An AI that calls itself Santa Claus versus one that calls itself his helper — which one is better at persuading children? Title: Persona and Persuasive Framing in AI Voice Agents: A 2×2 Field Experiment with Children URL: ❓ What did they actually test? The researchers embedded a research voice agent into a real German Santa hotline. Calls were randomly routed into one of four combinations — "Santa" or "Helper" persona, crossed with "persuasive" or "neutral" conversation — and they measured whether children spontaneously expressed a prosocial wish (one benefiting someone else). ❓ Did the persuasive framing actually work? Yes, strongly. 45.7% of children in the persuasive condition expressed a prosocial wish, versus just 11.6% in the neutral condition — an odds ratio of 6.4, and the effect held even after controlling for age and call duration. ❓ Was Santa more persuasive than the generic Helper? Surprisingly, barely at all. Under the persuasive framing, Santa (46.2%) and Helper (45.0%) were statistically indistinguishable. How the agent talked mattered far more than who it claimed to be. 💡 So did persona matter at all? Yes — just not for persuasion. Calls to the Helper ended within the first minute at a rate over 25 points higher than Santa's, and were more than 4x as likely to end without the child saying a word. Persona decided whether kids stayed on the line at all, not whether they complied once they did. The idea that a child-directed agent's persuasive power lives in its cheaply-rewritten script rather than its visible character feels like it should weigh heavily on how we think about regulating this. #AIEthics# #ChildrenAndAI#