Chatbot-based health interventions in low- and middle-income countries: effective access, early attrition, and design strategies from a mixed-methods study

Abstract Equity in digital health is commonly assessed through enrolment and coverage. For interventions requiring sustained participation, enrolment can substantially overstate access. We report a convergent mixed-methods optimisation study of a chatbot-based behaviour change intervention for caregivers (ParentText) delivered via WhatsApp and Facebook Messenger across three low- and middle-income countries: Jamaica, Malaysia, and the Philippines ( N  = 1293). Using engagement logs, participant and implementer interviews, and a structured consensus process, we examined attrition patterns and explored optimisations. Median engagement was two days and one module; 93.8% of participants completed less than 25% of available content over the five-week intervention. Fewer than one in five participants remained active at Day 7 in two of three sites. Structural barriers included platform policies restricting follow-up after non-response, high message burden, and onboarding design. A consensus panel identified 23 barriers and generated 29 optimisations (69% rated highly feasible within LMIC contexts), currently being tested in a subsequent trial. We introduce effective access as a complementary construct for digital health interventions requiring repeated participation and propose three design principles: 1) test whether users can resume participation under platform constraints; 2) reduce message burden without removing content; and 3) treat re-engagement as a core feature.