Who is actually using AI in Africa? It depends on whom you count

Kenya came first in a global ranking of ChatGPT use last year. In a survey published by DataReportal in July 2025, 42.1% of Kenyan Internet users aged 16 and above said they had used ChatGPT in the previous month—the highest share among the markets surveyed. South Africa came in at 34.3% and Nigeria at 27.9%. 

Microsoft’s ranking put the three countries in the opposite order. In the first quarter of 2026, it estimated that 23.1% of South Africans aged 15 to 64 had used a generative artificial intelligence (AI) tool, followed by 10.1% of Nigerians and 8.7% of Kenyans. The next quarter brought modest gains in all three countries, with South Africa reaching 23.9%, Nigeria 10.4% and Kenya 8.9%. 

The reversal can be confusing—and consequential—because AI adoption figures shape how investors size a market or how governments decide to deliver services. Yet if you look at the numbers closely, the two rankings describe different things. 

One shows how popular ChatGPT is among people who already use the internet. The other estimates how far a range of generative AI tools has reached across the working-age population. Then there is a third question, which neither headline figure answers: how many people have found a use for AI that brings them back? 

The people surveys leave out 

DataReportal’s ChatGPT figures come from GlobalWebIndex (GWI), which surveys internet users. Imagine two countries with 100 adults each. In the first, 40 are online, and half of them use ChatGPT. In the second, 80 are online, and a third use it. 

The first leads among internet users. The second has more ChatGPT users across the adult population. The ranking changes simply because the share of people who are online differs. 

The difference in access between Kenya and South Africa is substantial. In its Digital 2027 country reports, DataReportal puts Internet penetration at 40.6% in Kenya and 79.5% in South Africa. Nigeria sits at 47.5%.

Those figures cover the whole population and come from a different period from the original ChatGPT survey. They show why connectivity matters, but multiplying them by the ChatGPT percentages would mix age groups and timeframes, yielding a misleading national estimate. 

The denominator becomes even more important as the definition of AI widens. In its more recent global AI report, DataReportal says 97.3% of Kenyan internet users aged 16 and above reported using some form of AI in the previous month. In Nigeria, the figure was 96.6%. 

“It’s critical to stress that these figures represent the share of Internet users,” the report says. This time the survey was asking about AI broadly. The figures cannot be read as evidence that ChatGPT use has doubled, or that almost everyone in either country uses AI. 

Who gets online also affects the comparison. DataReportal found that, across the markets surveyed, 93.3% of internet users aged 16 to 24 said they had used AI in the previous month, compared with 52.2% of those aged 65 and above. 

If one country’s online population skews younger than another’s, that alone could lift its adoption rate. The published rankings leave open the question of how much of Kenya’s lead reflects the ages of the people surveyed and how much reflects differences in behaviour among people of the same age. 

What Microsoft can see

Microsoft starts with a wider population—everyone aged 15 to 64, whether they work, study or have internet access. It also counts activity across generative AI products. However, the data it observes directly are “primarily originating from users on Windows desktop (PC/tablet) platforms,” according to its technical paper. 

People can opt out of sending diagnostic data, and Microsoft uses other sources to estimate usage beyond Windows and account for activity on mobile devices. “These assumptions are difficult to validate,” the researchers write of their mobile adjustment. 

That matters when someone’s first encounter with generative AI may be through a chatbot on their phone. A survey can capture that person’s experience if they remember and report it; a model anchored in desktop activity has to estimate how much use it misses. 

Both approaches make assumptions, and the two headline figures offer little basis for saying exactly how much of the gap comes from internet access, the mix of products counted, or the way each study collects its data.

Microsoft’s tables also give reason to be cautious about drawing fine distinctions between countries. Its technical paper notes that it sometimes pools markets with limited data to produce regional averages. Nigeria shares its 10.4% second-quarter estimate with several countries, including Ghana, Benin, and Togo.  

The table alone offers little basis for treating Nigeria’s 1.5 percentage-point lead over Kenya as a precise measure of how differently people in the two countries use AI. South Africa’s much wider lead is the more striking result. 

Relying on chatbots

For a startup selling an AI writing tool to small businesses, familiarity with ChatGPT could make it easier to explain what the product does. Pricing it calls for a different kind of evidence. How often do those businesses need help writing? Do they already get enough from a free chatbot? Would a tool that works in the languages they use with customers be worth paying for? 

A high adoption rate among internet users tells a founder where to look for customers. It takes more to work out what those customers will buy.

For a government, the cost of getting the denominator wrong is different. If a ministry puts an AI assistant on its website to explain how to apply for a permit, a high rate of ChatGPT use among connected Kenyans may suggest some applicants will find it familiar. 

The ministry still needs to know how many applicants can reach the site, whether they can follow the instructions on a phone and how often they manage to complete the application. Even Microsoft’s wider estimate excludes people aged 65 and over, who may also need the service.

A student who asks ChatGPT to summarise one reading and a trader who uses it every week to draft replies to customers both count in a monthly survey. The distinction starts to matter when the trader changes how they handle enquiries because of the tool. Knowing how many people have reached that point would tell us more about what AI is changing in Africa than knowing which country came first.