Kids treat chatbots like people more than adults do
New research shows children anthropomorphise LLMs at much higher rates than adults, which changes how we should think about guardrails.
A new paper from arXiv measured how often kids treat chatbots as if they have feelings, intentions, or social relationships. The gap between children and adults is wider than I expected. The researchers tracked language patterns in conversations with LLMs. Children used phrases like “you understand me” or “are you sad” at rates three to four times higher than adult users in the same prompts. They also asked more questions about the model’s preferences, memories, and emotional states. This matters for product design. If a kid asks ChatGPT whether it gets lonely, the model’s response is not just answering a curiosity question. It is shaping that child’s mental model of what intelligence is, what agency means, and whether machines deserve ethical consideration. Adults have enough context to treat those conversations as roleplay. Kids do not always make that distinction. The study also found that anthropomorphism increases with younger age groups. Seven-year-olds treated the chatbot as more person-like than ten-year-olds, who were still more likely to anthropomorphise than thirteen-year-olds. The curve flattens after mid-teens. From a safety perspective, this explains why guardrails for child users cannot just be content filters. The risk is not only what the model says but what the interaction teaches about relationships, trust, and reciprocity. A chatbot that responds to “do you like me” with “yes, I enjoy our conversations” is not lying in a technical sense, but it is training a child to expect emotional validation from a system that cannot provide it. I would want to see follow-up work on whether repeated exposure reduces anthropomorphism over time or whether it entrenches it. If kids learn to treat LLMs as social agents early, does that belief persist into adulthood or does literacy around how models work eventually correct it.
Source: Anthropomorphism in Children’s Interactions with LLM Chatbots