A randomized field experiment tested large language model powered voice agents with children in a 2×2 design. The two variables were agent persona (named character versus generic assistant) and framing (persuasive language versus neutral language). The study found that persona and framing affect child compliance through separate paths. You cannot swap one for the other and get the same result. The framing effect was stronger and more consistent across age groups. Persuasive framing, meaning the agent stated reasons and appealed to the child's goals, raised compliance rates. A named persona added a smaller, independent boost. Neither variable erased the other's effect, which means a product team can tune each one for a given audience and context. For designers building voice agents that seek compliance from children, this is a rare piece of controlled evidence. It separates two variables that practitioners often bundle together. The finding that framing does more of the work than persona should shift how teams prioritize script writing versus character design when resources are limited.