AI Literacy in Zimbabwe: Why Using AI Is Not the Same as Thinking With It

What Is AI Literacy and Why Does Zimbabwe Need It?

For many Zimbabweans, AI was never something they went looking for. It arrived inside WhatsApp, in the search bar at the top of their chats. Nearly a third of internet users now use AI, most of them through Meta AI. For the connected population, access is no longer the main problem.

The problem is understanding. AI literacy is the ability to know what these tools are, use them for real work, check what they produce and recognise where they fail. Without it, an assistant becomes a magic box: a source of confident answers that flow into client messages, school notices and business decisions unchecked.

Research on how the brain engages when people write with and without AI points to the habit that matters most: thinking first and using AI second. That distinction, between depending on AI and thinking with it, is what AI literacy in Zimbabwe needs to mean.

How Zimbabweans Use AI: Meta AI, ChatGPT and the Tools Few People Know

The clearest picture comes from the 2025 ICT Access by Households and Use by Individuals Survey Report, produced jointly by ZIMSTAT and POTRAZ. As reported by NewZimbabwe, 31 percent of internet users now use AI applications. Meta AI is the most widely used platform at 53.7 percent of AI users, with ChatGPT a distant second at 29.1 percent. Learning drives adoption: 91.5 percent of AI users rely on it for education and academic work.

Why Meta AI Is the Most Used AI Tool in Zimbabwe

Meta AI's lead is a story about distribution, not preference. WhatsApp is used by 92.9 percent of Zimbabwe's internet users, according to the same survey, and Meta placed its assistant directly inside the app, including in the search bar at the top of your chats. Computers are found in only 11.2 percent of households, so the phone is where AI happens.

Techzim has noted that Meta AI could be used on a WhatsApp bundle, and that ChatGPT was unavailable in Zimbabwe for months after its launch, by which time many people had settled elsewhere.

Choosing the assistant already inside the app you use all day is rational, for the same reasons businesses run their operations through WhatsApp. It does mean that for many people, AI has one face, one voice and one place.

Why Claude, DeepSeek and Mistral Barely Register Outside Tech Circles

Statcounter measures something different: the share of website visits that arrive from each AI chatbot when people click links inside those tools.

Among the tools that send people out to websites, ChatGPT and Gemini dominate, and the assistants that technical users switch between are close to invisible. Mistral does not appear in Statcounter's list at all, and in our experience working with Zimbabwean businesses, Claude, DeepSeek and Mistral are names most people outside technology roles have not heard.

The AI Skill Gap in Zimbabwean Workplaces

The same survey found that only 43.7 percent of Zimbabweans have basic digital skills. AI has arrived faster than the skills people use to judge it, the same gap our analysis of AI in Zimbabwean business identified at institutional level. Inside a single business, it shows up as a habit: treating the assistant as an authority rather than a tool.

What Treating AI as a Magic Box Looks Like Inside a Business

A junior agent at an estate agency asks Meta AI how rental income is taxed and pastes the answer into a reply to a landlord. A school bursar asks for a fees reminder and sends it to parents without noticing the dates the assistant filled in. A sales officer asks which of two suppliers to choose and accepts the answer, without registering that the assistant knew nothing about either supplier's delivery record.

In each case the tool produced fluent, confident text, which is what it was built to do. The failure is the assumption that fluent means correct. AI assistants can be wrong about local tax rules, current prices, regulations and anything that changed recently, and they rarely say so. The business carries the consequence, because the message goes out under its name.

Why AI Answers Inside WhatsApp Feel Trustworthy

An assistant that sits in the same chat list as your family and your suppliers inherits some of their trust. It answers instantly, in a conversational tone and with apparent confidence. Nothing in that experience prompts anyone to ask where the answer came from, and for someone who knows only one assistant, there is nothing to compare it with.

What the MIT "Your Brain on ChatGPT" Study Found

The research most often cited on this question is "Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task", a 2025 MIT Media Lab study led by Nataliya Kosmyna. It is widely cited and widely misreported.

How the Three-Group EEG Essay Study Worked

Fifty-four students and staff from five Boston-area universities, aged 18 to 39, were split into three groups. Over three sessions spread across four months, each wrote a 20-minute essay on an SAT prompt. One group could use only ChatGPT, one could use any website but no AI, with AI answers removed from search results, and one could use no tools at all.

The researchers recorded brain activity with EEG, analysed the essays with language-processing tools, had them scored by human teachers and an AI judge, and interviewed every participant after each session. In a fourth session, 18 participants switched: former ChatGPT users wrote unaided, and the unaided group used ChatGPT, each on a topic they had written about before.

Cognitive Debt: Weaker Engagement, Weaker Recall, Less Ownership

The brain-only group showed the strongest and most widely distributed brain connectivity, search users showed moderate engagement, and ChatGPT users showed the weakest. The more the tool did, the less the brain did. Over four months, the ChatGPT group consistently underperformed at neural, linguistic and behavioural levels.

The first-session interviews show what that looked like.

The last row deserves a business owner's attention. Seventeen of 18 ChatGPT users were satisfied with essays that none of them could quote correctly minutes later. The quoting gap narrowed in later sessions, once participants expected the question, but it did not close. Feeling that a piece of work is finished is not evidence of understanding it.

The researchers call this cognitive debt: the convenience arrives immediately, and the cost appears later. When former ChatGPT users wrote without it in the fourth session, their brain activity showed signs of under-engagement. The use the study examined, writing as part of learning, is the use behind 91.5 percent of AI adoption in Zimbabwe, and those learners are the graduates who will join Zimbabwean businesses over the next few years.

Limitations of the MIT Study and What They Mean for Zimbabwe

The authors present the work as a preliminary guide, released before peer review. Their own list of limitations includes a narrow participant pool, a single AI model, a single task that was not broken into stages such as idea generation and drafting, and EEG's inability to see deeper brain structures. A commentary posted in January 2026 raised further concerns about sample size, reproducibility, EEG methods and inconsistencies in how some results are reported.

There is also a question of fit. The participants were young students and researchers at American universities writing timed essays, not a bursar in Gweru answering parents. The study suggests how people engage with AI; it does not measure Zimbabwean workers.

Does AI Use Cause Dementia? What the Evidence Actually Says

The study is often shared alongside claims that AI use leads to dementia. It says no such thing, and it did not measure dementia or long-term brain health. Its authors asked journalists not to describe the results with words such as harm, damage or brain rot.

The dementia link comes from elsewhere. UCL's Professor Aimee Spector has said that constantly using AI instead of working things out could increase dementia risk, because the brain needs exercise. Separately, the phrase cognitive debt was used years earlier in Alzheimer's research for a different idea, about repetitive negative thinking. These warnings describe a plausible long-term risk that has not been demonstrated, which is exactly the distinction AI users need to make about everything they read.

Thinking With AI vs Depending on AI

The most useful result in the MIT study gets the least attention. Participants who had written unaided for three sessions and then used ChatGPT in the fourth showed higher memory recall and activation patterns similar to the search group. They wrote more information-seeking prompts, and several described using ChatGPT to add evidence to an essay they had already thought through. The tool added to their thinking instead of replacing it.

Why Thinking First and Using AI Second Works Better

That pattern is the practical definition of working with AI. Dependency is asking the tool for the answer and using whatever comes back. Collaboration is forming your own view first, then using the tool to challenge, extend or tidy it, and checking the result before it leaves your hands. One participant in the ChatGPT group put it simply: the tool helped with grammar, but everything else came from the brain.

Earlier research reviewed in the paper points the same way: stronger learners used AI to revisit and pull material together, while weaker learners leaned on the instant answer. Given the study's size, this is a direction rather than a proven rule, but it matches our experience that people who know what a good answer looks like get far more from AI than people who hope the AI knows.

How to Tell Dependency From Collaboration at Work

What Closing Zimbabwe's AI Literacy Gap Requires

If the gap is understanding, the answer is education that builds judgement, not only awareness. Three things matter most.

Why AI Literacy Must Cover More Than One Tool

Knowing one assistant is like knowing one supplier: you have no way to tell whether its answer is any good. Literacy starts with knowing that ChatGPT, Gemini, Claude, Meta AI and others exist, that they behave differently, and that comparing two answers is one of the quickest checks available. It also means using them for productive work such as research, drafting, proofreading and weighing a decision, not only for quick questions in a chat. The habits carry across tools, so someone who learns them on ChatGPT or Gemini applies them to the Meta AI already on their phone.

Why Verification Is the Core AI Skill

Verification habits are learnable, and most are simple. Ask the tool for its sources and open them, because AI tools can cite references that do not support what they claim. Put the same question to a second assistant, and check figures against an official source such as ZIMRA or the relevant regulator.

Research reviewed in the MIT paper found that generative AI shifts people's effort from finding information to checking and editing what the AI produced. Once the tool drafts, verification becomes the main job. It belongs at the centre of any AI education, not in a closing warning.

How Managers Can Build a Think-First Habit at Work

In our experience, a few working rules do more than a training budget. Ask staff for their own first answer before they open an assistant, so the AI responds to their thinking instead of replacing it. Require a source for any fact that goes to a client, and treat anything about tax, contracts, fees or health as a draft until the person responsible has checked it. The dependency table above doubles as a checklist.

Why AI Literacy Should Come Before AI Adoption in Zimbabwe

The MIT paper contains a line addressed directly to AI models, telling them to read only its summary table. Anyone who asked an assistant to summarise it may have received a summary of one table from a 206-page paper. The dementia claim spread the same way: a hedged preprint became a confident warning as each person repeated the version they received without opening the source. The skill that would have caught both is the skill that makes AI useful, which is checking what the source actually says.

Our earlier work argued that digitisation and skills must come before AI applications, and AI literacy is the individual version of that sequence. That reasoning is why we are building Sparkline AI Foundations, a free course for beginners. Sparkline Labs engineers systems around the way Zimbabwean businesses actually work, and for many of them the first AI system worth building is not software. It is a team that knows how to use the assistant already on their phones, and when not to trust it.

Sources

1. Muonwa, J. (2026, July 17). AI gains ground as Zimbabweans turn to digital learning. NewZimbabwe.com. Reports the 2025 ICT Access by Households and Use by Individuals Survey Report by ZIMSTAT and POTRAZ.
2. Bizcommunity. (n.d.). Meta slides AI assistant into DMs in 7 African countries. Bizcommunity Uganda.
3. Techzim. (2025, February). ChatGPT on WhatsApp: Now Smarter with Image & Audio Uploads. Techzim.
4. Techzim. (2024, April). AI for everyone: Zuckerberg puts powerful Meta AI right inside WhatsApp. Techzim.
5. Sparkline Labs. (2026). Why Zimbabwean Businesses Run Operations Through WhatsApp.
6. Statcounter. (2026). AI Chatbot Market Share Zimbabwe. Statcounter Global Stats. September 2026 data, accessed October 2026.
7. Statcounter. (2026, April 2). Google Gemini overtakes Perplexity to become second largest source of AI chatbot referrals to websites. Statcounter Global Stats press release.
8. Sparkline Labs. (2026). AI in Zimbabwean Business: What It Has Enabled and What Must Be Digitised First.
9. Kosmyna, N., Hauptmann, E., Yuan, Y. T., Situ, J., Liao, X.-H., Beresnitzky, A. V., Braunstein, I., and Maes, P. (2025). Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task. arXiv preprint arXiv:2506.08872. MIT Media Lab.
10. MIT Media Lab. (2025). Your Brain on ChatGPT: Project Overview and FAQ. MIT Media Lab.
11. Stankovic, M., Hirche, E., Kollatzsch, S., and Doetsch, J. N. (2026, January). Comment on: Your Brain on ChatGPT: Accumulation of Cognitive Debt When Using an AI Assistant for Essay Writing Tasks. arXiv.
12. UCL News. (2026, July 14). Relying on ChatGPT may increase dementia risk. University College London. Quotes Professor Aimee Spector.
13. Marchant, N. L., and Howard, R. J. (2015). Cognitive Debt and Alzheimer's Disease. Journal of Alzheimer's Disease, 44, 755-770.