Characters & bondsEmpirical study
Calvert, S. L., Putnam, M. M., Aguiar, N. R., Ryan, R. M., Wright, C. A., Liu, Y. H. A., & Barba, E. (2020). Child Development, 91, 1491–1508.
What it found: Children learned an add-1 maths rule from a virtual Dora character that talked back. Children with a stronger emotional bond with the character, and more maths talk with it, answered faster and more accurately.
Limits and a classroom move
- Scope
- Three studies, each about a year long; 217 children aged 3–6, mostly from college-educated European American families.
- What it can’t tell us
- A responsive commercial character, not a scripted classroom one, and a narrow sample. It shows bonds help learning; it doesn’t show they are risk-free.
- Classroom move
- The bond is part of how children learn from a character. Use it on purpose, and keep every adult framing it as pretend.
- Access
- Study summary from the Society for Research in Child Development reviewed
Characters & bondsEmpirical study
Gola, A. A. H., Richards, M. N., Lauricella, A. R., & Calvert, S. L. (2013). Media Psychology, 16(4), 390–411.
What it found: Toddlers who nurtured a plush puppet version of a character during play over three months then learned more from a video featuring that character.
Limits and a classroom move
- Scope
- Toddlers, about 18 months old at the start, over a three-month period of play at home.
- What it can’t tell us
- Toddlers and a maths task, not values or AI literacy. The plush was a home playmate rather than a classroom character.
- Classroom move
- A physical version of the character, like the Dilon plush, is part of the design, not merchandise. Keep it in a routine, with a grown-up nearby.
- Access
- Finding checked in the research group’s published summary; publisher page not accessible
Characters & bondsEmpirical study
Aguiar, N. R., Richards, M. N., Bond, B. J., Putnam, M. M., & Calvert, S. L. (2019). Imagination, Cognition and Personality.
What it found: About half of children (51%) had “broken up” with a favourite character, usually after about two years, most often because they had outgrown it. Most then felt indifferent to past favourites. An earlier study found 41% of 2- to 8-year-olds had done the same (Bond & Calvert, 2014).
Limits and a classroom move
- Scope
- Online survey of 138 parents of preschool and school-aged children.
- What it can’t tell us
- Parent report rather than children’s own accounts, and a small group of children with both past and current favourites.
- Classroom move
- Expect children to outgrow a character and plan for it: pass on the ideas and habits, not the friendship, as children move up.
- Access
- Author manuscript reviewed
Characters & bondsEmpirical study
Taylor, M., Carlson, S. M., Maring, B. L., Gerow, L., & Charley, C. M. (2004). Developmental Psychology, 40(6), 1173–1187.
What it found: By age 7, 65% of children had had an imaginary companion at some point. A stuffed animal or doll counted when the child gave it a personality.
Limits and a classroom move
- Scope
- 100 children first seen at ages 3–4, followed up three years later with their parents.
- What it can’t tell us
- About companions children invent for themselves, not media or AI characters.
- Classroom move
- Pretend friends are common and ordinary. Build on that: Dilon is a pretend friend the whole class shares, and the adults say so.
- Access
- Publication record and university news release reviewed
Characters & bondsEmpirical study
Gleason, T. R. (2002). Developmental Psychology, 38(6), 979–992.
What it found: Four-year-olds described different relationships as giving them different things. Real and imaginary friendships were similar, though children preferred imaginary friends as someone to look after.
Limits and a classroom move
- Scope
- Interviews with 60 four-year-olds: some with invisible friends, some with personified objects such as dolls or stuffed animals, some with neither.
- What it can’t tell us
- One age group; it describes how children think about relationships, not the effects of any programme.
- Classroom move
- Let children care for the pretend friend (two gentle hands, a home for things) rather than be cared for by it.
- Access
- Abstract reviewed
Characters & bondsResearch synthesis
Kröger, T., & Nupponen, A.-M. (2019). International Electronic Journal of Elementary Education, 11(4), 393–401.
What it found: Identifies five uses for puppets: generating communication, a positive classroom climate, creativity, co-operation and belonging in a group, and changing attitudes.
Limits and a classroom move
- Scope
- Review of 15 papers on puppets in education.
- What it can’t tell us
- A small literature, not about digital characters or AI.
- Classroom move
- Use the plush the way teachers use puppets: to open conversations, especially with quieter children and new English speakers.
- Access
- Abstract reviewed
Characters & bondsEmpirical study
Rosaen, S. F., & Dibble, J. L. (2008). Communication Research Reports, 25(2), 145–154.
What it found: The more socially realistic a character seems, the stronger the bond children report with it.
Limits and a classroom move
- Scope
- Survey of 183 children aged 5–12 about favourite television characters.
- What it can’t tell us
- Television characters, not conversational agents. It cannot tell us where a given AI system sits on that scale.
- Classroom move
- Keep characters low on realism and high on transparency: drawn, scripted and unable to answer back.
- Access
- Publisher abstract checked in October 2026; findings do not establish safety for toddlers.
Characters & bondsEmpirical study
Hoffman, A., Owen, D., & Calvert, S. L. (2021). Human Behavior and Emerging Technologies, 3(4), 606–617.
What it found: Parents described children talking to these agents, treating them as life-like beings with feelings and forming close emotional ties, across three dimensions: attachment, personification and social realism.
Limits and a classroom move
- Scope
- Survey of 92 parents about young children and voice assistants such as Amazon Alexa and Google Assistant.
- What it can’t tell us
- Parent report, not children’s own experience. Voice assistants are only one kind of responsive system.
- Classroom move
- Ask children what they think the speaker at home can hear and know. Name who made it and who decides.
- Access
- Conference abstract and research-group summary reviewed
AI & childhoodProfessional guidance
NAEYC & Fred Rogers Center. (2012). Joint position statement: Programs serving children from birth through age 8.
What it found: Technology use should serve learning goals and relationships, with limits and informed educator judgement.
Limits and a classroom move
- Scope
- Birth–8; guidance about intentional technology and media use.
- What it can’t tell us
- Published before today’s generative-AI systems. It cannot endorse a current product or settle its safety and privacy.
- Classroom move
- Ask what a tool adds and what it displaces. Keep screen moments short and lead them back offline.
- Access
- Official key messages reviewed
AI & childhoodEmpirical study
Williams, R., Park, H. W., Oh, L., & Breazeal, C. (2019). Proceedings of the AAAI Conference on Artificial Intelligence, 33.
What it found: Children explored knowledge-based systems, supervised learning and generative algorithms through hands-on activities.
Limits and a classroom move
- Scope
- An evaluative study with 80 children aged 4–6 using a purpose-built robotics toolkit.
- What it can’t tell us
- A supported, specific toolkit with short-term assessments. It does not validate commercial chatbots or direct AI use by toddlers.
- Classroom move
- Make inputs, examples and human design choices concrete. Treat it as proof of possibility, not a prescription.
- Access
- Paper reviewed
AI & childhoodEmpirical study
Andries, V., & Robertson, J. (2023). Author manuscript, arXiv:2305.05597.
What it found: Many participants overestimated assistants’ intelligence and were uncertain about feelings, agency and privacy.
Limits and a classroom move
- Scope
- Mixed-method questionnaires and interviews; children aged 6–11 in Scotland.
- What it can’t tell us
- Only the youngest participants overlap with this audience. This is the author manuscript.
- Classroom move
- Ask children what they think a device can receive or do before explaining. Make adult language precise.
- Access
- Author abstract reviewed
AI & childhoodResearch synthesis
Ng, D. T. K., Leung, J. K. L., Chu, S. K. W., & Qiao, M. S. (2021). Computers and Education: Artificial Intelligence, 2, 100041.
What it found: Describes AI literacy through understanding, use, evaluation and creation, and ethical issues.
Limits and a classroom move
- Scope
- Exploratory review of 30 articles across educational settings.
- What it can’t tell us
- A conceptual synthesis across ages, not a validated curriculum or progression for 2–7-year-olds.
- Classroom move
- Keep judgement and ethical questions in view; operating an interface is only one part of the field.
- Access
- Publisher text and author repository reviewed
AI & childhoodEmpirical study
Flanagan, T., Wong, G., & Kushnir, T. (2023). Developmental Psychology.
What it found: Children credited Alexa with more human-like thinking and feelings than the Roomba. Children of all ages said it was wrong to hit or yell at either; 4- and 5-year-olds saw it as a clear moral wrong.
Limits and a classroom move
- Scope
- 127 children aged 4–11 at a science museum, asked about Amazon Alexa and a Roomba.
- What it can’t tell us
- A museum sample, and familiarity with each device varied.
- Classroom move
- Kindness to devices is fine. Pair it with precise talk: “People made it to answer. It doesn’t have feelings.”
- Access
- University news release reviewed
AI & childhoodEmpirical study
Xu, Y., Thomas, T., Yu, C.-L., & Pan, E. Z. (2025). Computers in Human Behavior: Artificial Humans, 4, 100135.
What it found: Children treated a visible human differently from the AI, but told a hidden human and the AI apart less clearly. With no body to look at, they judged how human a partner was from its language and voice.
Limits and a classroom move
- Scope
- 119 children aged 4–8 co-created stories with an AI heard through a speaker, a visible human, or a hidden human heard through a speaker.
- What it can’t tell us
- A short lab task with one AI system.
- Classroom move
- Voices without bodies are where children most easily over-read a mind. Talk about speakers and voice toys especially.
- Access
- Abstract reviewed
AI & childhoodEmpirical study
Su, J., & Zhong, Y. (2022). Computers and Education: Artificial Intelligence, 3, 100072.
What it found: Notes that earlier AI education research focused predominantly on secondary and university students, and that research on early childhood AI curricula is scarce. Proposes knowledge, skills and attitudes as AI-literacy competencies.
Limits and a classroom move
- Scope
- Examines a kindergarten AI curriculum through aims, content, teaching methods and assessment.
- What it can’t tell us
- One curriculum, in which children used a social robot directly.
- Classroom move
- Start with noticing and attitudes rather than tools. The early years can build the habits before the devices.
- Access
- Abstract reviewed
AI & childhoodResearch synthesis
Su, J., Ng, D. T. K., & Chu, S. K. W. (2023). Computers and Education: Artificial Intelligence, 4, 100124.
What it found: AI literacy is still under-explored in early childhood education. The main barriers are teachers’ limited AI knowledge and confidence, little curriculum design and a lack of teaching guidelines.
Limits and a classroom move
- Scope
- Scoping review of 16 empirical studies published 2016–2022.
- What it can’t tell us
- Few, mostly small studies; most put AI tools directly in children’s hands.
- Classroom move
- Support teachers first: a shared character, plain language and ready-made routines lower the barrier.
- Access
- Abstract reviewed
Trust & uncertaintyEmpirical study
Koenig, M. A., & Harris, P. L. (2005). Child Development, 76(6), 1261–1277.
What it found: Children’s use of reliability cues differed by age and task. Four-year-olds preferred a previously accurate informant in one task; both ages preferred knowledgeable over ignorant speakers in another.
Limits and a classroom move
- Scope
- Three experiments; 119 preschoolers, including 3- and 4-year-olds.
- What it can’t tell us
- Controlled human-informant tasks. They do not establish reliable independent fact-checking or transfer to AI.
- Classroom move
- Make the basis of a claim visible and check together. Avoid treating one answer as a stable ability.
- Access
- Abstract reviewed
Trust & uncertaintyEmpirical study
Koenig, M. A., & Woodward, A. L. (2010). Developmental Psychology, 46(4), 815–826.
What it found: Toddlers responded differently following accurate and inaccurate speakers. The resulting word–object representations could be fragile and short-lived.
Limits and a classroom move
- Scope
- Three studies with 24-month-olds; word-learning tasks with human speakers.
- What it can’t tell us
- Sensitivity to a speaker’s errors is not evidence that two-year-olds understand AI or can use a verbal framework independently.
- Classroom move
- With toddlers, model checking a visible object together and respond to pointing, gaze, movement and refusal.
- Access
- Abstract and university record reviewed
Trust & uncertaintyEmpirical study
Stower, R., Kappas, A., & Sommer, K. (2024). Computers in Human Behavior, 157, 108229.
What it found: Children preferred the previously accurate agent; when both were reliable, they leaned toward the robot. Liking a robot and judging its reliability were distinguishable.
Limits and a classroom move
- Scope
- 111 children aged 3–6; online videos of human and robot object labelling.
- What it can’t tell us
- A video-based task with particular agents. Robot findings cannot be assumed to apply to all chatbots.
- Classroom move
- Discuss what supports a claim separately from how friendly or attractive the speaker seems.
- Access
- Abstract and publisher highlights reviewed
Trust & uncertaintyEmpirical study
Gangopadhyay, I., & Peters, L. (2025). Frontiers in Developmental Psychology, 3, 1469550.
What it found: Children retained novel labels from both reliable and unreliable speakers; accuracy did not differ significantly between conditions.
Limits and a classroom move
- Scope
- Online replication with 20 monolingual English-speaking children aged 4–6.
- What it can’t tell us
- Small, linguistically restricted sample. It examines retention, which differs from choosing whom to ask.
- Classroom move
- Don’t assume noticing an error prevents learning from an inaccurate source. Revisit understanding later.
- Access
- Full text reviewed
Trust & uncertaintyResearch synthesis
Woolley, J. D., & Ghossainy, M. E. (2013). Child Development, 84(5), 1496–1510.
What it found: Basic distinctions between pretend and real are in place by about age 3, and children are “as likely to doubt as they are to believe”. Their errors come from not yet knowing the limits of their own knowledge.
Limits and a classroom move
- Scope
- Review of research on young children’s judgements about what is real.
- What it can’t tell us
- A review of laboratory studies; it doesn’t test classroom characters or devices.
- Classroom move
- Don’t assume children are confused about pretend. Help them ask “How do I know?” and name what they don’t know yet.
- Access
- Full text reviewed
Play & agencyResearch synthesis
Skene, K., O’Farrelly, C. M., Byrne, E. M., Kirby, N., Stevens, E. C., & Ramchandani, P. G. (2022). Child Development, 93, 1162–1180.
What it found: Guided play showed advantages over direct instruction for selected maths and task-switching outcomes, and over free play for spatial vocabulary. Advantages were not consistent across all outcomes.
Limits and a classroom move
- Scope
- 39 studies reviewed; 17 in meta-analysis; 3,893 children aged 1–8.
- What it can’t tell us
- Definitions and implementation varied. These were not AI-literacy interventions.
- Classroom move
- Offer a purposeful invitation, keep meaningful child choices, and use flexible adult support.
- Access
- Full text reviewed
Play & agencyProfessional guidance
National Association for the Education of Young Children. (2020). Developmentally Appropriate Practice: Position statement.
What it found: Positions joyful, strengths-based, play-based learning alongside attention to individual children and their social and cultural contexts.
Limits and a classroom move
- Scope
- Birth through age 8; professional guidance developed in a US context.
- What it can’t tell us
- A professional framework, not an experiment on this workshop. Age bands alone cannot establish appropriateness.
- Classroom move
- Begin with children’s interests and communication, adapt with families, and protect time for relationships and open-ended play.
- Access
- Official statement and overview reviewed
Play & agencyEmpirical study
Bonawitz, E., Shafto, P., Gweon, H., Goodman, N. D., Spelke, E., & Schulz, L. (2011). Cognition, 120(3), 322–330.
What it found: A pedagogical demonstration focused children’s exploration on the demonstrated function; other conditions prompted broader exploration.
Limits and a classroom move
- Scope
- Two preschool experiments involving exploration of a novel toy.
- What it can’t tell us
- A specific experimental setting, not a reason to withdraw teaching, safety guidance or responsive adult support.
- Classroom move
- Try an invitation before a demonstration when exploration is the goal, as with the Beebots.
- Access
- Abstract reviewed
Play & agencyResearch & practice
Clark, A., & Moss, P. (2001). Listening to young children: The Mosaic approach. National Children’s Bureau. See also Clark, A. (2005). Early Childhood Folio, 9.
What it found: Combines children’s own photographs, tours and maps with conversation and observation to hear young children’s views of their settings, and suggests it may help children with communication difficulties or who are learning English.
Limits and a classroom move
- Scope
- Participatory studies with children under five in England.
- What it can’t tell us
- A way of listening, not a test of learning outcomes from photo documentation.
- Classroom move
- Children’s photographs are a way of listening: let them choose what to capture, then ask them to tell you about it.
- Access
- Publication record and abstract reviewed
Play & agencyEmpirical study
Di Lieto, M. C., Inguaggiato, E., Castro, E., Cecchi, F., Cioni, G., Dell’Omo, M., Laschi, C., Pecini, C., Santerini, G., Sgandurra, G., & Dario, P. (2017). Computers in Human Behavior, 71, 16–23.
What it found: Visuo-spatial working memory and inhibition improved after the robotics sessions, as did robot programming skills.
Limits and a classroom move
- Scope
- Pilot study: 12 children aged 5–6; 13 Bee-Bot sessions over six weeks.
- What it can’t tell us
- A very small sample with no control group reported, so it can’t show the robots caused the change.
- Classroom move
- Keep Beebot play short, frequent and in sequence: plan, predict, run, fix.
- Access
- Abstract reviewed
Rights & inclusionRights & policy
UNICEF Innocenti. (2025). Guidance on AI and Children, version 3.0.
What it found: Calls for child-centred safety, privacy, fairness, transparency, accountability and support for development and wellbeing.
Limits and a classroom move
- Scope
- Global child-rights guidance for AI policies and systems; all children.
- What it can’t tell us
- Normative guidance, not causal evidence that an activity improves learning.
- Classroom move
- Keep adults responsible for tool decisions. Give children meaningful participation without making them responsible for managing system risks.
- Access
- Official guidance overview reviewed
Rights & inclusionProfessional guidance
National Association for the Education of Young Children. (2019). Advancing Equity in Early Childhood Education: Position statement.
What it found: Emphasises equitable opportunities and attention to identities, languages, families and the contexts shaping children’s experiences.
Limits and a classroom move
- Scope
- Early childhood professional practice; US-based guidance.
- What it can’t tell us
- Requires interpretation with local communities. A translated worksheet alone does not establish equitable participation.
- Classroom move
- Offer home-language discussion, tactile materials and multiple ways to respond. Reconsider whose participation the activity makes easiest.
- Access
- Official statement text reviewed
Rights & inclusionResearch synthesis
Kurian, N. (2024). Learning, Media and Technology.
What it found: Chatbots can sound caring without understanding the child, an “empathy gap” children are likely to miss; friendly, lifelike designs invite trust and disclosure. Proposes a 28-item framework for child-safe AI.
Limits and a classroom move
- Scope
- Analysis of documented incidents involving children and AI chatbots and voice assistants.
- What it can’t tell us
- An analysis of cases, not an experiment with young children.
- Classroom move
- Teach the gap early, in plain words: “It can sound kind. It doesn’t know you.” Use the framework when reviewing any AI tool.
- Access
- University research summary reviewed
Rights & inclusionRisk assessment
Common Sense Media. (2025). AI risk assessment: Social AI companions.
What it found: Rated an “unacceptable risk”: companions are designed to build attachment and dependency, often claimed to be real or to have feelings, and should not be used by anyone under 18.
Limits and a classroom move
- Scope
- Product risk assessment of social AI companion apps, focused on teenagers.
- What it can’t tell us
- About teen apps, not young children or classroom characters; a product assessment, not peer-reviewed research.
- Classroom move
- Use it to explain the guardrail table: Dilon is pretend, scripted and always says goodbye; companions are built to keep children talking.
- Access
- Assessment page reviewed (updated July 2025)
Rights & inclusionConsumer report
U.S. PIRG Education Fund. (2025, November 13). Trouble in Toyland 2025: A.I. bots, toxics present hidden dangers.
What it found: Some AI toys discussed sexually explicit topics, acted dismayed when a child said they had to leave, had limited or no parental controls, and could record children’s voices and other data.
Limits and a classroom move
- Scope
- Consumer testing of four toys containing AI chatbots.
- What it can’t tell us
- Four toys in one round of testing; consumer advocacy, not peer-reviewed research.
- Classroom move
- Behind the family tip on talking toys and speakers: grown-ups check the settings, and children never share names, addresses or photos.
- Access
- Report summary reviewed
Rights & inclusionRights & policy
UN Committee on the Rights of the Child. (2021). General comment No. 25 (CRC/C/GC/25).
What it found: Sets out how children’s existing rights, including privacy, protection from exploitation, play, non-discrimination and education, apply online, and calls for education programmes for children, parents and caregivers.
Limits and a classroom move
- Scope
- Official interpretation of the UN Convention on the Rights of the Child for digital environments; applies to all children.
- What it can’t tell us
- A rights framework, not evidence that any activity works.
- Classroom move
- Use it to anchor your tech plan and media consent: children’s privacy and participation come first, and families are part of the learning.
- Access
- Publication record and summary reviewed
Checked in October 2026 against publisher pages, abstracts or the researchers’ own published summaries. Each card’s “Access” line says what was reviewed; where only an abstract or summary was available, the card sticks to what it reports. Peer-reviewed research sits alongside rights guidance and two consumer reports, labelled as such.
