Study reveals health queries dominate general-purpose LLM use on WhatsApp

A new research project has shed light on how everyday users interact with general-purpose AI chatbots when accessed through familiar messaging apps. Researchers built and deployed WaLLM, an LLM-powered bot on WhatsApp, to observe real-world usage patterns that remain largely unstudied in academic settings.

What people asked for#

The study found that health and well-being topics dominated user queries, with people reaching out for medical advice, wellness information, and personal guidance. This suggests users view AI assistants as accessible sources for sensitive information, raising important questions about how such systems should be designed responsibly.

Engagement features that worked#

Different engagement tools attracted users in distinct ways:

• Proactive messaging (push notifications and active outreach) proved effective at keeping the service visible and correlated with sustained user activity
• Shared content lists helped users discover and share information within their networks, facilitating community interaction around the bot's capabilities

Why this matters#

Most current research focuses on narrow, specialized AI assistants—but real deployment through WhatsApp revealed gaps in our understanding of how people use open-ended, general-purpose LLMs. The messaging platform's affordances and constraints shaped how users interacted with the system, offering practical lessons for anyone building AI services for mobile-first, global audiences.

This work addresses a significant blind spot: while proprietary AI chatbots proliferate on messaging apps worldwide, the academic community has had little visibility into how these systems are actually used, who relies on them, and what design patterns encourage responsible adoption.