1. Introduction and problem

In the 3-to-6 years band — kindergarten, preschool, CENDI, infant school — a package of three artefacts now claims to count as Universal Design for Learning. The first is a platform that “personalizes”: a system that, with clicks, response times or a prior profile, delivers to each child a different path or stimulus. The second is an accessibility dashboard: a panel that activates captions, contrast or text-to-speech once someone has marked the child as a support user. The third is a model that “adapts” content: a classifier or a generator that, post hoc, rewrites the prompt or the story according to the label the system assigned. All three are visible and cheap in coordination time. They allow a centre to exhibit that it “already does UDL with artificial intelligence.” The enunciative leap is enormous: one moves from personalizing, tabulating or adapting to affirming that there is universal design. That leap is authorized neither by CAST’s UDL Guidelines 3.0 nor by implementation evidence in early childhood education.

The thesis of this article is restrictive. A platform that personalizes, an accessibility dashboard or a model that adapts content does not constitute UDL. UDL is design of barriers in advance — multiple means of representation, of action and expression, and of engagement — with the adult as designer of the environment; not a post hoc algorithmic adjustment on a classified child. CAST formulates the framework as a set of concrete suggestions so that all learners can access and participate in meaningful, challenging learning opportunities, and locates the goal in learner agency: purposeful and reflective, resourceful and authentic, strategic and action-oriented (CAST, 2024). Guidelines 3.0, released on 30 July 2024, do not ask for a prior profile of the child in order to “unlock” options. They ask that options be designed. Meyer and Rose, in the third edition of their foundational text, reiterate that UDL responds to learning differences through the design of inclusive goals, methods, materials, assessments and environments (Meyer & Rose, 2025). An algorithm that first classifies and then delivers is not designing the environment. It is administering a difference that design did not anticipate.

The problem is sharpened by a professional reason that product sheets do not mention. UDL is not an inclusion or assistive-ICT article as its unique axis. Nor is it individual differentiation. Chen and Dote-Kwan (2021) distinguish, in the preschool classroom, two overlapping but non-substitutable frameworks: UDL plans curriculum and instruction for the diverse group; differentiation addresses individual needs once the environment is already running. Confusing the second with the first is the category error this paper names. Rao, Gravel, Rose and Tucker-Smith (2023) locate UDL, in its third decade, in equity, inclusion and design: not in matching a learning style to a channel. Boysen (2024) warns that the analogy with “learning styles” has weakened the empirical base CAST exhibits. Inference, marked as such: the market for “AI for UDL” inherits the error and automates it. It turns variability — a fact of design — into a trait of the child that must be detected.

This paper does not recycle axes already treated in this series. The question is one of pedagogical category: what counts as UDL when an early childhood centre “does AI.” The contributions are three: reconstruct the state of the art that separates design of barriers in advance from personalization and from post hoc adaptation; examine three families of cases; and offer four tests for deciding when a kindergarten may claim that it designs universally, and not only that it personalizes, tabulates or adapts.

2. State of the art: from designing barriers to personalizing the child

Four strata that the “AI for UDL” market usually mixes should be kept apart. The first is UDL as a design framework: multiple means of engagement, representation, and action and expression, with learner agency as the goal (CAST, 2024). The second is still-uneven implementation evidence: a meta-analysis of 20 studies and 50 effects reports a Hedges’ g of 0.43 from pre-kindergarten to adulthood (King-Sears, Stefanidis, Evmenova, Rao, Mergen, Owen & Strimel, 2023) and a review of 32 PreK–12 studies documents uneven checkpoint coverage (Zhang, Carter, Greene and Bernacki, 2024). The third is AI in early childhood education: reviews that map prediction, classification and personalization as distinct affordances (Chen, 2024; Su & Yang, 2022; Ljungcrantz, 2026). The fourth is the rights and high-risk framework, which treats the 3–6-year-old as a subject of human oversight and, in European territory, of high-risk systems when AI determines access, admission, assignment or assessment.

In the framework stratum, Guidelines 3.0 are the most extensive change to the instrument since its origin. Gravel, Tucker-Smith, Rose, Fornauf, Hughes, Lester, Rao, Singer and Waitoller (2026) describe a four-year, community-driven process that prioritized learning from practitioners, researchers and young people themselves. CAST documents an advisory board, more than 40 focus groups and more than 1,100 unique titles. Version 3.0 acknowledges individual, institutional and system-level barriers; emphasizes interdependence; and shifts language from educator-centred to learner-centred (CAST, 2024; Gravel et al., 2026). The “who” of learning — multiple intersecting identities — is woven across the three principles. Inference, marked as such: a model that reduces the child to a difficulty vector does not operate that “who.” It compresses it.

In the evidence stratum, King-Sears et al. (2023) offer the best combined estimator available: a moderate positive effect (g = 0.43) of UDL-based instruction versus business-as-usual, with small groups (up to six) at g = 0.86 and large groups at g = 0.30. The range includes pre-kindergarten, but the meta-analysis does not isolate ages 3–6. Zhang et al. (2024) saturate the implementation picture: 32 PreK–12 studies (1999–2023), with alignment often absent between the intervention and UDL checkpoints, uneven coverage and a lack of theoretical guidance. Boysen (2024) shows that a sample of studies cited by CAST as support for three guidelines offers little evidence of choice, of measured learning or of brain function. Edyburn (2024) replies with a historical footnote on the initial base and on the relation between research and policy. Inference: UDL evidence does not authorize treating any flexible platform as UDL. It authorizes, cautiously, treating the intentional design of options as an instructional hypothesis. It does not authorize automating a profile.

In the AI-in-ECE stratum, Su and Yang (2022) review 17 eligible studies, 1995–2021, on AI in ECE. Chen (2024), in a stricter scoping, maps 18 articles from 11 countries and 16 journals (2005–2023; 14 of them in 2020–2023) covering 15,081 children aged 2 to 8. She extracts four affordances: (1) AI as tangible and intangible tools for interactive learning and information retrieval; (2) AI as technology for predicting or classifying children’s conditions; (3) AI as the object of learning that adapts and personalizes; and (4) a fourth axis of sense and perception. Ljungcrantz (2026) updates the 2020–2024 landscape: 39 studies, 20 of them from 2024, predominantly children aged 4–6 and mixed methods; 25 classified as “traditional” AI. Status: field finding, not UDL finding. Inference: affordances (2) and (3) — classifying the child and personalizing the path — are exactly what the market sells as “UDL.” CAST’s framework does not authorize them as equivalents. UDL does not begin with a diagnosis. It begins with an environment that has options.

In the rights stratum, the Recommendation on the Ethics of AI requires human oversight and particular attention when children are involved (UNESCO, 2021). Miao and Holmes (2023) set an age threshold for independent conversations with generative platforms — 13 years — and require ethical and pedagogical validation. UNICEF (2021) prioritizes the best interests of the child, development and well-being. Regulation (EU) 2024/1689 treats as high-risk, in Annex III, AI systems intended to determine access or admission to education and training institutions, to assign persons to those institutions, to evaluate learning outcomes or to determine the appropriate educational level (European Union, 2024). The European Commission (2022) and the U.S. Department of Education (2023) agree that the teacher must not be replaced. OECD (2021) anchors ECEC quality in process quality: everyday interactions with adults, peers, materials and space. OECD (2023) documents, with data from 30 countries and jurisdictions, opportunities and risks of digitalisation in ECEC. TALIS Starting Strong 2024 locates staff digital preparedness as a system lever, not as personalization of the child (OECD, 2025). The Tashkent Declaration calls for child-centred, play-based, fully inclusive pedagogies (UNESCO, 2022). Inference: a five-year-old is not the user of a dashboard that “adapts UDL for them.” They are the subject of an environment an adult designed with options.

3. Review method

A critical narrative review was conducted, not a meta-analysis. The purpose was not to estimate a homogeneous effect size among a personalization platform, a Liepāja preschool and a summer teacher institute, but to articulate an argument of pedagogical category with verified sources. Inclusion criteria: (a) 2021–2026, with CAST texts from 2024–2026 when they update the framework; (b) UDL as design of barriers in advance, or AI in early childhood education ages 3–6; (c) relevance for kindergarten, preschool, CENDI or ages 3–6; (d) peer-reviewed journal, DOI, or CAST, UNESCO, OECD, UNICEF, European Union, European Commission or ERIC report; (e) verifiable DOI or publisher page. Axes already used in this series were excluded as central objects, though some appear as limits.

The search was executed on 26 August 2026 on DOI pages, Springer, Elsevier, Wiley, SAGE, MDPI, CAST, OECD iLibrary, UNESDOC, UNICEF, EUR-Lex, ERIC and journal publisher sites. Each source was verified against at least one of those pages. The analysis distinguished three enunciative statuses. Empirical finding: what was observed or measured in the sample. Conceptual or normative framework: what a framework or a regulation prescribes. Pedagogical inference: the translation to kindergartens and CENDI centres, marked as such.

4. Case 1. Personalizing and classifying is not designing barriers in advance

Chen (2024) publishes in the Journal of Artificial Intelligence Research the scoping that best illustrates the leap the thesis rejects. With 18 articles from 11 countries and 16 journals, it covers 15,081 children aged 2 to 8 between 2005 and 2023; 14 of the 18 texts are from 2020–2023. Thematic analysis extracts four affordances of AI for use in ECE. The second is AI as technology for predicting or classifying children’s conditions. The third is AI as the object of learning that adapts and personalizes the path. Status of the evidence. Empirical field finding: the “AI for early childhood” literature already distinguishes classifying the child and personalizing content. It is not a UDL finding. Chen does not claim that those affordances are universal design. Pedagogical inference, marked as such: the gesture a kindergarten copies when it “does UDL with a platform” is exactly this. A signal is taken — click, error, time, label — a profile is assigned and different content is delivered. What exists is a post hoc adjustment. UDL, in CAST (2024) and in Meyer and Rose (2025), requires that options for representation, expression and engagement be in the environment before the child enters. A classifier does not put them there. It rations them.

Su and Yang (2022) saturate the picture with 17 studies from 1995 to 2021 on AI tools and impacts in ECE. Ljungcrantz (2026) updates to 39 publications 2020–2024, 20 of them in 2024, focused on ages 4–6 and mixed methods; “traditional” AI rises from 4 studies in 2023 to 12 in 2024. The author notes gaps in longitudinal studies, diverse populations and ethical frameworks. Status: finding of field growth, not of equivalence with UDL. Inference: the 2024 surge does not authorize translating “more AI in kindergarten” as “more UDL.” It authorizes asking what the AI does: whether it designs options in advance or classifies the child in order to deliver a channel.

Saborío-Taylor and Rojas-Ramírez (2024) explore the convergence of UDL and AI and present applications of multiple means. Bray, Devitt, Banks, Sanchez Fuentes, Sandoval, Riviou, Byrne, Flood, Reale and Terrenzio (2024), with 15 empirical studies from 1,253 records, show that, at second level, technology-and-UDL research has concentrated on the “easy wins” of the Representation principle — offering choice about how content is accessed — and leaves a gap in Engagement and in Action and expression. Status: second-level finding, not a 3–6 finding. Inference, marked as such: if even at second level “UDL technology” reduces to an access channel, the risk in kindergarten is greater. A dashboard that changes contrast or reads a PDF aloud has not designed multiple means of engagement or expression. It has activated an adjustment. King-Sears et al. (2023) measure UDL instruction, not platforms that call themselves UDL. Zhang et al. (2024) show that, without explicit alignment to checkpoints, what is called UDL often is not. Inference: a product sheet that says “personalization + accessibility = UDL” fails that test.

5. Case 2. What kindergarten does when there is UDL: the adult designs the environment

Tīģere, Bethere, Jurs and Ļubkina (2025) publish in Education Sciences the case study that best names, in this corpus, an empirical implementation of UDL Guidelines 3.0 in preschool. In Liepāja, Latvia, with the Center for Pedagogy and Social Work of Riga Technical University Liepaja Academy, they observe five children aged 4 to 8 from September 2023 to May 2024. Observation is aligned with Guidelines 3.0 and centres on communication and social interaction. Skill development is assessed with Vygotsky’s zone of proximal development. The programme designs, in advance, multiple means of engagement (interests and identities, effort and persistence, emotional capacity), of representation and of action and expression: visual schedules and timers, home bags for collecting materials with peers, independence tasks such as cooking and setting the table, visits to parks or professionals, and a repertoire of verbal, gestural, physical, visual and positional prompts. Status of the evidence. Empirical finding of environment design in a small N: options are placed in the activity before the child “fails.” It is not an AI finding. It is not a finding that an algorithm reproduces that design. It is not, and this article does not turn it into, a disability study as unique axis: the framework the authors invoke is UDL for all in an inclusive preschool, not a dashboard of individual supports. N = 5 and the 4–8 range limit generalization; the authors themselves do not inflate the data. Pedagogical inference, marked as such: this is the object an early childhood centre may properly call UDL. It is design of barriers in advance by an adult. A classifier does not collect materials with peers. An accessibility dashboard does not set the table.

Fundelius, Wade, Robbins, Wang, McConomy and Fumero (2023) translate the same gesture to storybook reading. “Book boxes” gather, in advance, tangible or abstract objects — photos, miniatures, real items — and sensory experiences, so that all children can participate. Status: instructional-design finding, not a randomized trial. Inference: the box is assembled before circle time. The story is not “adapted” after detecting who did not understand. Gauvreau, Lohmann and Hovey (2023) take UDL to circle time, a central kindergarten routine, so that all children participate. Hovey, Gauvreau and Lohmann (2022) detail multiple means of action and expression: tools, visual representations, choice and manipulation. Lohmann, Hovey and Gauvreau (2023) apply the same framework to preschool science: proactive planning, not repair. Chen and Dote-Kwan (2021) recall that UDL plans for the group and that differentiation addresses the individual. Inference: if the centre “does UDL” only once a child has already been labelled, it is not doing UDL. It is differentiating late, and often with an algorithm.

King-Sears et al. (2023) offer the quantitative contrast: combined g = 0.43, higher in small groups. The figure does not transfer, as such, to a kindergarten or a CENDI. It is used here as a ceiling of what the UDL-instruction literature — not platforms — has measured. Zhang et al. (2024) warn that, without documenting checkpoints, what is called implementation is not evaluable. Boysen (2024) obliges us not to inflate the neuroscience of the three networks. CAST (2024) and Gravel et al. (2026) shifted the language toward agency, belonging and system-level barriers. OECD (2021) locates the engine of development in process interactions. Inference: UDL evidence in early childhood is verified in environment design — book box, circle time with options, science with multiple means, prepared prompts — not in a personalization log.

6. Case 3. AI as the adult designer’s tool, not as a post hoc adjustment on the child

Evmenova, Borup and Shin (2024) publish in TechTrends the case that best illustrates the third artefact when it is placed on the right side of design. They analyse data from a summer 2023 teacher institute. At that point, most teachers had never used generative AI and were evenly split on whether it was a “friend” or “foe.” Forty-three percent stated that ChatGPT and other generative AI “will help make instruction more accessible for ALL learners”; 32% were undecided. The article discusses strategies for using generative AI to improve UDL: not to classify the student, but so that the adult produces, in advance, diverse representations, activity variants and expression supports. Status of the evidence. Empirical finding of teacher perceptions in a summer institute, not a trial in a 3–6 kindergarten. It is not a finding that generative AI “does UDL.” It is a finding that a set of teachers imagine AI as support for their own design. Pedagogical inference, marked as such: this is the only place where AI can approach UDL in early childhood education without contradicting the framework. The adult is the designer. The machine, if it enters, enters as a materials workshop, not as the child’s tutor. Miao and Holmes (2023) set the age-13 threshold for independent conversations with generative platforms. A four-year-old is not the interlocutor. OECD (2023, 2025) locates digitalisation in staff, materials and protection, not in an adaptive profile of the infant.

CAST (2024) and Gravel et al. (2026) reinforce that distribution of roles. At ages 3–6, agency is exercised in an environment an adult assembled: belonging, joy, play, identities, interdependence. Meyer and Rose (2025) recall accessible materials and universally designed environments. Rao et al. (2023) recall that UDL is a design framework, not a style match. Bray et al. (2024) show what happens when technology stays with the “easy win” of representation: an access channel is offered and that is called UDL. Inference: a kindergarten that asks a model to “adapt this story to this child” does the easy win backwards. It does not offer multiple means in the environment. It offers a single recalibrated means. That is personalization. It is not UDL.

The rights framework closes the case. UNESCO (2021) requires human oversight. UNICEF (2021) requires development and well-being. The European Commission (2022) requires that professional judgement not be replaced. Annex III of Regulation (EU) 2024/1689 includes automated evaluation of learning outcomes and determination of educational level (European Union, 2024). Inference: a system that, in a CENDI, classifies the child and assigns a “UDL level” is not a universal-design innovation. In that territory it is a high-risk practice if it determines level or assessment. Miao and Holmes’s (2023) age threshold is not “met” by lowering the user’s age. It is respected by not turning the early childhood child into the interlocutor of a platform or a classifier.

7. Inferential framework: four tests for claiming UDL, not an adjustment

The framework that follows is pedagogical inference of this article, anchored in the cases and in the verified instruments. It is not a new international standard. It distinguishes four tests. If a kindergarten, preschool, CENDI or infant school does not pass them, it cannot declare that a platform that personalizes, an accessibility dashboard or a model that adapts content constitutes UDL.

7.1. Test of design in advance, not of post hoc adjustment. CAST (2024) and Meyer and Rose (2025) locate UDL in the design of goals, methods, materials, assessments and environments. Fundelius et al. (2023) assemble the book box before reading. Gauvreau et al. (2023) design circle time so that all children participate. Tīģere et al. (2025) place visual schedules, prompts and materials in the activity. Chen (2024) shows that the AI literature, by contrast, predicts, classifies and personalizes. Inference: UDL evidence is verified in options present in the environment before the child “fails” or is labelled. If the “evidence” that the centre does UDL is an adaptation log or a profile that unlocks a mode, the centre has made an adjustment, not a design.

7.2. Test of the three principles, not of the access channel. Bray et al. (2024) document the reduction of “UDL technology” to the Representation principle. Hovey et al. (2022) and Lohmann et al. (2023) recall action/expression and engagement in the early childhood classroom. CAST (2024) weaves identity, belonging, joy, play and interdependence across the three principles. Zhang et al. (2024) require alignment to checkpoints, not to a slogan. Inference: a dashboard that changes contrast, font size or voice has not covered engagement or expression. It has covered, at best, a perception checkpoint. Declaring UDL by a channel is the easy win Bray names. At ages 3–6, engagement is played out in circle time, materials and relationship, not in a slider.

7.3. Test of the adult designer, not of the classified child. Chen and Dote-Kwan (2021) separate UDL (group plan) and differentiation (attention to the individual). Evmenova et al. (2024) locate generative AI in teacher design. Miao and Holmes (2023) exclude those under 13 as independent interlocutors of generative platforms. OECD (2021, 2023, 2025) locates quality in interactions and in staff. Inference: the 3–6-year-old is not the user of a model that “adapts UDL for them.” They inhabit an environment an adult designed. If AI enters, it enters as the adult’s workshop — story variants, pictograms, expression options — under pedagogical validation. It does not enter as tutor, classifier or personalizer of the infant.

7.4. Test of rights and high risk, not of the product dashboard. UNESCO (2021) requires human oversight. UNICEF (2021) requires the best interests of the child. Annex III of Regulation (EU) 2024/1689 includes access, admission, assignment, evaluation of learning outcomes and educational level (European Union, 2024). The European Commission (2022) requires that the teacher not be replaced. Inference: an early childhood centre cannot treat the child as the object of an automated “UDL profile” verdict. In European territory, a system that assigns level or assesses with AI is, unless the regulation’s safeguards apply, a high-risk practice. Outside that territory it remains a category error: UDL is not fulfilled by classifying better. It is fulfilled by designing options for all children before anyone asks for them.

The framework admits generative AI as the adult’s workshop (Evmenova et al., 2024); Guidelines 3.0 as a design tool, not a product checklist (CAST, 2024; Gravel et al., 2026); book boxes, circle times and science with multiple means (Fundelius et al., 2023; Gauvreau et al., 2023; Lohmann et al., 2023; Tīģere et al., 2025); and King-Sears et al.’s (2023) g = 0.43 as a cautious ceiling of UDL instruction, not of platforms. It rejects declaring UDL by a platform that personalizes, by a dashboard that unlocks accessibility after a label, or by a model that adapts content to the classified child (Chen, 2024; Bray et al., 2024; Miao & Holmes, 2023).

8. Discussion

Three tensions organize the discussion. The first is between personalizing and designing. It is a finding that the AI-in-ECE literature includes, as distinct affordances, predicting or classifying the child and personalizing learning (Chen, 2024); that the field grew to 39 studies in 2020–2024, with a 2024 peak and a focus on ages 4–6 (Ljungcrantz, 2026); and that 17 earlier studies already reported tools and impacts (Su & Yang, 2022). It is a framework that UDL designs options in advance (CAST, 2024; Meyer & Rose, 2025; Rao et al., 2023). It is not a finding that personalizing produces the environment Tīģere et al. (2025), Fundelius et al. (2023) and Gauvreau et al. (2023) observe. The policy of the three artefacts — platform, dashboard, model — measures what engineering knows how to measure and declares what only barrier design would authorize.

The second is between the child as profile and the environment as design. Chen and Dote-Kwan (2021) show that UDL plans for the group and that differentiation addresses the individual. King-Sears et al. (2023) measure instruction, not matching. Zhang et al. (2024) show that, without documented checkpoints, the “UDL” label is not evaluable. Boysen (2024) shows the risk of treating UDL as learning styles. Inference: insisting that the child “have their path” while the environment remains single-means is inverted pedagogy. The profile blames the child for what design did not anticipate. CAST (2024) had shifted the language toward system-level barriers. The personalization model does the inverse operation: it extracts the child from the environment and returns them as an itinerary.

The third is between the adult’s workshop and the child interlocutor. Evmenova et al. (2024) show teachers who imagine generative AI as support for their UDL. Miao and Holmes (2023) exclude those under 13 as independent users. OECD (2023, 2025) locates digitalisation in staff and in protection. Inference: the only AI use that does not contradict UDL at ages 3–6 is the one that remains on the adult designer’s side and undergoes pedagogical validation. A story with pictograms and objects prepared by the educator can be UDL. A model that rewrites the story because the system classified the child is not.

9. Limits

This review is narrative. It does not apply PRISMA or estimate combined effects. Chen (2024) maps AI affordances in ECE ages 2–8, not UDL implementations; transfer to the restrictive thesis is inference. Su and Yang (2022) cover 1995–2021 and do not isolate UDL. Ljungcrantz (2026) covers 2020–2024. Tīģere et al. (2025) is an N = 5 in Liepāja, range 4–8; it is not used as a disability article and does not authorize generalization. Fundelius et al. (2023), Gauvreau et al. (2023), Hovey et al. (2022) and Lohmann et al. (2023) are teaching-technique articles, not trials. King-Sears et al. (2023) include pre-kindergarten but do not isolate it; g = 0.43 does not translate into a kindergarten effect. Zhang et al. (2024) cover PreK–12. Boysen (2024) and Edyburn (2024) debate CAST’s base. Bray et al. (2024) are second-level. Evmenova et al. (2024) is a teacher institute, not a 3–6 classroom. Saborío-Taylor and Rojas-Ramírez (2024) are conceptual. CAST, Meyer and Rose, Gravel et al., Rao et al., UNESCO, UNICEF, the OECD, the European Union and the 2021–2023 guidances are framework sources. Chen and Dote-Kwan (2021) are used for the UDL/differentiation distinction, not as an inclusion axis. No Latin American trials of UDL and AI in kindergarten measuring barrier design versus personalization were located with the same degree of DOI. Section 7 inferences are hypotheses of pedagogical category, not implementation evidence.

10. Conclusions

A platform that personalizes, an accessibility dashboard or a model that adapts content does not constitute Universal Design for Learning in an early childhood centre. Verified evidence does not authorize that declaration. Eighteen studies covering 15,081 children aged 2 to 8 distinguish, as AI affordances, predicting or classifying the child and personalizing learning; that is the AI-in-ECE field, not UDL (Chen, 2024). Seventeen and thirty-nine reviews saturate field growth without equating it to barrier design (Su & Yang, 2022; Ljungcrantz, 2026). At second level, “UDL technology” concentrates on the easy win of representation (Bray et al., 2024). The meta-analysis of UDL instruction reports g = 0.43 and does not measure platforms (King-Sears et al., 2023). PreK–12 implementation stumbles on undocumented checkpoints (Zhang et al., 2024). By contrast, when there is UDL in early childhood, there is an adult who designs the environment in advance: multimodal book boxes (Fundelius et al., 2023), circle times for all children (Gauvreau et al., 2023), science and expression with multiple means (Lohmann et al., 2023; Hovey et al., 2022), and a Liepāja preschool aligned with Guidelines 3.0 (Tīģere et al., 2025). Generative AI, when it approaches UDL, approaches as the teacher’s workshop, not as the infant’s classifier (Evmenova et al., 2024). Current rights require human oversight, the best interests of the child, an age threshold for independent conversations with generative platforms and, in the European Union, treating as high-risk the AI that determines access, admission, assignment, assessment or educational level (UNESCO, 2021; UNICEF, 2021; Miao & Holmes, 2023; European Union, 2024; European Commission, 2022).

Where the sources do not measure a kindergarten, this article does not assert it. Where they measure personalization or an access channel, it does not translate them into UDL. Accompanying children aged three to six with UDL is designing barriers in advance — multiple means of representation, of action and expression, and of engagement — with the adult as designer of the environment. The rest is an adjustment on a classified child. It is not UDL, and it must not be presented as what it is not.

Laboratorio Editorial de NEXTECH.IA / Ingeniero Mitre.

References

  1. Boysen, G. A. (2024). A critical analysis of the research evidence behind CAST’s universal design for learning guidelines. Policy Futures in Education, 22(7), 1219–1238. https://doi.org/10.1177/14782103241255428
  2. Bray, A., Devitt, A., Banks, J., Sanchez Fuentes, S., Sandoval, M., Riviou, K., Byrne, D., Flood, M., Reale, J., & Terrenzio, S. (2024). What next for Universal Design for Learning? A systematic literature review of technology in UDL implementations at second level. British Journal of Educational Technology, 55(1), 113–138. https://doi.org/10.1111/bjet.13328
  3. CAST. (2024). Universal Design for Learning Guidelines version 3.0 [graphic organizer]. CAST. https://udlguidelines.cast.org/
  4. Chen, D., & Dote-Kwan, J. (2021). Preschoolers with visual impairments and additional disabilities: Using universal design for learning and differentiation. Young Exceptional Children, 24(2), 70–81. https://doi.org/10.1177/1096250620922205
  5. Chen, J. J. (2024). A scoping study on AI affordances in early childhood education: Mapping the global landscape, identifying research gaps, and charting future research directions. Journal of Artificial Intelligence Research. https://doi.org/10.1613/jair.1.16882
  6. Edyburn, D. (2024). Research or policy – which lever pushes practice faster? A universal design for learning historical footnote: A response to Guy Boysen. Policy Futures in Education, 22(7), 1239–1260. https://doi.org/10.1177/14782103241260902
  7. European Commission. (2022). Ethical guidelines on the use of artificial intelligence (AI) and data in teaching and learning for educators. Publications Office of the European Union. https://doi.org/10.2766/153756
  8. European Union. (2024). Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 laying down harmonised rules on artificial intelligence. Official Journal of the European Union. https://eur-lex.europa.eu/eli/reg/2024/1689/oj
  9. Evmenova, A. S., Borup, J., & Shin, J. K. (2024). Harnessing the power of generative AI to support ALL learners. TechTrends, 68(4), 820–831. https://doi.org/10.1007/s11528-024-00966-x
  10. Fundelius, E., Wade, T., Robbins, A., Wang, S., McConomy, M. A., & Fumero, K. (2023). Universal design principles for multimodal representation in literacy activities for preschoolers. Inclusive Practices, 2(1). https://doi.org/10.1177/27324745221140380
  11. Gauvreau, A. N., Lohmann, M. J., & Hovey, K. A. (2023). Circle is for everyone: Using UDL to promote inclusion during circle times. Young Exceptional Children, 26(1), 3–15. https://doi.org/10.1177/10962506211028576
  12. Gravel, J. W., Tucker-Smith, N., Rose, D. H., Fornauf, B., Hughes, M. M., Lester, L., Rao, K., Singer, S., & Waitoller, F. R. (2026). UDL Guidelines 3.0: A community-driven, research-based process toward fulfilling the promise of Universal Design for Learning. International Journal of Universal Design and Universal Design for Learning, 1(2). https://doi.org/10.34874/PRSM.ijududl-vol1iss2.6592
  13. Hovey, K. A., Gauvreau, A. N., & Lohmann, M. J. (2022). Providing multiple means of action and expression in the early childhood classroom through a Universal Design for Learning framework. The Journal of Special Education Apprenticeship, 11(2), Article 7. https://scholarworks.lib.csusb.edu/josea/vol11/iss2/7
  14. King-Sears, M. E., Stefanidis, A., Evmenova, A. S., Rao, K., Mergen, R. L., Owen, L. S., & Strimel, M. M. (2023). Achievement of learners receiving UDL instruction: A meta-analysis. Teaching and Teacher Education, 122, 103956. https://doi.org/10.1016/j.tate.2022.103956
  15. Ljungcrantz, L. (2026). The interaction of AI and early childhood education. A state-of-the-art review 2020–2024. Early Childhood Education Journal, 54, 3565–3581. https://doi.org/10.1007/s10643-025-02079-3
  16. Lohmann, M. J., Hovey, K. A., & Gauvreau, A. N. (2023). Universal Design for Learning (UDL) in inclusive preschool science classrooms. Journal of Science Education for Students with Disabilities, 26(1), Article 4. https://doi.org/10.14448/jsesd.15.0005
  17. Meyer, A., & Rose, D. H. (2025). Universal Design for Learning: Principles, framework, and practice (3rd ed.). CAST Professional Publishing. https://www.cast.org/books-media/universal-design-for-learning-principles-framework-practice-third-edition/
  18. Miao, F., & Holmes, W. (2023). Guidance for generative AI in education and research. UNESCO. https://doi.org/10.54675/EWZM9535
  19. OECD. (2021). Starting Strong VI: Supporting meaningful interactions in early childhood education and care. OECD Publishing. https://doi.org/10.1787/f47a06ae-en
  20. OECD. (2023). Empowering young children in the digital age (Starting Strong). OECD Publishing. https://doi.org/10.1787/50967622-en
  21. OECD. (2025). Results from TALIS Starting Strong 2024: Strengthening early childhood education and care. OECD Publishing. https://doi.org/10.1787/20af08c0-en
  22. Rao, K., Gravel, J. W., Rose, D. H., & Tucker-Smith, N. (2023). Universal design for learning in its 3rd decade: A focus on equity, inclusion, and design. In R. J. Tierney, F. Rizvi, & K. Erkican (Eds.), International encyclopedia of education (4th ed., pp. 712–720). Elsevier. https://doi.org/10.1016/B978-0-12-818630-5.14079-5
  23. Saborío-Taylor, S., & Rojas-Ramírez, F. (2024). Universal design for learning and artificial intelligence in the digital era: Fostering inclusion and autonomous learning. International Journal of Professional Development, Learners and Learning, 6(2), ep2408. https://doi.org/10.30935/ijpdll/14694
  24. Su, J., & Yang, W. (2022). Artificial intelligence in early childhood education: A scoping review. Computers and Education: Artificial Intelligence, 3, 100049. https://doi.org/10.1016/j.caeai.2022.100049
  25. Tīģere, I., Bethere, D., Jurs, P., & Ļubkina, V. (2025). Developing inclusive preschool education for children with autism applying universal learning design strategy. Education Sciences, 15(6), 638. https://doi.org/10.3390/educsci15060638
  26. UNESCO. (2021). Recommendation on the ethics of artificial intelligence. UNESCO. https://unesdoc.unesco.org/ark:/48223/pf0000381137
  27. UNESCO. (2022). Tashkent Declaration and Commitments to Action for Transforming Early Childhood Care and Education. UNESCO. https://unesdoc.unesco.org/ark:/48223/pf0000384045
  28. UNICEF. (2021). Policy guidance on AI for children 2.0. UNICEF Innocenti. https://www.unicef.org/innocenti/reports/policy-guidance-ai-children
  29. U.S. Department of Education, Office of Educational Technology. (2023). Artificial intelligence and the future of teaching and learning: Insights and recommendations. U.S. Department of Education. https://www.ed.gov/sites/ed/files/documents/ai-report/ai-report.pdf
  30. Zhang, L., Carter, R. A., Jr., Greene, J. A., & Bernacki, M. L. (2024). Unraveling challenges with the implementation of Universal Design for Learning: A systematic literature review. Educational Psychology Review, 36, 35. https://doi.org/10.1007/s10648-024-09860-7