Bridging the Digital Divide: How South African AI is Fighting Misinformation in Indigenous Languages
The internet is a global marketplace of ideas, but it is also a breeding ground for a dangerous commodity: misinformation. In the English-speaking world, artificial intelligence (AI) has become a critical first line of defense, with sophisticated algorithms scanning social media, news articles, and search results to flag false or misleading content. Yet, this technological shield is profoundly unequal. When the same falsehoods are expressed in isiZulu, Sepedi, or any of South Africa’s other official languages, the AI that protects English speakers often goes blind. These languages are considered “low-resource” in the field of natural language processing, meaning there is insufficient digital text available for AI models to learn from, making them virtually invisible to automated detection systems. This linguistic digital divide leaves millions of South Africans vulnerable to a tsunami of online misinformation, often with devastating real-world consequences, from vaccine hesitancy to ethnic tensions. A groundbreaking PhD study conducted at North-West University by Dr Seani Rananga, recently profiled in the Public Sector Manager, is now taking direct aim at this problem, demonstrating that it is possible to build AI systems that operate across English, isiZulu, and Sepedi to detect misinformation and create a vital foundation for a safer digital ecosystem in South Africa.
At the heart of Rananga’s research is a sophisticated multilingual framework designed to identify misleading information with equal accuracy across the three target languages. The study used the Covid-19 pandemic—a period described by the World Health Organization as an “infodemic” due to the sheer volume of false information—as a rigorous test case. The findings were unequivocal: AI trained to work multilingually can effectively detect misleading information in English, isiZulu, and Sepedi, proving that a one-size-fits-all approach is not a prerequisite for safety. The research delved deeper into the mechanics of how these systems work, revealing that the quality of automated translation is the single most significant factor in improving AI performance for languages with limited digital content. In essence, for AI to understand and flag a false claim in Sepedi, the system must first accurately translate the nuances of Sepedi into a form that the core AI model can process. This finding underscores the need for significant investment in machine translation tools tailored to African languages. “The framework provides a foundation for building more inclusive and trustworthy AI technologies for South African languages,” Rananga told the Public Sector Manager. The implication is clear: we do not need to wait for a perfect, English-dominated internet; we can build the tools to protect speakers of every language.
The motivation behind the research stems from a stark reality: South Africa is one of the most linguistically diverse countries on Earth, yet the digital realm has long ignored this diversity. “South Africa is one of the most linguistically diverse countries in the world, yet most AI systems for detecting misinformation are developed primarily for English. This leaves speakers of indigenous languages at a disadvantage when it comes to identifying false or misleading information online,” said Rananga, a lecturer at the University of Pretoria. This disadvantage is not merely a matter of inconvenience; it is a threat to democracy, public health, and social cohesion. When citizens rely on community media, WhatsApp, and Facebook in their mother tongue for news, they are navigating a minefield of unmoderated content. Without AI moderation in their language, they are more likely to fall for false narratives about elections, health protocols, or natural disasters. Rananga’s research is thus a form of technological activism, aiming to ensure that the benefits of the AI revolution are not reserved for a global English-speaking elite but are shared by all South Africans. It acknowledges that credible information is a public good, and that everyone deserves the right to make informed decisions, regardless of the language they speak.
While the research was validated using health misinformation, the framework’s applications are remarkably broad. The most immediate and critical use case is elections. South Africa has seen misinformation used as a weapon to undermine electoral integrity, and a system that can flag false claims in isiZulu or Sepedi in real-time could be a game-changer for the Electoral Commission and civil society organizations. Similarly, in public emergencies like floods or pandemics, the rapid detection of false safety advice could save lives. The study suggests that government departments, fact-checking organizations such as Africa Check, and social media platforms can all leverage this framework. For platforms, it offers a path toward more responsible content moderation; for fact-checkers, it reduces the manual burden of monitoring multiple languages; and for government, it provides a tool to monitor public sentiment and counter misinformation with authoritative, accurate responses. The framework is designed to be scalable, meaning that the underlying architecture can be adapted with new data to tackle different types of false information, from agricultural scams to mining safety rumors. It represents a scalable, low-cost solution to a problem that is currently being managed, rather than solved, through ad-hoc human moderation.
The academic community has taken notice. Rananga’s groundbreaking work has earned her the prestigious Google PhD Fellowship, a globally competitive award that supports exceptional doctoral research in AI and is renowned for identifying the technology leaders of tomorrow. The fellowship not only provides funding but also opens doors to collaboration with Google’s research teams. Her research poster was also honored with the Best Poster Award at the 2025 Deep Learning IndabaX South Africa, which earned her funding to present the work at the 2026 Deep Learning Indaba at Pan-Atlantic University in Lagos—the premier African conference on AI. “Receiving the Google PhD Fellowship was a tremendous honour and a strong validation of the importance of conducting AI research for African languages,” said Rananga. “It reinforced my commitment to developing technologies that address challenges faced by multilingual and low-resource communities while also creating opportunities to collaborate with leading researchers around the world.” This international recognition signals a shift in the global AI research agenda, confirming that the challenges of low-resource languages are not peripheral but central to the future of an inclusive AI ecosystem.
Looking forward, Rananga has an ambitious and inspiring roadmap. She plans to expand the research to include more of South Africa’s official languages, moving beyond isiZulu and Sepedi to ensure no language community is left without digital protection. Her future work will also pivot to the next frontier of misinformation: deepfakes, the hyper-realistic but fake videos and images generated by AI. Developing methods to detect these synthetic media in a multilingual context will be crucial. Beyond detection, her vision is to build trustworthy multilingual AI systems, including large language models for African languages that can understand and generate text; retrieval-augmented systems that check facts against authoritative sources before answering; and knowledge graphs that map the connections between entities and claims. These tools have potential applications across healthcare, education, agriculture, and governance. “My long-term vision is to ensure that African languages are fully represented in the next generation of AI technologies while contributing to safer and more reliable digital spaces for everyone,” she said. As South Africa grapples with the double-edged sword of technological advancement and information chaos, Rananga’s research offers a blueprint for a future where safety online is not a privilege of language, but a universal right.
