Anthropomorphism in Children's Interactions with LLM Chatbots: A Systematic Review of Drivers and Outcomes

A systematic review of 35 empirical studies (2022–2025) examining how children anthropomorphize LLM-based chatbots. Uses Epley's SEEK theory (Elicited Agent Knowledge, Effectance Motivation, Sociality Motivation) as its theoretical anchor and follows PRISMA guidelines.

Four drivers of anthropomorphism emerge. Human-Like Persona Construction (n=19) — roleplay characters with backstories, voice cues, and communicative competence that activate children's self-knowledge about humans. Supportive Companionship (n=16) — non-judgmental peer positioning and emotional responsiveness that satisfy sociality needs. Adaptive Scaffolding (n=14) — cognitive and affective support that reduces uncertainty, triggering effectance motivation. Non-Human Embodiment Design (n=10) — a novel fourth driver extending SEEK: fantasy archetypes (whimsical animals, nature embodiments) where cross-modal responsiveness overrides surface appearance, showing LLM conversational coherence alone can trigger anthropomorphism without human-like visuals.

Five outcome themes, all carrying both benefits and risks. Social Ties Construction — children form bonds from strong (parental figures, romantic partners at the formal operational stage) to weak (teacher, entertainer). Social Boundary Exploration — testing what the "social other" can and can't do. Human Narrative Attribution — explaining chatbot failures with human-like intentions rather than technical causes. Paradoxical Social-Moral Responses — treating the chatbot as both socially real and a mere machine. Dual Consciousness Formation — holding both "it's not human" and "I'll respond as if it were" frameworks simultaneously.

Developmental stage modulates outcomes: the romantic-partner sub-theme appeared exclusively in adolescents at Piaget's formal operational stage. The evidence base skews short-term, US-centric (18 of 35 studies), and GPT-dominated (28 of 35), limiting generalizability to sustained developmental trajectories. The authors call for longitudinal work, stage-differentiated design, and governance frameworks that go beyond safety thresholds.