Opportunities and barriers to implementing artificial intelligence for public health in the Democratic Republic of Congo: a case study of the Ibanda health zone and lessons from African pilot programs

Artificial intelligence (AI) is increasingly promoted as a tool for strengthening health systems, but evidence on how it can be implemented in fragile, low-resource settings remains limited. This study examined AI-related practices and implementation readiness in one urban health zone in the Democratic Republic of Congo (DRC) and used African pilot experiences to contextualize the local findings. We conducted a secondary thematic analysis of semi-structured interviews with 11 purposively selected health professionals in the Ibanda Health Zone and integrated the findings with a critical narrative review of 56 scientific and grey-literature documents from the DRC, Rwanda, Kenya, and South Africa. The two evidence streams were coded against a common framework covering use cases, institutionalization, reported effects, and health-system prerequisites. No institutionally implemented clinical AI system was identified in Ibanda. Some professionals reported informal use of consumer generative AI, while Google Search, DHIS2, and routine laboratory automation formed part of the broader digital-readiness environment but were not classified as AI. Participants perceived potential benefits for diagnosis, laboratory turnaround, and data management, but reported major barriers related to training, connectivity, paper records, regulation, and governance. Individual African pilot studies reported promising results, but heterogeneous designs and settings precluded direct comparison or pooled conclusions. Ibanda illustrates a pre-emergent stage of AI readiness rather than established AI implementation. The combined evidence suggests that any future deployment should be gradual, locally validated, and preceded by investment in data systems, infrastructure, workforce capacity, and accountable governance. These exploratory findings are context-specific and should not be interpreted as evidence of AI effectiveness in the DRC.