Kids ended up trusting AI more than an actual teacher. That's the real finding here.

A new study, published in Technological Forecasting and Social Change in May 2026, recruited 86 upper-grade elementary school students from eastern China (84 valid after excluding two who failed attention checks, mean age 10.24) to test trust in AI versus teacher recommendations.

Most existing research on trust in AI recommendations has focused on adults -- there have been relatively few experimental studies directly comparing how children specifically evaluate and respond to advice from an AI versus a human. That's a meaningful gap, given how much AI-powered tutoring, educational chatbots, and recommendation tools are now aimed squarely at kids.

Kids were randomly assigned to receive clothing-coordination advice from either an AI or a teacher. The task was intentionally low-stakes: "compared with high-risk tasks such as healthcare and finance, clothing is a more everyday and low-risk selection context" -- chosen specifically so kids would show their honest trust responses, not their anxiety about a high-stakes decision.

"Study 1 showed no significant difference in initial trust between AI and teacher recommenders; however, post-interaction trust was significantly higher for AI." Kids started out neutral about which source to trust more -- but after actually interacting with each one, trust in the AI pulled ahead of trust in the teacher. Trust itself was also significantly associated with whether kids actually adopted the advice they were given, meaning this wasn't just a feeling that stayed abstract -- it changed what kids actually did.

A second experiment found that "anthropomorphic AI was more likely than mechanical AI to induce changes in children's initial choices" -- a more human-like-looking AI moved kids' minds more than a robotic-looking one. That lines up with a related pattern researchers in this space have flagged elsewhere: kids tend to assume an AI system doesn't make mistakes, even when it's demonstrably no more accurate than a human giving the same advice, and tend to prefer AI guidance in ambiguous situations as a result.

The researchers were deliberate about keeping the task low-stakes, and that matters for how far this finding should be stretched. A clothing recommendation carries essentially no real consequence if it's wrong. Whether the same trust gap shows up -- and whether it's a good or bad thing -- in higher-stakes contexts like homework help, health questions, or emotional support is a separate question this study doesn't answer.

The researchers frame this as evidence that trust in AI isn't fixed by identity alone -- it's shaped dynamically through interaction and design cues, which has real implications for how AI systems built for children should be designed and evaluated. Because kids are especially susceptible to both privacy risks and exposure to inappropriate content through AI tools, and because human-like design choices appear to make kids more persuadable, researchers argue anthropomorphic design in children's AI products deserves real scrutiny -- not just as a usability choice, but as a factor that can measurably shift what a child believes and decides.

Source: https://www.sciencedirect.com/science/article/abs/pii/S0040162526001848 (Technological Forecasting and Social Change, May 2026)