NeFut Logo NeFut
中 Admin Login

[CS.AI] Persona and Persuasive Framing in AI Voice Agents: A $2\times2$ Field Experiment with Children

Published at: 2026-10-01 22:00 Last updated: 2026-10-06 12:11
#AI #Machine Learning #LLM

Conversational agents are increasingly interacting with children, yet most evidence on how design influences children’s susceptibility to persuasion comes from laboratory settings. We embedded a $2\times2$ randomized field experiment in a public German Santa‑Claus telephone hotline, routing incoming calls to one of four LLM voice agents.\

The agents varied along two dimensions: persona (high‑authority “Santa” vs. low‑authority “Helper”) and framing (persuasive nudges toward prosocial wishes vs. neutral prompts). Out of 1,072 logged calls, 89 conversations (median age 6) met the inclusion criteria.\

Persuasive framing dramatically increased the probability of a prosocial wish from $11.6\%$ to $45.7\%$, a result that remained robust after adding controls. Persona authority showed a near‑zero effect: Santa did not outperform the Helper. However, persona shaped engagement; children hung up on the Helper within the first minute at a rate of $65\%$, compared to $39\%$ for Santa.\

When the context already confers legitimacy on the agent (as with a Santa hotline), how the agent speaks influences children’s compliance more than who it claims to be. Designers of child‑focused voice systems should therefore prioritize persuasive language over merely projecting authority.\

Review

Original Source: https://arxiv.org/abs/2609.38782

[h] Back to Home