CoChat
Sign Up
Cited by 17Innovative Human-Technology InteractionChild Development and Digital TechnologyEthics and Social Impacts of AI

Child-AI Co-Creation: A Review of the Current Research Landscape and a Proposal for Six Design Considerations

Open access
Cai, Zhenyao, Han, Ariel, Zhou, Xiaofei, Gazulla, Eva Durall, Peppler, Kylie·2025·Proceedings of the 24th Interaction Design and Children
DOI 10.1145/3713043.3731506
Open this paper in CoChat — free accountOpen PDF

Abstract

As generative AI becomes increasingly integrated into children's creative lives, designing responsible and meaningful tools for Child-AI co-creation is a growing concern in HCI and education.This paper presents a scoping review of 20 studies on Child-AI co-creation, analyzing the types of creative activities involved, the age groups studied, and the roles AI systems play.Based on this synthesis, we identify six key design considerations: protecting child data privacy, minimizing bias and hallucinations, fostering appropriate reliance on AI, balancing support and creative freedom, encouraging peer and family collaboration, and making AI's creative process understandable.This work contributes an initial framework to guide the design of child-centered and developmentally appropriate AI co-creation systems, and highlights directions for future research.

Integrity signals

Registry

Crossref · 10.1145/3713043.3731506

Cites

57 of 57 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

0 of 68 cited sources checked · 0 supported · 0 flagged

  • Findingown data

    The scoping review identified 20 relevant studies on child-AI co-creation.

    “This paper presents a scoping review of 20 studies on Child-AI co-creation, analyzing the types of creative activities involved, the age groups studied, and the roles AI systems play.”
    • Also stated in: § 3 Method · plain
      “In total, 20 studies were included in the final analysis.”
    Quote in the paperSubstantial claim
  • Findingown data

    Two of 20 reviewed studies included teens aged 13-20.

    “or teens (13-20) (two out of 20 studies)”
    Quote in the paperNot a substantial claim
  • Findingown data

    Most reviewed studies (13 of 20) involved children aged 7-12.

    “Most studies focused on children aged 7-12 (thirteen out of 20 studies)”
    Quote in the paperNot a substantial claim
  • Findingown data

    Storytelling was a common activity in nine of 20 reviewed studies.

    “Common activities included storytelling (oral and written stories) (nine out of 20 studies)”
    Quote in the paperNot a substantial claim
  • Findingown data

    Visual creation was a common activity in nine of 20 reviewed studies.

    “and visual creation (drawing and other artistic creation) (nine out of 20 studies)”
    Quote in the paperNot a substantial claim
  • Classified as probably not a substantial claim. 33
  • May combine two claims. 15

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    Eight of 20 reviewed studies included children aged 3-7.

    “with some including younger children (3-7) (eight out of 20 studies)”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingcitation

    The reviewed papers raised concerns about using children's data to train large language models and about unauthorized data retention.

    “raising concerns about the use of children's data to train large language models and the potential issues of unauthorized data retention.”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites [2] source

      Not checked in the free run.

    • cites [61] source

      Not checked in the free run.

  • May combine two claims.

    Findingcitation

    Several reviewed papers raised concerns about bias, harmful content, and hallucinations in AI-generated outputs, which can affect children's learning, self-perception, and trust in technology.

    “Several papers raised concerns about bias, harmful content, and hallucinations in AI-generated outputs, which can significantly affect children's learning, self-perception, and trust in technology.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    In the Collaborator role, both AI and child actively contribute to the final product, building on each other's contributions.

    “3) Collaborator: Both the AI and the child actively contribute to the final product, building on each other's contributions.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    In the Enhancer role, AI works from a child-generated base and augments or transforms it.

    “4) Enhancer: The AI works from a child-generated base and augments or transforms it, such as converting a drawing into a 3D model.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingnone given

    In the Generator role, AI produces content with minimal input from the child.

    “1) Generator: The AI produces content with minimal input from the child.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    In the Facilitator role, AI does not directly contribute to the final artifact but supports the child through prompts, questions, hints, or brainstorming.

    “2) Facilitator: The AI does not directly contribute to the final artifact but supports the child through prompts, questions, hints, or brainstorming ideas.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    Image generation tools such as DALL•E, Stable Diffusion, and LCM were used in the reviewed studies.

    “followed by image generation tools such as DALL•E, Stable Diffusion, and LCM.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    Backward and forward snowballing of key papers yielded six additional studies.

    “To supplement the database search, we also conducted backward and forward snowballing of key papers, which yielded six additional studies.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Findingnone given

    In the reviewed papers, the boundaries of what stage of creation a tool intends for are blurry.

    “In the reviewed papers however, the boundaries of what stage of creation it intends for is blurry.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    After removing 11 duplicates, 116 unique studies remained.

    “After removing 11 duplicates, 116 unique studies remained.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    All included publications were published in 2019 or later.

    “All included publications were published in 2019 or later.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    The initial search resulted in 127 studies (113 from Web of Science and 14 from ACM Digital Library).

    “The initial search resulted in 127 studies (113 from Web of Science and 14 from the ACM Digital Library).”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Findingreasoning

    The definition of appropriate reliance remains somewhat ambiguous in the reviewed studies.

    “However, the definition of “appropriate reliance” remains somewhat ambiguous.”
    • Also stated in: § 4.2.3 Foster Appropriate Reliance on AI and · plain
      “What counts as appropriate is rarely articulated clearly in the reviewed studies.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    The first author screened titles and abstracts, resulting in 24 papers selected for full-text review.

    “The first author screened titles and abstracts for eligibility, which resulted in 24 papers selected for full-text review.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    AI unpredictability may confuse children, disrupt their creative flow, or lead them to accept inaccurate outputs as correct, especially when AI is perceived as a knowledgeable partner.

    “This unpredictability may confuse children, disrupt their creative flow, or lead them to accept inaccurate outputs as correct-especially when the AI is perceived as a knowledgeable partner.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    In one reviewed study [61], AI created unusual connections between common story elements, such as using clouds to add speed to a journey.

    “In [61], for example, the authors mentioned that AI created unusual connections between common story elements, such as “using clouds to add speed to a journey.””
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [61] source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    One reviewed study [6] explored how image generation models reflect stereotypical representations of gender and race, encouraging children to reflect critically on AI's limitations.

    “For example, [6] explicitly explored how image generation models reflect stereotypical representations of gender and race, encouraging children to reflect critically on the AI's limitations.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [6] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationcitation

    Intended constraints are deliberately designed by developers, such as AI-generated suggestions or facilitation strategies that may steer creativity in a specific direction.

    “Intended constraints are those deliberately designed by developers, such as AI-generated suggestions or facilitation strategies that, while helpful, may end up steering creativity in a specific direction [62].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [62] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    Kant defined creative genius as the ability to produce works that are both original (novel) and exemplary (valuable), a definition accepted in HCI research.

    “Early philosophers such as Kant define creative genius, which is also accepted in HCI research [3,37], as the ability to produce works that are not only “original” (novel) but also “exemplary” (valuable) ( [28] as cited in [42]).”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites [3] source

      Not checked in the free run.

    • cites [37] source

      Not checked in the free run.

    • cites [42]

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    Self-directed engagement with GenAI has potential benefits, such as expanding access to new forms of expression and augmenting creativity.

    “While such self-directed engagement has potential benefits, as suggested by recent articles, such as expanding access to new forms of expression and augmenting creativity [16,18]”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [16] source

      Not checked in the free run.

    • cites [18] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Divergent thinking, as proposed by Torrance, includes four components: flexibility, fluency, originality, and elaboration.

    “A key cognitive process in creativity is divergent thinking, which, as what Torrance proposed, includes four components: flexibility (the ability to shift between categories of ideas), fluency (the number of ideas), originality (the uniqueness of ideas), and elaboration (the level of detail) [49].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [49]

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

    Premisecitation

    Creativity is a complex phenomenon that requires integrating a variety of theoretical perspectives to understand.

    “Creativity is a deeply complex phenomenon, and requires to integrate a variety of theoretical perspectives to understand [42].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [42]

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationreasoning

    These constraints on creativity can be either intended or unintended.

    “These constraints can be either intended or unintended.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    In Kantosalo and Riihiaho [29], children who co-wrote poems with AI reflected that something felt 'missing' compared to working with peers, suggesting a lack of social cues and shared meaning-making typical in peer collaboration.

    “In the study by Kantosalo and Riihiaho [29], children who co-wrote poems with AI reflected that something felt “missing” compared to working with peers-suggesting the lack of social cues, and shared meaning-making typical in peer collaboration.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [29] source

      Not checked in the free run.

  • May combine two claims.

    Interpretationcitation

    These forms of social interaction help with development of creativity and learning and might not be replaceable by AI.

    “These forms of social interaction, which help with development of creativity and learning, might not be able to be replaced by AI, which underscores the importance of designing AI systems that preserve space for human connection, rather than encouraging isolated AI use.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisecitation

    Self-directed engagement with GenAI presents risks related to privacy, safety, and ethical considerations and may influence children's cognitive and social development in complex ways.

    “it also presents risks related to privacy, safety, and ethical considerations and may influence children's cognitive and social development in complex ways [8,34,38,41].”
    Quote in the paperSubstantial claim0 of 3 sources checked
    • cites [8] source

      Not checked in the free run.

    • cites [34] source

      Not checked in the free run.

    • cites [38] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    A Common Sense Media report found that nearly half of children self-reported using GenAI tools for activities such as writing documents or creating multimedia content.

    “A recent Common Sense Media report found that nearly half of children (between [13][14][15][16][17][18] self-reported using GenAI tools for these activities, such as writing documents or creating other multimedia content 1 .”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Premisecitation

    GenAI's core ability is recognizing patterns from trained data and generating content, allowing human-AI co-creative tasks like writing, drawing, and filmmaking.

    “One of the core abilities of GenAI is recognizing patterns from trained data and generating content [1], allowing for human-AI co-creative tasks and activities like writing [21,22,31], drawing [19,36], filmmaking [7,27,56,57] and more.”
    Quote in the paperSubstantial claim0 of 3 sources checked
    • cites [1]

      Not checked in the free run.

    • cites [21] source

      Not checked in the free run.

    • cites [22] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductcitation

    The review followed PRISMA guidelines for screening and selection.

    “We followed the PRISMA guidelines for screening and selection [40].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [40]

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    In HCI, Shneiderman identified four stages of creative tasks and eight activities that tools can support, including consulting with peers and composing artifacts.

    “Shneiderman [20,47] identified four stages of creative tasks and eight activities that tools can support, including consulting with peers and composing artifacts, and that tools can be designed differently depending on which activity or stage of co-creation they target.”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites [20] source

      Not checked in the free run.

    • cites [47] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Recent CHI and IDC workshops examine how to translate child-centered AI principles into practice.

    “This conversation has been extended in recent CHI and IDC workshops, which examine how to translate these principles into practice [5,54]”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [5] source

      Not checked in the free run.

    • cites [54] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    Xu [58] suggests that children engage in more complex conversations with humans, particularly because humans can use nonverbal cues such as nodding and eye contact, which AI currently lacks.

    “Similarly, Xu [58], in her article in Child Development Perspectives, suggests that children engage in more complex conversations with humans, particularly because humans can use nonverbal cues (e.g., nodding, eye contact), which AI currently lacks.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [58] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationnone given

    Some design considerations (e.g., #1, #2) are broadly applicable to generative AI tools for children, such as tutoring platforms and educational assistants.

    “Some of these considerations (e.g., #1, #2) are broadly applicable to generative AI tools for children, such as tutoring platforms and educational assistants.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Wang et al. [55] synthesized ten design principles across six regulatory frameworks, emphasizing fairness, transparency, privacy, and developmental alignment.

    “Wang et al. [55] synthesized ten design principles across six regulatory frameworks, emphasizing themes such as fairness, transparency, privacy, and developmental alignment.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [55] source

      Not checked in the free run.

  • Classified as probably not a substantial claim. May combine two claims.

    Premisecitation

    One reviewed study [32] found that teens viewed using AI to generate the entire output as inappropriate but supported its use for brainstorming and inspiration.

    “One study [32] found that teens viewed using AI to generate the entire output as inappropriate, but supported its use for brainstorming and inspiration.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [32] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Researchers have observed a 'fourth-grade slump,' a period during 3rd-4th grade when children's divergent thinking abilities tend to decline.

    “In child development, researchers have observed the “fourth-grade slump”, a period during 3th-4th grade when children's divergent thinking abilities tend to decline [50].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [50] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductown data

    Full-text screening was conducted by two authors, with discrepancies resolved through discussion.

    “Full-text screening was then conducted by two authors, with discrepancies resolved through discussion.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductreasoning

    The review did not detail which specific challenges appeared in each paper or explicitly trace how those challenges informed each design recommendation.

    “However, given the scope of this work-in-progress, we did not detail which specific challenges appeared in each paper or explicitly trace how those challenges informed each design recommendation.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

    Study design and conductnone given

    The review remains largely descriptive.

    “Second, while our review identifies key trends, it remains largely descriptive.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    The proposed design considerations' real-world implementation remains speculative.

    “Finally, although the proposed design considerations offer practical starting points, their real-world implementation remains speculative.”
    Quote in the paperNot a substantial claim
41 of 41

  • The paper relies on this without support.

    The 20 included studies are representative enough of the current child-AI co-creation research landscape to support the synthesis.

    “we reviewed the current research landscape on child-AI co-creation and identified 20 relevant papers”
    “In total, 20 studies were included in the final analysis.”

  • The paper relies on this without support.

    The challenges identified across the reviewed studies can be addressed through the proposed design considerations.

    “what design considerations can help mitigate them?”
    “derived six design considerations to inform future research”

  • The paper relies on this without support.

    The four AI roles identified are exhaustive of the roles AI can play in child-AI co-creation.

    “identified four distinct roles”

  • The paper relies on this without support.

    The 20 reviewed studies are representative of the current research landscape on child-AI co-creation.

    “This paper presents a preliminary synthesis of 20 studies on child-AI co-creation, examining how children engage in creative activities with AI systems.”

  • The paper relies on this without support.

    AI's creative process can be made transparent and understandable to children.

    “Designing AI that makes its creative process transparent and understandable to children.”
5 of 5

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

  • The model table lists two rule-based studies, not one

    Minor

    The results call study [29] the unique rule-based study, but the model table classifies both [15] and [29] as rule-based.

  • The reported childhood age bands do not match the table

    Minor

    The results use ages 3-7 and 7-12, whereas the age table labels the corresponding groups 3-8 and 9-12. The boundary ages therefore differ between the two accounts.

2 of 2

Cites

Flagged first, then by citations

Research works printed but not linked: 5

  1. Josh Achiam Steven Adler Sandhini Agarwal Lama Ahmad Ilge Akkaya Florencia Leoni Aleman Diogo Almeida Janko Altenschmidt Sam Altman Shyamal Anadkat et al. 2023. Gpt-4 technical report. arXiv preprint arXiv:https://arXiv.org/abs/2303.08774 (2023). · arXiv
  2. Zhenyao Cai, Shiyao Wei, Ariel Han, and Kylie Peppler. 2025. "Hi Kids, Let’s Talk About How Snakes Hunt": Understanding Children’s Engagement and Challenges in Educational Video Creation. In Proceedings of the 24th Annual ACM Interaction Design and Children Conference). ACM, (To appear).
  3. Callie Y. Kim, Arissa J. Sato, Nathan Thomas White, Hui-Ru Ho, Christine P. Lee, Yuna Hwang, and Bilge Mutlu. 2025. Bridging Generations using AI-Supported Co-Creative Activities. In CHI Conference on Human Factors in Computing Systems (CHI ’25). ACM, Yokohama, Japan, 1–15.
  4. Elliot Samuel Paul and Dustin Stokes. 2024. Creativity. In The Stanford Encyclopedia of Philosophy (Spring 2024 ed.), Edward N. Zalta and Uri Nodelman (Eds.). Metaphysics Research Lab, Stanford University.
  5. Lymanshan Ye Jiandong Jiang Yuhan Liu Yihan Ran Pengfei Liu and Danni Chang. 2024. Colin: A Multimodal Human–AI Co-Creation Storytelling System To Support Children’s Multi-Level Narrative Skills. arXiv preprint arXiv:https://arXiv.org/abs/2405.06495 (2024). https://arxiv.org/abs/2405.06495 Last revised 17 Mar 2025. · arXiv

Cited by

Sorted by citations

Related

Authors
Cai, Z., Han, A., Zhou, X., Gazulla, E. D., & Peppler, K.
Affiliations
  • University of California, Irvine (US)
  • University of Southern California (US)
  • University of Rochester (US)
  • INTERACT, University of Oulu, Oulu, Finland
ORCID
Venue
Proceedings of the 24th Interaction Design and Children
Publisher
ACM
Published
2025-06-23
Pages
916-922
DOI
10.1145/3713043.3731506
OpenAlex
W4411558490
Semantic Scholar
e4c180de20cdc7e897dff586c482b18f4fda380d
Cited by
17
References
57
Citations by year
2026: 15 · 2025: 2
FWCI
16.93
Type
conference paper
Language
en
Open access
Locations
ACM International Conference on Interaction Design and Children (IDC) (publishedVersion)
Topics
  • Physical Sciences › Computer Science › Human-Computer Interaction › Innovative Human-Technology Interaction
  • Social Sciences › Education › Child Development and Digital Technology
  • Social Sciences › Safety Research › Ethics and Social Impacts of AI
Keywords
generative AI, children's creative activity, AI reliance, creative freedom, explainable AI

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-06.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-06.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review