Remember life before the internet?
Health advice in many homes lived on a shelf. Families kept books such as Your Health and You, written by two doctors “in collaboration with 38 leading medical specialists.” When a child had a fever, parents opened the book and trusted it, because real experts had put their names to every page.
A well-used copy of Your Health and You, Volume 2, by Harold Shryock and Hubert O. Swartout (The Stanborough Press, reprinted 1980).
Then the World Wide Web arrived
A parent with a feverish child could now type their worry into a search engine and get pages of answers in seconds. But who had written them? In 1997, Impicciatore and colleagues in Milan found 41 web pages advising parents on a child’s fever. Only four closely followed medical guidelines, and the researchers called for urgent checks on health information online. That same year, a JAMA editorial warned Caveant lector et viewor, let the reader and viewer beware. The Medical Journal of Australia put it more bluntly, “anybody with a computer and internet connection can be an author, editor and publisher rolled into one.” Yet checking health websites was left largely to voluntary codes, such as the HONcode launched in 1996.
We have been here before
Nearly thirty years later, we are asking the same questions about AI, with the same excitement, confusion and distrust. Who checks these tools? Who pays for them? Who is accountable when they fail?
This time, the shift is bigger. The book had an author. The web page had an address you could trace, if you knew how. Built into software, AI can read scans, support diagnosis and advise clinicians. And unlike a book, it can be updated after it is in use, changing how it works.
The internet did fix a great deal. It moved journals online and let evidence travel far beyond the institutions that produced it. But access was never equal. In 2000, Africa had roughly one person online for every 250, against a world average of one in 35, and 56% of health institutions in the poorest countries had no subscription to any international journal. Africa got in late, on terms set elsewhere, and mostly as a reader rather than an author.
Building on what the internet started
That is changing. By 2025, more than one in three Africans were online and nearly all internet users in Kenya and Nigeria now use an AI tool each month. The question, therefore, is no longer whether Africa gets in. It is whether Africa helps shape what is inside.
At HealthTech Hub Africa (HTHA), a pan-African innovation platform working with healthtech innovators, funders and governments, we see that work up close. This year, health officials from 21 African countries met twice through our Intergovernmental Working Group (IWG) to share how AI is entering their health systems.
AI is already being tested and used in their health systems, from Ebola chatbots in Uganda, ultrasound and tuberculosis tools in Zambia and early detection of HIV and diabetes in Nigeria. But AI is only as good as the data it learns from. If health data is fragmented or missing, AI will reproduce the old gaps, only faster. As one official put it, AI is a mirror of health system maturity.
That is why Africa cannot arrive late this time. Equity has to be built, through clear governance, local data, local capacity and solutions designed for the systems they will run in. Officials’ clearest message was that AI should not bypass the system. It should strengthen it. Governments are setting rules too. Nigeria has endorsed a national digital health architecture designed so that all AI tools in health “plug into a unified system”. Kenya is developing an AI policy for health. Rwanda’s National AI Policy covers healthcare, and Egypt has adopted a Charter for Responsible AI. Countries move at different speeds, but they face the same questions from healthtech innovators.
The questions innovators ask, and what governments are telling us
Innovators ask us the same things. Who will approve my tool? Which regulations apply? Why did my pilot not scale? Through the Intergovernmental Working Group, we have been putting these questions to governments. Their answers point to five lessons.
Regulatory pathways are still being defined. A single AI tool can fall under health regulation, data protection, professional standards and a national AI mandate at once, leaving it unclear who evaluates, approves, registers or monitors it. Countries are clarifying these pathways. Zambia has mapped how an AI product would move from approval to registration to monitoring, and Kenya is developing an AI for health framework. Innovators need to understand every institution their product falls under, and to avoid over-reliance on a single champion, whose support can end with a change of administration.
AI needs regulation, validation and post-market surveillance. Some governments want AI in health regulated in the same way as medical devices and medicines, with checks that tools are safe, appropriate and trained on relevant data. Lab results are not enough. Governments need evidence that is local, credible and relevant to the real implementation context, and AI that complements human expertise rather than replacing human judgment.
Data governance comes first. Weak data governance, including data access, privacy and interoperability, is the barrier officials name most often, ahead of fragmented systems. Digital public infrastructure such as registries, identifiers and health information exchanges must be strengthened, and solutions will need to connect to these shared national components.
Scale needs a transition pathway. Slow and complex procurement processes keep many AI pilots from scaling. Pilots stall not because they fail, but because they never make it into a government budget. “We proved it works,” one innovator told us, “but that did not solve who pays, who owns, or who maintains it.” In countries such as Kenya, Uganda and Zambia, health budgets are shaped six months to a year before the financial year begins, and budget approval does not always mean funds are released. Engage while budgets are being shaped, and plan who will own, pay for and maintain your solution from the start.
Trust is institutional. Public trust and cultural hesitation remain real barriers, which is why countries such as Uganda, which used AI-supported chatbots in its Ebola response, are adopting AI in phases. The gap also shows up in language. Innovators say “bureaucracy.” Officials say “checks and balances.” Public servants are accountable for what happens in public health systems, and backing a solution that fails carries real professional risk. For them, trust means accountability.
Build with the system, not around it
Together, these lessons point one way. African governments are setting the direction for digital health and AI, and every solution will need to meet national regulatory and data requirements. Those seeking to scale through the public sector must also align with national priorities, connect to national digital infrastructure, and plan from the start for public ownership, financing and recurrent costs.
Each actor has a role. Governments can clarify regulatory pathways, articulate demand earlier and make planning, budgeting and procurement cycles visible. Funders can avoid parallel systems and invest in the transition from pilot to long-term public ownership and financing. Innovators can meet regulatory and data requirements early, generate local evidence and, where they seek public-sector scale, engage in planning and budgeting cycles and become procurement-ready before they pitch. The HTHA Resource Hub brings together country policy analysis, compliance and public sector scaling checklists, and guidance on budget and procurement cycles to support this work.
We once trusted the book on the shelf because someone stood behind it. AI will earn that trust only if someone stands behind it too. This time, Africa must be an author, not just a reader.
HealthTech Hub Africa
This article is part of Digital Health Week 2026, convened by Transform Health.
