressed at the response level. Their framework for the Jill Watson Large Language Models (LLMs)/RAG (Retrieval-Augmented Generation) tutor combines learner-selected preferences (abstraction, verbosity, perception, processing, understanding) with system-inferred cognitive demand (Bloom's Taxonomy) to produce 96 micro-profiles adapted at each interaction via structured prompt conditioning — no retraining, no domain-specific authoring. This is a hybrid of adaptability (learner-driven preference selection) and adaptivity (system-driven cognitive assessment), showing that personalization of how content is presented can be both scalable and perceptible to learners.

Terminological ambiguity

A recurring problem is that "personalized learning" is a broad, loosely defined umbrella term. Systematic reviews (Khalifeh et al., 2026) find that adaptive learning, individualized instruction, customized learning, and personalized learning are used interchangeably, with no universally accepted definition — a source of conceptual ambiguity that complicates research synthesis and evidence-based practice. The field increasingly calls for a unified framework and definition so that "personalized" denotes a precise, evidence-backed claim rather than a vague label (a point reinforced by the knowledge base's critique of weak construct use).

Common Misconceptions

"Personalized learning and adaptive learning are the same thing." Adaptive systems are one means toward personalization, not its definition. Personalization can equally be achieved through static learner profiles, choice-based pathways, or human-tutor tailoring that never adapts in real time. Treating the two as synonyms makes a school that offers rich learner choice but no adaptive engine look unpersonalized, and makes a tightly adaptive drill look more learner-centered than it is.

"Personalization pays off by covering more ground." Where the diagnosis behind a path is actually validated, it pays off by lowering load rather than by coverage: shortest-path remediation averaged 3.82 steps and cut study time 22.0% (57.6 versus 73.8 minutes), with cognitive load carrying the effect. The benefit is fewer unnecessary steps, not more content per hour.

"A model that can generate a coherent curriculum can personalize it." Generation and adaptation are different capabilities, and the gap between them is large. Curriculum generators were strong on prerequisite-and-hallucination validity — DeepSeek-V3.1 scored 90.9% — but weak on adapting to the learner: adaptivity peaked at 44.7%, and DeepSeek-V3.1's final pass rate fell to 29.5% in the Basic condition. Prerequisite validity is not personalization.

"Generating the material removes the constraint real content imposes." Generation relaxes one constraint and cannot relax others. DagFlow-AI generates synthetic statistics datasets matched to a student's stated interest, the course objectives and the intended skill level — a three-way match real datasets rarely satisfy — but its authors state the limit plainly: synthetic data cannot teach data collection, measurement or study design, so the tool complements real data rather than replacing it.

"Reported learning gains are what validate these systems." In the RL and personalization studies surveyed here, policies beat non-adaptive baselines most often on guidance-related adaptation and on affective variables (63% of tested studies), while learning gain — especially normalized learning gain — was the least commonly reported outcome. A literature that reports engagement and affect far more than learning is not evidence that learning dominates.

Implications for Practice

For instructors

• Ask what is being adapted before believing the word "personalized": the content, the sequence, the pace, the affect, or the presentation. Different adaptations carry different evidence, and only some have learning outcomes attached.
• Prefer systems that shorten the path over systems that add content. The measured benefit in the remediation study was 22% less study time, not more coverage.
• Keep the parts of the discipline a generated dataset cannot supply — data collection, measurement, study design — as real work, and let personalization operate on the practice around them.

For developers

• Report learning outcomes, not only engagement and affect. The survey finding that learning gain is the least-reported measure is a description of what the field is currently shipping, not a template to copy.
• Distinguish generation quality from adaptation quality in your own evaluation, ideally reporting both separately, because strong prerequisite validity can coexist with adaptivity under 50%.
• Treat the reward objective as a personalization choice in its own right: an RL coach trained on the learner's independent competence cut lap time 27.9% (p = 0.005) where rule-based fading produced no reliable change.

For researchers

• Stop conflating personalization with adaptation in the constructs you measure, and say explicitly which mechanism produced the effect.
• Report normalized learning gain as a primary outcome; its scarcity in the literature is a gap to close rather than a convention to inherit.

Connected Concepts

• Adaptive Learning — Adaptive systems that tailor content, pacing, and difficulty to the learner in real time
• Intelligent Tutoring — Tutoring systems that model the learner and deliver individualized instruction
• Learner Modeling and Adaptive Instruction — Representing learner knowledge, skills, and states that drive adaptation
• Knowledge Tracing — Inferring mastery of knowledge components from performance over time
• Cognitive Diagnosis — Diagnosing latent learner knowledge and attributes from responses
• Scaffolding — Support and fading calibrated to individual learner needs
• Student Experience — The learner's lived experience of personalization
• Learning Analytics — Data-driven measurement of learning that informs adaptation
• Formative Assessment — Ongoing assessment that signals what to adapt next
• Summative Assessment — Endpoint assessment whose comparability personalization complicates
• Generative AI — LLM-based conversational personalization
• Edtech Platform — Platforms that deliver personalized learning at scale
• Higher Education — Higher-education context for personalization
• Online Teaching and Learning — Online Teaching and Learning
• Recommender Systems and Learning Paths

Connected Articles

• Generative AI for Learning Individualization and Student Support in Sustainable Higher Education: A Systematic Review of Creative Agency, Equitable Learning, and Implications for SDG 4 and SDG 10 — 61 studies: 75.4% coded learner-directed control, 8.2% system-driven adaptation (Aybar-Bellido et al., 2026)
• Bespoke: Generating MOOC-Quality Industry-Personalized Lecture Videos at Scale — Industry-personalized lecture video regeneration from a seed transcript: audience-level tailoring, rated by domain experts (Puech et al. 2026)
• A Prompt-Engineering Approach to Develop Scalable, Flexible, and Real-Time Hybrid Micro-Level Personalization in a General Purpose AI Teaching Assistant — Prompt-engineering micro-personalization of an AI teaching assistant (Basu, Kakar & Goel 2026)
• AI Tutoring is Not a Monolith: What We Actually Know — AI Tutoring is Not a Monolith: What We Actually Know (Stanford SCALE/NSSA brief)
• Control vs. Agency: Exploring the History of AI in Education — Distinguishes two forms of personalization (uniform vs diverse outcomes)
• Redefining personalized learning in the artificial intelligence era: an updated systematic review from 2019 to 2025 — Redefining personalized learning: systematic review
• DeepTutor: Towards Agentic Personalized Tutoring — Agent-native personalization substrate for tutoring
• Ontology-based layered hybrid AI-driven knowledge model for personalized e-learning — Ontology-based layered hybrid knowledge model for personalized e-learning
• Exploring Fraction Comprehension and Interest in Elementary Education Through AI-Powered Personalized Learning — Personalized adaptive learning for elementary fractions
• AI Coaching for Accelerating Human Skill Development with Reinforcement Learning — Reinforcement-learning coaching for skill development
• Students prefer personalized, AI-generated educational videos over non-personalized, human-recorded videos — Students prefer personalized AI-generated videos over non-personalized human-recorded ones (Tomlinson et al. 2026)
• Bayesian cognitive diagnosis optimizes personalized learning paths via mediation of cognitive load and Hidden Markov Model state transitions — Bayesian cognitive diagnosis for personalized learning paths
• Intelligent tutoring in dynamic domains: a graph-based system for comparative analysis of adaptive algorithms with intuitionistic fuzzy logic and forgetting — Graph-Based Intelligent Tutoring for Dynamic Domains (2026)
• Instructor and AI roles in the chemistry classroom: Future science teachers' perceptions in a ChatGPT-enhanced formative assessment — Instructor and AI roles in ChatGPT-enhanced formative assessment
• Marked pedagogies: Examining linguistic biases in personalized automated writing feedback — Marked Pedagogies: bias in personalized automated feedback
• Mapping artificial intelligence integration in higher education: A systematic review using the FACETS and SAMR frameworks — Systematic review: adaptive pathways & recommenders are a top AI integration use case in higher ed
• Reinforcement Learning in Education: A Systematic Literature Review
• Flawed but Memorable: Student Critical Reception of Interest-Personalized GenAI Analogies in Computing Education — Flawed but Memorable: Student Critical Reception of Interest-Personalized GenAI Analogies in Computing Education
• The personalization paradox: how AI-driven adaptive learning environments are associated with college students' academic emotions and self-regulation learning—a moderated mediation model — Personalization paradox: perceived adaptive personalization linked to lower self-regulated learning via academic emotions, buffered by AI literacy (Li, Lin & Qiu 2026)
• PersonaPath: Towards Knowledge-Centric Personalized Learning Path Planning — PersonaPath: LLM planners reach 90.9% validity but no model exceeds 44.7% adaptivity when personalizing paths to a stated learner goal (Liu et al. 2026)
• AI-based personalized learning and academic performance in computing and engineering higher education: a systematic review and meta-analysis of controlled studies — Four studies, a large pooled effect, and very low certainty