Study finds AI health messages in Africa no better than traditional campaigns

A new study comparing AI-generated health messages with traditional campaigns in Kenya and Nigeria found that neither approach proved superior for communicating about vaccines and maternal healthcare. The research analyzed 120 health messages and revealed that while AI was more creative in incorporating cultural references, it often produced shallow or inaccurate content, while traditional campaigns remained authoritative but rigid and sometimes reinforced colonial-era communication patterns.

What the study found: Researchers from The Conversation analyzed 80 traditional health messages from ministries and NGOs alongside 40 AI-generated messages, focusing on vaccine hesitancy and maternal healthcare communication.

• AI-generated messages included more cultural references than traditional campaigns, attempting to use local metaphors, farming analogies, and community-centered language.
• However, these cultural references were often shallow and sometimes inaccurate, with AI referencing local customs without truly understanding them.
• AI-generated images frequently produced warped, distorted faces, reflecting persistent problems with AI systems trained on insufficient diverse examples.

Traditional campaign limitations: Human-created health materials had their own significant shortcomings despite being developed by well-resourced organizations with local presence.

• Traditional campaigns often stuck to clinical, western medical language and reinforced external medical expertise.
• They gave limited space to community knowledge and traditional health practices.
• Most striking was that both AI and traditional approaches positioned people as “passive recipients of expert knowledge rather than active participants in their own health decisions.”

Technical performance issues: The WHO’s health-focused AI tool S.A.R.A.H (Smart AI Resource Assistant for Health) demonstrated notable functionality problems during testing.

• The system often produced incomplete responses and sometimes required resets to function properly.
• Its use of a white female avatar raises questions about representation in global health AI design.
• These issues occurred despite the tool being designed specifically for health communication.

Why this matters now: AI adoption in African health systems is accelerating rapidly, with significant implications for public health outcomes.

• In sub-Saharan Africa, 31.7% of AI deployments in health focus on telemedicine, 20% on sexual and reproductive health, and 16.7% on operations.
• The stakes are particularly high for vaccine hesitancy and maternal health, where “trust, cultural sensitivity and community buy-in can literally mean the difference between life and death.”
• Success stories are emerging, including Kenya’s AI Consult platform reducing diagnostic errors and AI tools improving healthcare access in Nigeria.

The path forward: The solution involves developing AI systems with genuine community engagement rather than abandoning the technology entirely.

• Health organizations should build community feedback loops into AI message development and test content with target communities.
• Training AI systems using locally relevant data and knowledge systems is essential for cultural accuracy.
• African-led platforms like the digital healthcare assistant AwaDoc demonstrate how locally developed AI can better understand cultural context while maintaining medical accuracy.

What experts are saying: The research highlights broader patterns in global health communication that extend beyond AI implementation.

• The study noted how international organizations can “inadvertently replicate colonial-era patterns of external ‘experts’ telling local communities what to do.”
• This pattern was reinforced during COVID-19 when “high-income countries blocked efforts to waive intellectual property rules and hoarded vaccine doses,” leaving many low- and middle-income countries struggling for access.
• According to the researchers, “AI’s future in global health communication will be determined not just by how smart these systems become, but by how well they learn to genuinely listen to and learn from the communities they aim to serve.”