Researchers across domains have examined children’s use of LLM-based chatbots, yet studies remain fragmented on anthropomorphism—the tendency to attribute human traits to non‑human chatbots. This systematic review analyzes 35 empirical studies published between 2022–2025. Findings reveal four key drivers: human‑like persona construction, adaptive scaffolding, supportive companionship, and non‑human embodied design. These factors foster children’s anthropomorphic interactions with the chatbots.

The review also identifies five notable outcomes. Children exhibit paradoxical social and moral responses, develop a dual consciousness about the chatbot’s agency, form varying social ties, explore social boundaries, and attribute human narratives to conversation breakdowns.

Both benefits and risks emerge from these findings. While anthropomorphic cues can enhance engagement and learning, they also risk fostering unrealistic expectations or misattributing agency. The study underscores the importance of aligning chatbot features with developmental stages and ensuring adaptive scaffolding does not reinforce false beliefs. These insights inform ethical guidelines and user interface strategies for child‑friendly chatbots.

Designers should balance anthropomorphic cues with transparency about the chatbot’s non‑human nature to mitigate misunderstandings. Future research should explore longitudinal effects and culturally diverse contexts to refine guidelines for child‑centric LLM chatbot design.