Text this: Controlled Benchmark Evaluation of a Geographically Grounded Digital Health Chatbot Against a General-Purpose Model.