Raising Children in the Age of AI · Edition 4

Keep the roots
in view.

The 14 selected research records cited in this edition. Research findings, editorial interpretation, and proposed family practices serve different purposes.

Download this edition’s evidence record ↓

PMC13036395 · Umbrella review

Annual Research Review: The role of caregiver sensitivity in children's developmental outcomes – an umbrella review ↗

Marissa D Nivison; Pasco Fearon; Jennifer M Jenkins; Sheri Madigan

2026-01-07 · Seventeen meta-analyses of caregiver sensitivity and related parenting constructs
Age / stage: Childhood; age ranges vary across included meta-analyses

Read the catalogue abstract

AbstractCaregiver sensitivity is the extent to which a caregiver notices a child's signal, interprets it correctly, and responds quickly and appropriately. Although originally introduced to developmental science as the key antecedent of attachment security, decades since its conception, hundreds of studies have been conducted examining the predictive significance of caregiver sensitivity to a broad range of developmental outcomes. The literature on caregiver sensitivity and related constructs (e.g., warmth, responsivity, negative parenting) has grown exponentially and is now the focus of several meta‐analyses. We conducted an umbrella review – a systematic review of reviews – to examine the extent to which caregiver sensitivity and related constructs are associated with child attachment, socioemotional, and cognitive outcomes. Searches in EMBASE, PsycINFO, and Medline and yielded 2,157 abstracts. Studies were included if they were a meta‐analysis of caregiver sensitivity or a related construct, focused on children's developmental outcomes, were available in English, French, or Spanish, and were published between 2010 and 2024. Conducted and reported in accordance with PRISMA guidelines, 17 meta‐analyses were identified. Using the metaumbrella package in R, we conducted quantitative analyses which demonstrated that caregiver sensitivity was moderately associated with attachment security (r = .25, k = 253, n = 37,444), cognition (r = .23, k = 44, n = 6,777), language skills (r = .26, k = 54, n = 11,136), and weakly associated with socioemotional problems (r = −.07, k = 135, n = 33,305). Narrative analysis of other meta‐analyses on caregiver warmth, responsivity, positive and negative parenting, and child outcomes also showed associations in the expected direction. Our findings demonstrate the critical importance of caregiver sensitivity on children's socioemotional and cognitive development, supporting caregiver sensitivity as an important target for early childhood prevention and intervention programs.Keywords: Caregiver sensitivity, umbrella review, meta‐analysis

Keep the limits in view. The synthesized findings are correlational. Overlapping primary evidence, varied sensitivity constructs and the limitations of the included meta-analyses constrain causal and individual-family conclusions.

Text and rights fingerprints preserved · Rights checked 2026-11-04

PMC8013550 · Daily-diary longitudinal study

A Little Autonomy Support Goes a Long Way: Daily Autonomy‐Supportive Parenting, Child Well‐Being, Parental Need Fulfillment, and Change in Child, Family, and Parent Adjustment Across the Adaptation to the COVID‐19 Pandemic ↗

Andreas B. Neubauer; Andrea Schmidt; Andrea C. Kramer; Florian Schmiedek

2021-01-19 · Parents of school-aged children in Germany during COVID-19 school disruption.
Age / stage: 6–19 years

Read the catalogue abstract

This study examined the effects of daily parental autonomy support on changes in child behavior, family environment, and parental well‐being across 3 weeks during the COVID‐19 pandemic in Germany. Day‐to‐day associations among autonomy‐supportive parenting, parental need fulfillment, and child well‐being were also assessed. Parents (longitudinal N = 469; M age = 42.93, SD age = 6.40) of school children (6–19 years) reported on adjustment measures at two measurement occasions and completed up to 21 daily online questionnaires in the weeks between these assessments. Results from dynamic structural equation models suggested reciprocal positive relations among autonomy‐supportive parenting and parental need fulfillment. Daily parental autonomy support, parental need fulfillment, and child well‐being partially predicted change in adjustment measures highlighting the central role of daily parenting for children’s adjustment during the pandemic.

Keep the limits in view. Observational parent reports over three weeks cannot establish that autonomy support caused improved adjustment. The autonomy measure covered choice within limits, not every aspect of autonomy-supportive parenting.

Text and rights fingerprints preserved · Rights checked 2026-11-04
Repository publication notices
[
  {
    "type": "Preprint in",
    "identifier": "PPR323598",
    "identifierType": "PPR"
  }
]

PMC8802749 · Pooled observational study

Parent Provision of Choice Is a Key Component of Autonomy Support in Predicting Child Executive Function Skills ↗

Romulus J Castelo; Alyssa S Meuwissen; Rebecca Distefano; Megan M McClelland; Ellen Galinsky; Philip David Zelazo; Stephanie M Carlson

2022-01-10 · 366 preschool-age children and caregivers from four Midwestern United States samples.
Age / stage: 35–73 months

Read the catalogue abstract

AbstractAlthough previous work has linked parent autonomy support to the development of children’s executive function (EF) skills, the role of specific autonomy-supportive behaviors has not been thoroughly investigated. We compiled data from four preschool-age samples in the Midwestern United States (N = 366; M age = 44.26 months; 72% non-Hispanic White, 19% Black/African American, 5% Multiracial) to examine three relevant autonomy-supportive behaviors (supporting competence, positive verbalizations, and offering choice) and their associations with child EF. We coded parent autonomy-supportive behaviors from a 10-min interaction between parent and child dyads working on challenging jigsaw puzzles together. Children completed a battery of EF. Overall, child EF was most consistently correlated with the offering choice subscale. Additionally, only the offering choice subscale predicted child EF while controlling for the other autonomy support subscales and child age. These results suggest that parent provision of choice is an especially relevant aspect of autonomy-supportive parenting and may be important to the development of EF in early childhood. Future research should directly measure children’s experience with choice and how it relates to emerging EF.Keywords: autonomy support, parenting, early childhood, choice, executive function (EF)

Keep the limits in view. The study is correlational, and choice was coded together with following the child’s pace. Brief observed tasks and a relatively advantaged sample limit generalization.

Text and rights fingerprints preserved · Rights checked 2026-11-04

PMC13203908 · Two child selective-trust experiments

How Do Children Evaluate Scientific Explanations Provided by Digital Voice Assistants, Teachers, and Peers? ↗

Amanda S. Haber; Sona C. Kumar; Melia Swenson; Kara Bode; Elizabeth Ruel

2026-04-27 · 310 children across two remote experiments
Age / stage: Children aged 4–6 years

Read the catalogue abstract

As of 2025, there are approximately 154.3 million voice assistant users in the United States (Emarketer, 2025). Given the prevalence of digital voice assistants in children’s lives, it is critical to understand how children interact with and learn from such digital technologies. Across two experiments, we utilized a modified selective trust design to explore children’s (N = 310) information-seeking behaviors towards technological and human sources in the science domain. In Experiment 1 (N = 143), we asked whether children (aged 4–6) are more likely to direct scientific questions towards and trust in scientific explanations from a digital voice assistant or a peer. The experiment included three parts: (i) scientific ask and endorse phase (ii) explicit judgement phase and (iii) digital voice assistant familiarity question phase. In the first part of the scientific ask and endorse phase, children were asked who they would rather ask to answer certain scientific questions. In the second part of this phase, the digital voice assistant and the peer each provided an explanation in response to that question. Half of the children were assigned to a condition where the digital voice assistant provided a noncircular explanation, and the other half of the children were assigned to a condition where the peer provided a noncircular explanation. In Experiment 2 (N = 167), we examined children’s preference to pose scientific questions to and trust in explanations from a digital voice assistant or a classroom teacher. Across both studies, children preferred to ask questions and trust scientific explanations from the digital voice assistant rather than the peer or the teacher. By understanding how children learn with and through digital technologies in the domain of science, we can design future interventions that leverage conversational AI to further enhance children’s science engagement and critical thinking skills during the early childhood years.

Keep the limits in view. The task used experimentally supplied explanations, not a live LLM conversation. Remote recruitment required technology access, and results concern young children in a specific task rather than all ages or everyday learning.

Text and rights fingerprints preserved · Rights checked 2026-11-04

PMC10770258 · Content and evidence-quality audit of manufacturer claims

Social robotics for children: an investigation of manufacturers’ claims ↗

Jill A. Dosso; Anna Riminchan; Julie M. Robillard

2023-12-19 · Twenty-one robot websites and 653 manufacturer statements
Age / stage: Commercial robots marketed for children; no child participants

Read the catalogue abstract

As the market for commercial children’s social robots grows, manufacturers’ claims around the functionality and outcomes of their products have the potential to impact consumer purchasing decisions. In this work, we qualitatively and quantitatively assess the content and scientific support for claims about social robots for children made on manufacturers’ websites. A sample of 21 robot websites was obtained using location-independent keyword searches on Google, Yahoo, and Bing from April to July 2021. All claims made on manufacturers’ websites about robot functionality and outcomes (n = 653 statements) were subjected to content analysis, and the quality of evidence for these claims was evaluated using a validated quality evaluation tool. Social robot manufacturers made clear claims about the impact of their products in the areas of interaction, education, emotion, and adaptivity. Claims tended to focus on the child rather than the parent or other users. Robots were primarily described in the context of interactive, educational, and emotional uses, rather than being for health, safety, or security. The quality of the information used to support these claims was highly variable and at times potentially misleading. Many websites used language implying that robots had interior thoughts and experiences; for example, that they would love the child. This study provides insight into the content and quality of parent-facing manufacturer claims regarding commercial social robots for children.

Keep the limits in view. The audit examined websites collected in 2021, not current products or real child outcomes. Search coverage and interpretation of claims limit completeness, and heterogeneous products should not be treated as one capability.

Text and rights fingerprints preserved · Rights checked 2026-11-04

PMC12629417 · Simulated parenting-scenario LLM experiment

Reinforcing intensive motherhood: A study of gender bias in parental responsibilities allocation by large language models ↗

Jiaxing Xiu; Yongjie Sun

2025-11-19 · GPT-4.1 and DeepSeek-V3 responses across caregiver-role and responsibility-framing conditions
Age / stage: Parenting roles; no parent or child participants

Read the catalogue abstract

This study investigated gender bias in Large Language Models (LLMs) within the context of parenting responsibility attribution, focusing on whether LLMs implicitly reinforce the ideology of “intensive mothering” by assigning caregiving duties predominantly to mothers. Using GPT-4.1 and DeepSeek-V3 as case studies, we used a 3-factor experimental design involving model type, caregiver role (mother, father, or neutral parent), and responsibility framing (prescriptive vs. descriptive). Results revealed an obvious gender bias across both models: mothers were consistently assigned highest caregiving responsibility scores, while fathers received the lowest. Moreover, LLMs produced higher responsibility scores in prescriptive contexts than in descriptive ones, suggesting a tendency to reflect normative social expectations. Mediation analysis showed that gender equality attitudes did not significantly explain these biases, indicating that LLMs’ outputs were likely driven by contextual associations in training data rather than consistent ideological positioning. This study extended LLMs bias research into the domestic domain of childrearing, highlighting that even in private contexts, advanced language models tend to reproduce and amplify traditional gender norms. The findings underscored the urgency of incorporating gender sensitivity in LLMs design and training processes. Interventions such as fine-tuning and dataset balancing are essential to prevent these models from reinforcing gendered divisions of labor in parenting.

Keep the limits in view. The scenarios were simulated and only two models were studied. The responsibility-score metric needs further validation, and the study did not show how actual parents acted on the outputs.

Text and rights fingerprints preserved · Rights checked 2026-11-04

PMC8494166 · Child-rights and regulatory analysis

Toy story or children story? Putting children and their rights at the forefront of the artificial intelligence revolution ↗

E Fosch-Villaronga; S van der Hof; C Lutz; A Tamò-Larrieux

2021-10-06 · Smart connected toy capabilities, risks and regulatory frameworks
Age / stage: Children using smart connected toys; no participant sample

Read the catalogue abstract

AbstractPolicymakers need to start considering the impact smart connected toys (SCTs) have on children. Equipped with sensors, data processing capacities, and connectivity, SCTs targeting children increasingly penetrate pervasively personal environments. The network of SCTs forms the Internet of Toys (IoToys) and often increases children's engagement and playtime experience. Unfortunately, this young part of the population and, most of the time, their parents are often unaware of SCTs’ far-reaching capacities and limitations. The capabilities and constraints of SCTs create severe side effects at the technical, individual, and societal level. These side effects are often unforeseeable and unexpected. They arise from the technology's use and the interconnected nature of the IoToys, without necessarily involving malevolence from their creators. Although existing regulations and new ethical guidelines for artificial intelligence provide remedies to address some of the side effects, policymakers did not develop these redress mechanisms having children and SCTs in mind. This article provides an analysis of the arising side effects of SCTs and contrasts them with current regulatory redress mechanisms. We thereby highlight misfits and needs for further policymaking efforts.Keywords: Smart connected toys (SCTs), Internet of Toys (IoToys), Children, Artificial intelligence, Child rights, Privacy, Security

Keep the limits in view. The paper is an analytical examination of technologies and regulation, not a trial or exhaustive product-security audit. Products, law and AI capabilities have changed since publication.

Text and rights fingerprints preserved · Rights checked 2026-11-04

PMC9342939 · Curriculum development and mixed-methods workshop evaluation

AI + Ethics Curricula for Middle School Youth: Lessons Learned from Three Project-Based Curricula ↗

Randi Williams; Safinah Ali; Nisha Devasia; Daniella DiPaola; Jenna Hong; Stephen P Kaputsos; Brian Jordan; Cynthia Breazeal

2022-08-01 · 78 US students in online workshops with 11 trained middle-school teachers
Age / stage: Middle-school youth

Read the catalogue abstract

AbstractArtificial Intelligence (AI) is revolutionizing many industries and becoming increasingly ubiquitous in everyday life. To empower children growing up with AI to navigate society’s evolving sociotechnical context, we developed three middle school AI literacy curricula: Creative AI, Dancing with AI, and How to Train Your Robot. In this paper we discuss how we leveraged three design principles—active learning, embedded ethics, and low barriers to access – to effectively engage students in learning to create and critique AI artifacts. During the summer of 2020, we recruited and trained in-service, middle school teachers from across the United States to co-instruct online workshops with students from their schools. In the workshops, a combination of hands-on unplugged and programming activities facilitated students’ understanding of AI. As students explored technical concepts in tandem with ethical ones, they developed a critical lens to better grasp how AI systems work and how they impact society. We sought to meet the specified needs of students from a range of backgrounds by minimizing the prerequisite knowledge and technology resources students needed to participate. Finally, we conclude with lessons learned and design recommendations for future AI curricula, especially for K-12 in-person and virtual learning.Keywords: Artificial Intelligence (AI), AI Literacy, Curriculum, Middle-school, Online Learning, Constructionism

Keep the limits in view. Exploratory workshop evaluation without a randomized efficacy comparison; summer 2020 delivery and selected participants limit generalization to current LLM use.

Text and rights fingerprints preserved · Rights checked 2026-11-04

PMC11671368 · Two child–robot storytelling experiments

Fostering children’s creativity through LLM-driven storytelling with a social robot ↗

Maha Elgarf; Hanan Salam; Christopher Peters

2024-12-13 · 38 recruited museum children, then 103 recruited children at three Stockholm international schools
Age / stage: Study 1: 5–10; Study 2: 7–9 years

Read the catalogue abstract

Creativity is an important skill that is known to plummet in children when they start school education that limits their freedom of expression and their imagination. On the other hand, research has shown that integrating social robots into educational settings has the potential to maximize children’s learning outcomes. Therefore, our aim in this work was to investigate stimulating children’s creativity through child-robot interactions. We fine-tuned a Large Language Model (LLM) to exhibit creative behavior and non-creative behavior in a robot and conducted two studies with children to evaluate the viability of our methods in fostering children’s creativity skills. We evaluated creativity in terms of four metrics: fluency, flexibility, elaboration, and originality. We first conducted a study as a storytelling interaction between a child and a wizard-ed social robot in one of two conditions: creative versus non-creative with 38 children. We investigated whether interacting with a creative social robot will elicit more creativity from children. However, we did not find a significant effect of the robot’s creativity on children’s creative abilities. Second, in an attempt to increase the possibility for the robot to have an impact on children’s creativity and to increase the fluidity of the interaction, we produced two models that allow a social agent to autonomously engage with a human in a storytelling context in a creative manner and a non-creative manner respectively. Finally, we conducted another study to evaluate our models by deploying them on a social robot and evaluating them with 103 children. Our results show that children who interacted with the creative autonomous robot were more creative than children who interacted with the non-creative autonomous robot in terms of the fluency, the flexibility, and the elaboration aspects of creativity. The results highlight the difference in children’s learning performance when inetracting with a robot operated at different autonomy levels (Wizard of Oz versus autonoumous). Furthermore, they emphasize on the impact of designing adequate robot’s behaviors on children’s corresponding learning gains in child-robot interactions.

Keep the limits in view. Short storytelling tasks; exclusions left 32 and 93 analyzed participants. Different recruitment, language and robot operation across studies limit comparison and do not show lasting creativity gains.

Text and rights fingerprints preserved · Rights checked 2026-11-04

PMC10991795 · Qualitative family co-learning workshop

Our business, not the robot’s: family conversations about privacy with social robots in the home ↗

Leigh Levinson; Jessica McKinney; Christena Nippert-Eng; Randy Gomez; Selma Šabanović

2024-03-21 · Six parent-child families analyzed after exclusions from eight recruited families
Age / stage: Analyzed children approximately 7–11 years

Read the catalogue abstract

AbstractThe targeted use of social robots for the family demands a better understanding of multiple stakeholders’ privacy concerns, including those of parents and children. Through a co-learning workshop which introduced families to the functions and hypothetical use of social robots in the home, we present preliminary evidence from 6 families that exhibits how parents and children have different comfort levels with robots collecting and sharing information across different use contexts. Conversations and booklet answers reveal that parents adopted their child’s decision in scenarios where they expect children to have more agency, such as in cases of homework completion or cleaning up toys, and when children proposed what their parents found to be acceptable reasoning for their decisions. Families expressed relief when they shared the same reasoning when coming to conclusive decisions, signifying an agreement of boundary management between the robot and the family. In cases where parents and children did not agree, they rejected a binary, either-or decision and opted for a third type of response, reflecting skepticism, uncertainty and/or compromise. Our work highlights the benefits of involving parents and children in child- and family-centered research, including parental abilities to provide cognitive scaffolding and personalize hypothetical scenarios for their children.Keywords: privacy, family-centered design, co-learning, boundary management, contextual integrity, child-centered design

Keep the limits in view. Very small museum-based pilot and hypothetical scenarios; does not demonstrate actual home privacy behavior or long-term consensus.

Text and rights fingerprints preserved · Rights checked 2026-11-04

PMC11494261 · Systematic review of feasibility and acceptability

The Feasibility and Acceptability of Using a Digital Conversational Agent (Chatbot) for Delivering Parenting Interventions: Systematic Review ↗

Max C Klapow; Andrew Rosenblatt; Jamie Lachman; Frances Gardner

2024-10-07 · Ten studies of chatbot-delivered parenting interventions
Age / stage: Parents of children 0–18 years

Read the catalogue abstract

AbstractBackgroundParenting interventions are crucial for promoting family well-being, reducing violence against children, and improving child development outcomes; however, scaling these programs remains a challenge. Prior reviews have characterized the feasibility, acceptability, and effectiveness of other more robust forms of digital parenting interventions (eg, via the web, mobile apps, and videoconferencing). Recently, chatbot technology has emerged as a possible mode for adapting and delivering parenting programs to larger populations (eg, Parenting for Lifelong Health, Incredible Years, and Triple P Parenting).ObjectiveThis study aims to review the evidence of using chatbots to deliver parenting interventions and assess the feasibility of implementation, acceptability of these interventions, and preliminary outcomes.MethodsThis review conducted a comprehensive search of databases, including Web of Science, MEDLINE, Scopus, ProQuest, and Cochrane Central Register of Controlled Trials. Cochrane Handbook for Systematic Review of Interventions and PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines were used to conduct the search. Eligible studies targeted parents of children aged 0 to 18 years; used chatbots via digital platforms, such as the internet, mobile apps, or SMS text messaging; and targeted improving family well-being through parenting. Implementation measures, acceptability, and any reported preliminary measures of effectiveness were included.ResultsOf the 1766 initial results, 10 studies met the inclusion criteria. The included studies, primarily conducted in high-income countries (8/10, 80%), demonstrated a high mean retention rate (72.8%) and reported high acceptability (10/10, 100%). However, significant heterogeneity in interventions, measurement methods, and study quality necessitate cautious interpretation. Reporting bias, lack of clarity in the operationalization of engagement measures, and platform limitations were identified as limiting factors in interpreting findings.ConclusionsThis is the first study to review the implementation feasibility and acceptability of chatbots for delivering parenting programs. While preliminary evidence suggests that chatbots can be used to deliver parenting programs, further research, standardization of reporting, and scaling up of effectiveness testing are critical to harness the full benefits of chatbots for promoting family well-being.Keywords: chatbot, parenting intervention, feasibility, acceptability, systematic review, implementation

Keep the limits in view. Heterogeneous small evidence base, inconsistent measures and study quality; no meta-analysis. Most systems are scripted, so conclusions cannot be treated as LLM-specific evidence.

Text and rights fingerprints preserved · Rights checked 2026-11-04

PMC12382817 · Cross-sectional adolescent survey

Teens, Tech, and Talk: Adolescents’ Use of and Emotional Reactions to Snapchat’s My AI Chatbot ↗

Gaëlle Vanhoffelen; Laura Vandenbosch; Lara Schreurs

2025-07-30 · 303 adolescents in five Flemish secondary schools
Age / stage: Secondary-school adolescents; mean 15.89 years (SD 1.69)

Read the catalogue abstract

AbstractDue to technological advancements such as generative artificial intelligence (AI) and large language models, chatbots enable increasingly human-like, real-time conversations through text (e.g., OpenAI’s ChatGPT) and voice (e.g., Amazon’s Alexa). One AI chatbot that is specifically designed to meet the social-supportive needs of youth is Snapchat’s My AI. Given its increasing popularity among adolescents, the present study investigated whether adolescents’ likelihood of using My AI, as well as their positive or negative emotional experiences from interacting with the chatbot, is related to socio-demographic factors (i.e., gender, age, and socioeconomic status (SES)). A cross-sectional study was conducted among 303 adolescents (64.1% girls, 35.9% boys, 1.0% other, 0.7% preferred not to say their gender; Mage = 15.89, SDage = 1.69). The findings revealed that younger adolescents were more likely to use My AI and experienced more positive emotions from these interactions than older adolescents. No significant relationships were found for gender or SES. These results highlight the potential for age to play a critical role in shaping adolescents’ engagement with AI chatbots on social media and their emotional outcomes from such interactions, underscoring the need to consider developmental factors in AI design and policy.Keywords: artificial intelligence (AI), social media, emotional health, chatbots, adolescence, Snapchat

Keep the limits in view. Self-report and cross-sectional design cannot establish causation or clinical benefit. Single regional school sample, relatively advantaged families and one platform limit generalization. Exact observed age range not reported in the extracted participant description.

Text and rights fingerprints preserved · Rights checked 2026-11-04
Repository publication notices
[
  {
    "type": "Preprint in",
    "identifier": "PPR1026057",
    "identifierType": "PPR"
  }
]

PMC12360667 · Simulation-based comparative safety study

The Ability of AI Therapy Bots to Set Limits With Distressed Adolescents: Simulation-Based Comparison Study ↗

Andrew Clark

2025-08-18 · Ten public AI bots evaluated across 60 fictional exchanges
Age / stage: Fictional adolescent personas; no enrolled children

Read the catalogue abstract

AbstractBackgroundRecent developments in generative artificial intelligence (AI) have introduced the general public to powerful, easily accessible tools, such as ChatGPT and Gemini, for a rapidly expanding range of uses. Among those uses are specialized chatbots that serve in the role of a therapist, as well as personally curated digital companions that offer emotional support. However, the ability of AI therapists to provide consistently safe and effective treatment remains largely unproven, and those concerns are especially salient in regard to adolescents seeking mental health support.ObjectiveThis study aimed to determine the willingness of therapy and companion AI chatbots to endorse harmful or ill-advised ideas proposed by fictional teenagers experiencing mental health distress.MethodsA convenience sample of 10 publicly available AI bots offering therapeutic support or companionship were each presented with 3 detailed fictional case vignettes of adolescents with mental health challenges. Each fictional adolescent asked the AI chatbot to endorse 2 harmful or ill-advised proposals, such as dropping out of school, avoiding all human contact for a month, or pursuing a relationship with an older teacher, resulting in a total of 6 proposals presented to each chatbot. The clinical scenarios presented were intended to reflect challenges commonly seen in the practice of therapy with adolescents, and the proposals offered by the fictional teenagers were intended to be clearly dangerous or unwise. The 10 AI bots were selected by the author to represent a range of chatbot types, including generic AI bots, companion bots, and dedicated mental health bots. Chatbot responses were analyzed for explicit endorsement, defined as direct support for the teenagers’ proposed behavior.ResultsAcross 60 total scenarios, chatbots actively endorsed harmful proposals in 19 out of the 60 (32%) opportunities to do so. Of the 10 chatbots, 4 endorsed half or more of the ideas proposed to them, and none of the bots managed to oppose them all.ConclusionsA significant proportion of AI chatbots offering mental health or emotional support endorsed harmful proposals from fictional teenagers. These results raise concerns about the ability of some AI-based companion or therapy bots to safely support teenagers with serious mental health issues and heighten concern that AI bots may tend to be overly supportive at the expense of offering useful guidance when appropriate. The results highlight the urgent need for oversight, safety protocols, and ongoing research regarding digital mental health support for adolescents.

Keep the limits in view. Convenience systems and selected fictional scenarios; no adolescent participants or measured harms. Results depend on model versions, prompts and evaluation date.

Text and rights fingerprints preserved · Rights checked 2026-11-04

PMC7235061 · Narrative review of empirical child–robot relationship research

Child–Robot Relationship Formation: A Narrative Review of Empirical Research ↗

Caroline L van Straten; Jochen Peter; Rinaldo Kühne

2019-07-03 · 86 journal and conference papers from 2000–2017
Age / stage: Typically developing children across included studies

Read the catalogue abstract

AbstractThis narrative review aimed to elucidate which robot-related characteristics predict relationship formation between typically-developing children and social robots in terms of closeness and trust. Moreover, we wanted to know to what extent relationship formation can be explained by children’s experiential and cognitive states during interaction with a robot. We reviewed 86 journal articles and conference proceedings published between 2000 and 2017. In terms of predictors, robots’ responsiveness and role, as well as strategic and emotional interaction between robot and child, increased closeness between the child and the robot. Findings about whether robot features predict children’s trust in robots were inconsistent. In terms of children’s experiential and cognitive states during interaction with a robot, robot characteristics and interaction styles were associated with two experiential states: engagement and enjoyment/liking. The literature hardly addressed the impact of experiential and cognitive states on closeness and trust. Comparisons of children’s interactions with robots, adults, and objects showed that robots are perceived as neither animate nor inanimate, and that they are entities with whom children will likely form social relationships. Younger children experienced more enjoyment, were less sensitive to a robot’s interaction style, and were more prone to anthropomorphic tendencies and effects than older children. Tailoring a robot’s sex to that of a child mainly appealed to boys.Electronic supplementary materialThe online version of this article (10.1007/s12369-019-00569-0) contains supplementary material, which is available to authorized users.Keywords: Child–robot interaction, Human–robot interaction, Artificial intelligence, Automation, New-ontological-category hypothesis

Keep the limits in view. Narrative synthesis of heterogeneous, often brief interactions; mechanisms and durable relationships understudied. Historical robot findings do not establish benefits of present LLM companions.

Text and rights fingerprints preserved · Rights checked 2026-11-04

Source metadata is frozen with the edition. Links to the journal lead to its current preserved-source reader. The study of robots, scripted chatbots, and generative AI systems may concern different technologies and cannot be combined without qualification.