1. Introduction and problem

In the 3-to-6 band —kindergarten, preschool, CENDI, early childhood school— a package of five artefacts has been installed that purport to count as support for symbolic play, pretend play or sociodramatic play: a chatbot that “plays roles” or converses in character with the child; an app that generates fiction-play scripts or “pretend-play scenarios”; an autonomous voice agent —not adult-teleoperated— that directs roleplay; a computer-vision model that scores “symbolic-play quality”; and a generator that produces dramatic-play units from a prompt. All five are visible to coordination and allow centres to exhibit that they “already do pretend play with AI.” The leap —from conversational character, script, score or unit to affirming support for symbolic play— is not authorised by evidence from AI-in-ECE mappings nor by play pedagogy when there is peer role negotiation, object substitution, metacommunication, prolonged time, open-ended objects and adult scaffolding of shared imagination.

This article’s thesis is restrictive. A chatbot that “plays roles” or converses in character with the child, an app that generates fiction-play scripts or “pretend-play scenarios,” an autonomous voice agent (not adult-teleoperated) that directs roleplay, a computer-vision model that scores “symbolic-play quality,” or a generator that produces dramatic-play units from a prompt do not constitute support for symbolic play in early childhood. In the early years symbolic play is situated practice among peers (and sometimes with an adult), with open-ended objects, time, role negotiation, metacommunication and human scaffolding of shared imagination; not a generated script, a pretense score or an AI companion that replaces the peer or the adult. Thompson and Goldstein (2022) document pretend-play progression in preschool (N = 73; 34 h of observation) across five behaviours —from object substitution to metacommunication—. White, Thibodeau-Nielsen, Palermo and Mikulski (2021) show that engagement in social pretend play predicts inhibitory-control gains across the school year (N = 132 Spanish-speaking Head Start preschoolers): a finding of EF as a lateral outcome, not of a chatbot. Smits-van der Nat, van der Wilt, Meeter and van der Veen (2024) meta-analyse 34 studies (ages 3–8) and find a positive pretend play ↔ social competence relationship, with quality outweighing quantity and a majority of cross-sectional designs. That authorises asking what was measured: character, script or score —not the practice when a child says “let’s play store,” negotiates who is the seller, uses a block as a phone and sustains the episode with peers.

The problem is aggravated by five category confusions. First: symbolic play / pretend play / sociodramatic play is not generic “free play” —although free play may host it, the construct requires substitution, roles, metacommunication and shared fiction; that generic axis was already treated in this series and appears here only as a boundary—. Second: it is not artistic creativity as axis. Third: it is not orality or narration as product —telling a story is not negotiating roles with a block as a phone—. Fourth: it is not SEL as a generic programme —Smits-van der Nat et al. (2024) and White et al. (2021) show social competence and EF as lateral outcomes; they do not authorise selling SEL as if it were pretend play—. Fifth: it is not a pretense score nor automated assessment of “symbolic-play quality.” This work does not recycle articles on generic free play, generative AI and play scaffolding as a previously published axis, early scientific inquiry, STEM/robotics, numeracy, emergent literacy, orality/narration, UDL, inclusion, SEL as main topic, documentation, child participation/voice, planning, executive functions as axis, artistic creativity, music, motor skills, joint attention, adult–child interactions as axis, teacher education, environmental/sustainability or formative assessment. The question is symbolic play in the early years: what counts as such when a centre “does AI and pretend play.”

There is, moreover, a coordination economy: in-character chatbot, printed script, pretense score and dramatic-play unit fit on a slide; prolonged peer negotiation with open-ended objects and metacommunication does not. Lillard, Lerner, Hopkins, Dore, Smith and Palmquist (2013) critically anchor that strong causal evidence for unique pretend-play benefits is insufficient and that equifinality and epiphenomenon operate: a critical anchor, not “proof” of exclusive benefits. NAEYC (2022) and OECD (2021, 2023) require meaningful interactions, developmentally appropriate practice and subordinated digitalisation. Inference: support for symbolic play is not fulfilled by generating scripts or scoring pretense. The contributions are three: separate the craft from the artefact package; examine three case families —including adult-teleoperated StoryCarnival as a contrast of peripheral good use (Currin et al., 2023a, 2023b)—; and offer four tests for deciding when a kindergarten may affirm support for symbolic play, and not only chatbot, score, vision or prompt.

2. State of the art: from situated pretense practice to the artefact on display

Four strata that the “AI for pretend play in ECE” market usually mixes should be separated. The first is the construct of symbolic play / pretend play / sociodramatic play at ages 3–6 as practice of role negotiation, object substitution, attribution of pretend properties, metacommunication and prolonged peer episodes (Thompson & Goldstein, 2022; White et al., 2021; Jaggy et al., 2023; Smits-van der Nat et al., 2024; Lillard et al., 2013). The second is the pedagogical craft that cultivates it —open-ended objects, protected time, adult scaffolding of shared imagination, social pretend-play training, dramatic pretend play games— (Jaggy et al., 2023; Goldstein & Lerner, 2018, with marked transfer; Richard, Clerc-Georgy & Gentaz, 2023; Currin et al., 2023a, 2023b; NAEYC, 2022; OECD, 2021, 2023). The third is evidence on AI affordances in ECE, child-centred GenAI and story/voice-agent design —without equating them to situated peer pretense— (Chen, 2024; Su & Yang, 2022; Su & Zhong, 2022; Ljungcrantz, 2026; Nikolopoulou, 2025; Currin et al., 2023a, 2023b). The fourth is the rights, systems and developmentally appropriate practice framework that treats the 3–6-year-old as a subject of situated shared imagination, not as a signal vector for a scoring pipeline nor as a passive recipient of scripts or an autonomous AI companion (UNESCO, 2021; Miao & Holmes, 2023; U.S. Department of Education, 2023).

In construct and craft, Thompson and Goldstein (2022) observe pretend-play progression across the preschool years (N = 73; five behaviours; 34 h): empirical finding of object substitution → metacommunication. White et al. (2021) predict inhibitory-control gains from social pretend play (N = 132 Spanish-speaking Head Start): empirical finding of EF as lateral outcome. Jaggy et al. (2023) report an experiment of social pretend play → social development. Smits-van der Nat et al. (2024) meta-analyse 34 studies (ages 3–8): positive relation with social competence; quality > quantity; majority cross-sectional. Richard et al. (2023) evaluate pretend play-based training at ages 5–6. Goldstein and Lerner (2018) —marked transfer; 2018 anchor— and Vasilopoulos et al. (2026) —marked transfer for LSAC cohort— contribute lateral outcomes. Lillard et al. (2013) forbid reading strong causality or uniqueness of benefits. NAEYC (2022) and OECD (2021, 2023) anchor play and meaningful interactions. Inference: a chatbot does not negotiate roles with a block; a peer and a scaffolding adult do.

In the artefact stratum, Chen (2024), Su and Yang (2022), Su and Zhong (2022) and Ljungcrantz (2026) map affordances, curricula and the state of the art of AI in ECE without equating them to situated sociodramatic play. Nikolopoulou (2025) balances promises and challenges of child-centred GenAI and includes symbolic play / storytelling under teacher mediation: a caution framework, not authorisation of an autonomous companion chatbot. Currin et al. (2023a) document StoryCarnival with an adult-teleoperated voice agent in Montevideo and increased social play: empirical contrast of peripheral use —the adult operates the agent—. Currin et al. (2023b) draw narrative-design lessons for collaborative roleplay and situate LLMs as assisted tools, not substitutes. In the systems stratum, UNESCO (2021) requires human oversight. Miao and Holmes (2023) set pedagogical validation and age thresholds for generative AI —direct framework for the prompt-based dramatic-play unit generator—. U.S. Department of Education (2023) requires that AI support, not replace, professional judgement.

3. Review method

A critical narrative review was conducted, not a primary meta-analysis. The purpose was not to estimate a homogeneous effect size of role-companion chatbots, script apps, autonomous voice agents, pretense-quality classifiers or unit generators, but to articulate a pedagogical-category argument with verified sources. Inclusion criteria: (a) 2021–2026, with Lillard et al. (2013) and Goldstein and Lerner (2018) accepted as critical/experimental anchors, and with transfer explicitly marked when the sample or level does not equate to kindergarten ages 3–6; (b) pretend play, symbolic play, sociodramatic play, dramatic pretend play, metacommunication, object substitution, or AI in early childhood education relevant to the artefact/craft contrast; (c) relevance to kindergarten, preschool, CENDI or ages 3–6; (d) peer-reviewed journal, DOI or NAEYC, UNESCO, OECD or education-department report; (e) verifiable DOI or publisher page. Axes already used in this series were excluded as central objects —generic free play, generative AI and play scaffolding as a prior article, early scientific inquiry, STEM/robotics, numeracy, emergent literacy, orality/narration, UDL, inclusion, SEL as main programme, pedagogical documentation, child participation/voice, curricular planning, executive functions as axis, artistic creativity, music, motor skills, joint attention, adult–child interactions as axis, teacher education, environmental/sustainability and formative assessment— although some appear as category boundaries. EF and SEL appear only as lateral outcomes of social pretend play (White et al., 2021; Smits-van der Nat et al., 2024; Richard et al., 2023), not as axis.

The search was executed on 2 September 2026 (slot 13:02 America/Mexico_City) on DOI pages, Crossref, Springer, Elsevier, Taylor & Francis, ACM Digital Library, JAIR, OECD iLibrary, UNESDOC, NAEYC and publisher sites. Each source was verified against at least one of those pages. Empirical finding, conceptual or normative framework, and pedagogical inference marked as such were distinguished. Priority was given to the distinction symbolic play / generic free play / artistic creativity / orality-narration / SEL as product / pretense score, and to the caution not to translate in-character chatbots, generated scripts or pretense scores into pedagogy of role negotiation and peer metacommunication. No N, d, r, AUC or DOI were invented: when an artefact (computer vision of symbolic-play quality; autonomous role-companion chatbot) lacks a verified study in the corpus with scoring metrics at ages 3–6, it is discussed as a category ceiling supported by general AI-in-ECE mappings (Chen, 2024; Su & Yang, 2022; Ljungcrantz, 2026) and by the teleoperated StoryCarnival contrast (Currin et al., 2023a), not as an invented empirical finding.

4. Case 1. Role-companion chatbot, script app, autonomous voice agent, pretense scoring or prompt-based units do not constitute support for symbolic play

Chen (2024), Su and Yang (2022) and Ljungcrantz (2026) saturate the artefact-ceiling portrait when AI in ECE is presented as if it were support for pretend play. Chen (2024) maps global AI affordances in early childhood education: a scoping finding on emerging uses —tutoring, analytics, content generation—, not a finding that an in-character chatbot cultivates peer metacommunication. Su and Yang (2022) review the AI-in-ECE field: a synthesis finding on trends, not on object substitution or role negotiation. Ljungcrantz (2026) reviews the AI–ECE interaction 2020–2024: a state-of-the-art finding, not of situated sociodramatic play. Pedagogical inference, marked as such: the gesture “the child conversed with the chatbot in character = there was symbolic play” is a generative-conversation ceiling. Thompson and Goldstein (2022) and White et al. (2021) require social pretense with peers and observable progression; a useful dialogue with a bot may coexist with absence of negotiation among children, open-ended objects and prolonged time.

Nikolopoulou (2025) and Miao and Holmes (2023) name the risk of the dramatic-play unit generator and of pretend-play scripts from a prompt. Nikolopoulou (2025) balances promises and challenges of child-centred GenAI and includes symbolic play / storytelling under teacher mediation: a caution framework. Miao and Holmes (2023) require pedagogical validation and age thresholds for generative AI. Status: normative and review framework, not a trial of prompt-based dramatic-play units versus peer sociodramatic episodes. Inference: producing a script or unit from a prompt may be subordinated teacher preparation; support for symbolic play begins when there is “let’s play…,” negotiated roles, open-ended objects and an adult who scaffolds shared imagination without directing the script as an autonomous director. Su and Zhong (2022) propose AI curriculum design in ECE as a future direction —an AI-literacy curriculum, not situated pretend play—. Boundary inference: an AI curriculum does not sign metacommunication.

The autonomous voice agent that directs roleplay —distinct from the teleoperated agent— and the computer-vision model that scores “symbolic-play quality” lack, in the verified corpus, trials with reported N, d, r or AUC for pretense scoring or autonomous roleplay direction in preschool ages 3–6; figures are not invented. Chen (2024), Su and Yang (2022) and Ljungcrantz (2026) map analytics, agents and classification as an AI-in-ECE trend, without equivalence to sociodramatic play. Currin et al. (2023a) offer the decisive empirical contrast: StoryCarnival + adult-teleoperated voice agent in Montevideo is associated with increased social play precisely because the adult operates the agent —peripheral use, not an autonomous companion chatbot—. Currin et al. (2023b) situate LLMs as assisted tools for collaborative narrative design, not as peer substitutes. Restrictive inference: adult-teleoperated agent ≠ autonomous directing agent; assisted LLM ≠ role-companion chatbot; teacher empowerment ≠ child pretense score. UNESCO (2021) and U.S. Department of Education (2023) require human oversight. The five artefacts share one substitution grammar: the bot’s character replaces the peer; the script replaces negotiation; the autonomous agent replaces human scaffolding; the score replaces pedagogical observation; the prompt replaces prolonged time with open-ended objects.

The ceiling should be specified without inventing effects. A chatbot can converse in character; Thompson and Goldstein (2022) measure object substitution and metacommunication among children, not turns with a model. An app can print a scenario; Jaggy et al. (2023) intervene social pretend play among preschoolers, not scripts. A classifier can label “high pretense”; Smits-van der Nat et al. (2024) meta-analyse pretend-play quality in human studies, not vision AUC. Inference: “high quality according to the model” or “unit according to the prompt” is measurement or generation, not signed support for symbolic play.

5. Case 2. What the kindergarten does when there is support for symbolic play: role negotiation, open-ended objects, metacommunication and scaffolding —with teleoperated StoryCarnival as peripheral contrast

Thompson and Goldstein (2022) saturate the craft from prolonged observation (N = 73; 34 h): pretend-play progression includes object substitution, attribution of pretend properties and metacommunication —the “let’s play…” that organises the episode—. Status: empirical finding of progression. Pedagogical inference, marked as such: this is the object a centre may call support for symbolic play. Pretend play is not a pretense score: it is growing mastery of negotiating roles among peers, substituting objects, attributing pretend properties, metacommunicating the fiction frame, sustaining the episode over time and receiving adult scaffolding of shared imagination. A chatbot that marks “in-character turn” without peers may celebrate a labelled dialogue and, at once, empty the drama corner of negotiation. A quality classifier may label “pretense” on video and have failed to capture whether the adult protected time, offered open-ended objects or invited transferring the hospital role to the market.

White et al. (2021) document that engagement in social pretend play predicts inhibitory-control gains (N = 132 Spanish-speaking Head Start preschoolers): a finding of EF as lateral outcome —not this article’s axis—. Jaggy et al. (2023) show experimental impact of social pretend play on social development. Smits-van der Nat et al. (2024) confirm, across 34 studies (ages 3–8), a positive relation with social competence and primacy of quality —majority cross-sectional—. Richard et al. (2023) evaluate pretend play-based training at ages 5–6. Goldstein and Lerner (2018) and Vasilopoulos et al. (2026) —marked transfer— contribute lateral outcomes. Lillard et al. (2013) require reading them with equifinality. NAEYC (2022) and OECD (2021, 2023) anchor play and meaningful interactions.

The peripheral good-use contrast is StoryCarnival. Currin et al. (2023a) report stories and voice agents to inspire preschoolers’ social play in Montevideo: the voice agent is adult-teleoperated; increased social play is interpreted under human mediation, not under an autonomous companion chatbot. Currin et al. (2023b) design stories to inspire creative, collaborative roleplay and situate LLMs as assisted tools. Inference: subordinated digital —an adult operating the agent, a teacher using an LLM to prepare a story— may inspire peer roleplay; it does not authorise declaring that an autonomous agent directing roleplay or a companion chatbot constitutes the craft. Mediation —NAEYC (2022), OECD (2021)— is adult presence that organises time, offers open-ended objects, hears metacommunication, protects negotiation and expands without imposing a script. Inference: pedagogical observation of pretend play is professional reading, not a log of pretense scores. A dashboard does not negotiate who is the doctor; a peer and a scaffolding adult do.

6. Case 3. Symbolic play ≠ generic free play, artistic creativity, orality-narration, SEL as product or pretense score

The first category boundary is generic free play. Free play may host pretense episodes; it does not, by itself, constitute symbolic play of role negotiation, object substitution and metacommunication (Thompson & Goldstein, 2022). An in-character chatbot is also not “digital free play”: it is generative conversation. The second boundary is artistic creativity as axis: painting or collage do not equate to sociodramatic play. The third is orality/narration: telling a story may feed pretend play; it does not convert narration into role negotiation with open-ended objects (Currin et al., 2023b —narrative tool, not identity of constructs—). The fourth is SEL as product: Smits-van der Nat et al. (2024), White et al. (2021) and Richard et al. (2023) show social competence, EF and socioemotional outcomes as lateral, not construct identity. The fifth is the pretense score: scoring “symbolic-play quality” inverts the pedagogy. UNESCO (2021), Miao and Holmes (2023) and U.S. Department of Education (2023) require human oversight. Inference: the only AI use coherent with ages 3–6 remains on the adult’s side —teleoperated StoryCarnival or assisted LLM (Currin et al., 2023a, 2023b)— under pedagogical validation. It does not enter as autonomous scorer, companion chatbot, voice agent that directs without teleoperation or generator that replaces peer negotiation. Lillard et al. (2013) reinforce the epistemic boundary: without strong causal evidence of unique benefits, selling score or chatbot as a “development guarantee via pretend play” is a double invalid leap.

7. Inferential framework: four tests to affirm support for symbolic play, not an artefact

The framework that follows is this article’s pedagogical inference, anchored in the cases and verified instruments. It is not a new international standard. It distinguishes four tests. If a kindergarten, preschool, CENDI or early childhood school does not pass them, it cannot declare that a role-companion chatbot, a pretend-play script app, an autonomous voice agent, computer vision of pretense quality or a prompt-based dramatic-play unit generator constitute support for symbolic play.

7.1. Test of situated practice of role negotiation, object substitution and peer metacommunication, not of the in-character chatbot or generated script. Thompson and Goldstein (2022) and White et al. (2021) define the craft as progressive, observable social pretense. Chen (2024), Su and Yang (2022) and Ljungcrantz (2026) map chatbots and content generation as affordances, not as sociodramatic play. Inference: evidence of support for symbolic play is verified in whether the child negotiated roles, substituted objects and metacommunicated the fiction frame with peers. If the centre’s “evidence” is a chatbot dialogue or a printed script, the centre has done generative conversation or planning, not support for pretend play.

7.2. Test of prolonged time, open-ended objects and adult scaffolding of shared imagination, not of the prompt-based dramatic-play unit. Jaggy et al. (2023), Richard et al. (2023) and NAEYC (2022) situate training, implementation and developmentally appropriate practice. Miao and Holmes (2023) require pedagogical validation of generative AI. Inference: producing a unit from a prompt does not demonstrate protected time, open-ended objects or human scaffolding. A script may exist; it does not sign the sociodramatic episode.

7.3. Test of human mediation —including the peripheral contrast of the teleoperated agent—, not of the pretense dashboard or the autonomous directing agent. OECD (2021, 2023) require meaningful interactions and subordinated digitalisation. Currin et al. (2023a, 2023b) anchor teleoperated StoryCarnival and assisted LLMs. Inference: “high pretense” or “roleplay directed by the autonomous agent” may raise a threshold without raising mediation quality. Pedagogical observation reads negotiation and metacommunication; the dashboard counts a label. The adult-teleoperated agent is legitimate periphery; the autonomous director is not.

7.4. Test of category distinction and professional judgement, not of the product catalogue. Symbolic play ≠ generic free play, artistic creativity, orality/narration, SEL as product or pretense score. Smits-van der Nat et al. (2024) and White et al. (2021) forbid confusing lateral outcomes with construct identity. Lillard et al. (2013) forbid selling causal uniqueness. UNESCO (2021), Miao and Holmes (2023), U.S. Department of Education (2023), NAEYC (2022) and OECD (2021, 2023) require human oversight, pedagogical validation and not replacing professional judgement. Inference: a centre cannot treat the child as a pretense vector nor as exclusive interlocutor of a companion chatbot. Support for symbolic play is not fulfilled by scoring “pretend-play quality” better. It is fulfilled by practising peer role negotiation, object substitution, metacommunication, prolonged time, open-ended objects and adult scaffolding of shared imagination.

The framework admits digital when subordinated to teacher preparation and validated human mediation (Currin et al., 2023a, 2023b; Chen, 2024; Su & Yang, 2022; Ljungcrantz, 2026). It rejects declaring support for symbolic play via autonomous companion chatbot, script app that replaces negotiation, autonomous directing voice agent, pretense computer vision or unit generator that replaces the peer episode (Miao & Holmes, 2023; Nikolopoulou, 2025; UNESCO, 2021). The four tests are read together: passing only the first without mediation is not enough; passing only the fourth without situated practice is not enough either.

8. Discussion

Three tensions organise the discussion. The first is between exhibiting chatbot, script, autonomous agent, scoring or generation and exercising support for symbolic play. It is a finding that pretend-play progression (N = 73; 34 h), social pretend play predictive of EF (N = 132), the social-development experiment, the 34-study meta-analysis and pretend play-based training sustain the situated craft (Thompson & Goldstein, 2022; White et al., 2021; Jaggy et al., 2023; Smits-van der Nat et al., 2024; Richard et al., 2023). It is a framework that AI in ECE grows without equivalence to peer sociodramatic play (Chen, 2024; Su & Yang, 2022; Ljungcrantz, 2026) and that teleoperated StoryCarnival inspires social play under adult mediation (Currin et al., 2023a). It is not a finding that the five artefacts produce the drama-corner craft: they typify, script, direct, score or generate what engineering can deliver and declare what only mediated practice would authorise.

The second is between automated assessment of “symbolic-play quality” and everyday pretense pedagogy. Chen (2024), Su and Yang (2022) and Ljungcrantz (2026) objectify analytics and classification affordances as a trend; they do not report in the corpus verified trials validating computer-vision pretense scoring in preschool ages 3–6 —hence metrics are not invented—. Inference: insisting that the kindergarten “already supports symbolic play” because a model classifies pretense or a chatbot converses in character is inverted pedagogy. An algorithmic or conversational proxy is made to stand for role negotiation, object substitution and metacommunication. The child becomes a signal emitter for the classifier or the bot’s interlocutor; the adult, a dashboard supervisor.

The third is between generated unit/script and integral peer craft. Miao and Holmes (2023) and Nikolopoulou (2025) set GenAI limits; Currin et al. (2023b) situate LLMs as assisted. Inference: selling a prompt-based dramatic-play unit as “pretend play with AI” confuses textual product with situated practice. An autonomous agent because the system needs to direct turns does not replace the adult who protects time, offers a block as a phone and hears “you are the mum and I am the doctor.” Smits-van der Nat et al. (2024), White et al. (2021) and Lillard et al. (2013) confirm lateral outcomes and equifinality without construct identity; UNESCO (2021), Miao and Holmes (2023) and U.S. Department of Education (2023) subordinate AI to professional judgement.

The four tests in section 7 read these tensions with operational criteria: 7.1 requires negotiation, object substitution and metacommunication; 7.2 asks for time, open-ended objects and scaffolding; 7.3 distinguishes human mediation and peripheral teleoperation from the dashboard or autonomous agent; 7.4 forbids confusing symbolic play with generic free play, artistic creativity, orality/narration, SEL as product or pretense score. The empirical craft contrast —progression N = 73; social pretend play N = 132; Jaggy et al. experiment; 34-study meta-analysis; Richard et al. training; teleoperated StoryCarnival— defines the floor the five artefacts do not reach alone (Thompson & Goldstein, 2022; White et al., 2021; Jaggy et al., 2023; Smits-van der Nat et al., 2024; Richard et al., 2023; Currin et al., 2023a, 2023b). Inference: support for symbolic play erodes not by using AI, but by treating chatbot, script, autonomous agent, scoring or generation as if they were the practice. Subordinated digital is legitimate —teleoperated StoryCarnival; LLM to prepare a story with caution—; inverting the sequence and installing an artefact to declare pretend play fulfilled is not. It is verified among peers, with open-ended objects, time and a scaffolding adult (NAEYC, 2022; OECD, 2021, 2023).

9. Limits

This review is narrative. It does not apply PRISMA nor estimate primary combined effects. Thompson and Goldstein (2022) observe N = 73 (34 h), not chatbots. White et al. (2021) predict EF from social pretend play (N = 132), not chatbot-versus-peer comparison. Jaggy et al. (2023), Smits-van der Nat et al. (2024) and Richard et al. (2023) measure human craft, not artefacts. Goldstein and Lerner (2018) and Vasilopoulos et al. (2026) —marked transfer— are lateral anchors. Lillard et al. (2013) are a critical review. Currin et al. (2023a, 2023b) document teleoperation and assisted LLMs: peripheral contrast. Chen (2024), Su and Yang (2022), Su and Zhong (2022), Ljungcrantz (2026) and Nikolopoulou (2025) map AI in ECE, not pretense scoring. No verified trials of computer vision of symbolic-play quality or autonomous companion chatbots in CENDI ages 3–6 were located; they are discussed as category ceiling. NAEYC, UNESCO and OECD are framework sources. Section 7 inferences are pedagogical-category hypotheses, not implementation evidence.

10. Conclusions

A chatbot that “plays roles” or converses in character with the child, an app that generates fiction-play scripts or “pretend-play scenarios,” an autonomous voice agent (not adult-teleoperated) that directs roleplay, a computer-vision model that scores “symbolic-play quality,” or a generator that produces dramatic-play units from a prompt do not constitute support for symbolic play in an early childhood education centre. Verified evidence does not authorise that declaration. Chen (2024), Su and Yang (2022) and Ljungcrantz (2026) confirm AI affordances in ECE without equivalence to situated sociodramatic play. Nikolopoulou (2025) and Miao and Holmes (2023) set GenAI limits. Currin et al. (2023a, 2023b) require reading teleoperated StoryCarnival and assisted LLMs as peripheral contrast —teacher support ≠ companion chatbot or pretense score—. By contrast, when there is support for symbolic play in the early years, there is situated practice: object substitution → metacommunication progression (Thompson & Goldstein, 2022); social pretend play and lateral EF (White et al., 2021); social-development experiment (Jaggy et al., 2023); social-competence meta-analysis (Smits-van der Nat et al., 2024); pretend play-based training (Richard et al., 2023); DPPG and longitudinal evidence —marked transfer— (Goldstein & Lerner, 2018; Vasilopoulos et al., 2026); critical equifinality anchor (Lillard et al., 2013); DAP and interactions (NAEYC, 2022; OECD, 2021, 2023). Symbolic play is distinct from generic free play, artistic creativity, orality/narration, SEL as product and pretense score. Guidelines require human oversight and not replacing professional judgement (UNESCO, 2021; Miao & Holmes, 2023; U.S. Department of Education, 2023; Su & Zhong, 2022).

Where sources do not measure a kindergarten, this article does not affirm it. Where they measure AI mappings, GenAI or teleoperated agents, it does not translate them into pedagogical support for symbolic play. Accompanying three-to-six-year-olds in pretend play is to exercise peer role negotiation, object substitution, attribution of pretend properties, metacommunication, prolonged time, open-ended objects and adult scaffolding of shared imagination. Everything else is in-character chatbot, scenario script by prompt, autonomous directing agent, pretense scoring and algorithmic “quality” classification. It is not support for symbolic play in early childhood education, and it must not be presented as what it is not.

Laboratorio Editorial de NEXTECH.IA / Ingeniero Mitre.

References

  1. Chen, J. J. (2024). A scoping study on AI affordances in early childhood education. Journal of Artificial Intelligence Research, 81, 701–740. https://doi.org/10.1613/jair.1.16882
  2. Currin, F. H., et al. (2023a). Stories and voice agents to inspire preschool children’s social play: An experience with StoryCarnival. En Proceedings of the 22nd Annual ACM Interaction Design and Children Conference (pp. 543–547). ACM. https://doi.org/10.1145/3585088.3593893
  3. Currin, F. H., et al. (2023b). Designing stories to inspire preschoolers’ creative, collaborative roleplay. En Proceedings of the 22nd Annual ACM Interaction Design and Children Conference. ACM. https://doi.org/10.1145/3582515.3609516
  4. Goldstein, T. R., y Lerner, M. D. (2018). Dramatic pretend play games uniquely improve emotional control in young children. Developmental Science, 21(4), e12603. https://doi.org/10.1111/desc.12603
  5. Jaggy, A. K., Kalkusch, I., Bossi, C. B., Weiss, B., Sticca, F., y Perren, S. (2023). The impact of social pretend play on preschoolers’ social development: Results of an experimental study. Early Childhood Research Quarterly, 64, 13–25. https://doi.org/10.1016/j.ecresq.2023.01.012
  6. Lillard, A. S., Lerner, M. D., Hopkins, E. J., Dore, R. A., Smith, E. D., y Palmquist, C. M. (2013). The impact of pretend play on children’s development: A review of the evidence. Psychological Bulletin, 139(1), 1–34. https://doi.org/10.1037/a0029321
  7. 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
  8. Miao, F., y Holmes, W. (2023). Guidance for generative AI in education and research. UNESCO. https://doi.org/10.54675/EWZM9535
  9. NAEYC. (2022). Developmentally appropriate practice in early childhood programs serving children from birth through age 8 (4.ª ed.). NAEYC. https://www.naeyc.org/resources/pubs/books/dap-fourth-edition
  10. Nikolopoulou, K. (2025). Child-centered integration of generative AI in early learning: Balancing promises and challenges. AI, Brain and Child. https://doi.org/10.1007/s44436-025-00023-1
  11. OECD. (2021). Starting Strong VI: Supporting meaningful interactions in early childhood education and care. OECD Publishing. https://doi.org/10.1787/f47a06ae-en
  12. OECD. (2023). Empowering young children in the digital age (Starting Strong). OECD Publishing. https://doi.org/10.1787/50967622-en
  13. Richard, S., Clerc-Georgy, A., y Gentaz, E. (2023). Pretend play-based training improves some socio-emotional competences in 5–6-year-old children: A large-scale study assessing implementation. Acta Psychologica, 238, 103961. https://doi.org/10.1016/j.actpsy.2023.103961
  14. Smits-van der Nat, M., van der Wilt, F., Meeter, M., y van der Veen, C. (2024). The value of pretend play for social competence in early childhood: A meta-analysis. Educational Psychology Review, 36, Article 46. https://doi.org/10.1007/s10648-024-09884-z
  15. Su, J., y 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
  16. Su, J., y Zhong, Y. (2022). Artificial Intelligence (AI) in early childhood education: Curriculum design and future directions. Computers and Education: Artificial Intelligence, 3, 100072. https://doi.org/10.1016/j.caeai.2022.100072
  17. Thompson, B. N., y Goldstein, T. R. (2022). Observing the developmental progression of pretend play across the preschool years. Journal of Cognition and Development, 23(4), 482–502. https://doi.org/10.1080/15248372.2022.2058508
  18. 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
  19. UNESCO. (2021). Recommendation on the ethics of artificial intelligence. UNESCO. https://unesdoc.unesco.org/ark:/48223/pf0000381137
  20. Vasilopoulos, F., Grummitt, L., Bailey, S., et al. (2026). Longitudinal evidence of the relationship between pretend play and mental health in the early years. Early Childhood Education Journal, 54, 4387–4397. https://doi.org/10.1007/s10643-026-02150-7
  21. White, R. E., Thibodeau-Nielsen, R. B., Palermo, F., y Mikulski, A. M. (2021). Engagement in social pretend play predicts preschoolers’ executive function gains across the school year. Early Childhood Research Quarterly, 56, 103–113. https://doi.org/10.1016/j.ecresq.2021.03.005