A message lands in a family WhatsApp group. It is written in isiZulu, and it warns that a certain vaccine will cause infertility, or that a well-known political figure has secretly resigned. It sounds urgent, and many people forward it without checking. In English, automated fact-checking tools and artificial intelligence systems might flag that message as false within seconds. But in isiZulu, Sepedi, or many of South Africa’s other indigenous languages, the same misinformation often slips through unnoticed. That is because most AI systems used to detect misleading content are built around English and a handful of other well-resourced languages. South Africa is one of the most linguistically diverse countries in the world, yet its speakers of indigenous languages are largely invisible to these technologies. This is not a small problem. Misinformation can affect health decisions, election outcomes, and public safety. It can make people refuse life-saving treatment, or distrust legitimate information, or vote based on lies. The gap has serious consequences, and for a long time, little was done to fix it.
That is what makes the work of Dr Seani Rananga so significant. Rananga, a lecturer at the University of Pretoria, completed a PhD at North-West University that set out to build AI that can identify misinformation in more than just English. Her research focused on three languages: English, isiZulu, and Sepedi. The goal was not simply to translate existing detection tools, but to design a multilingual framework that treats each language fairly and understands the ways misinformation travels within different language communities. According to a recent article in Public Sector Manager, the study showed that AI can effectively detect misleading information across all three languages, even though isiZulu and Sepedi have far less digital content available for training AI systems. That matters because many low-resource languages are left behind in AI development. The more data an AI system has, the better it performs; when data is scarce, the technology struggles. Rananga’s research did not pretend this obstacle away. Instead, it confronted it directly and offered a way forward. It demonstrated that with carefully designed methods and an emphasis on improving automated translation, it is possible to create tools that work for languages that have historically been excluded. The implications go far beyond academic publishing. They open the door to safer digital spaces for millions of South Africans who communicate primarily in their home languages.
The motivation behind the research is deeply rooted in fairness and inclusion. South Africa’s constitution recognises twelve official languages, and the country is home to a rich variety of cultures and traditions. But when it comes to technology, not all languages are treated equally. English dominates the internet, the media, and the algorithms that shape our digital lives. For people who speak isiZulu or Sepedi as their first language, this creates an everyday disadvantage. They may have access to smartphones and social media, but the tools meant to protect them from false information are often simply not there. Rananga recognised this early on. “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,” she said. “This leaves speakers of indigenous languages at a disadvantage when it comes to identifying false or misleading information online.” Her research was driven by a desire to change that. She wanted to make sure that, as AI becomes increasingly central to how we understand and navigate the world, speakers of indigenous languages are not left behind. That is not just a technical challenge. It is a question of equity and dignity. In a democracy, people deserve access to information they can trust in a language they understand. If technology only serves English speakers, it fails to serve the nation as a whole.
To test her framework, Rananga used misinformation about the Covid-19 pandemic as a case study. The pandemic was a period of intense fear, confusion, and uncertainty. Every day brought new claims about treatments, vaccines, government restrictions, and conspiracy theories. Some of that misinformation was harmless, but a great deal of it was not. It led people to drink dangerous substances, to avoid hospitals, and to spread panic. The pandemic proved how quickly false information can spread and how serious the consequences can be. By using pandemic-related misinformation as her test case, Rananga was able to see whether her multilingual AI framework could hold up under real-world pressure. The findings were encouraging. The AI was able to identify misleading information across English, isiZulu, and Sepedi. The study also highlighted the importance of good automated translation. For languages with limited digital content, translation quality plays a huge role in how well the AI performs. When translation is poor, the AI loses context and nuance, and its ability to detect misinformation weakens. But when translation is improved, the AI performs significantly better. This insight is crucial. It shows that tackling misinformation in low-resource languages is not just about building bigger models; it is about building smarter ones that respect the structures and meanings of each language. The methods were designed with flexibility in mind, so they can be adapted to other sectors where misinformation poses a public risk, not just health. Elections, natural disasters, and public emergencies all create fertile ground for false information.
The potential real-world impact is enormous. Imagine a government department trying to issue a public health warning in Limpopo, or an election commission trying to counter a claim that polling stations are being moved in KwaZulu-Natal. In English, official responses can be amplified and checked using AI tools. In isiZulu or Sepedi, those responses might go unnoticed while false content spreads organically through social networks. Rananga’s framework could help change that. By enabling earlier detection of misleading information, it gives fact-checking organisations, social media platforms, and community leaders a better chance to respond quickly and effectively. It also improves access to credible information. When people can receive accurate news in their own language, they are better equipped to make informed decisions about their health, their families, and their votes. This is not only about protecting individuals. It is about strengthening trust in institutions, in the media, and in the democratic process itself. Misinformation flourishes when people feel ignored or excluded. When communities see that technology is designed with them in mind, they are more likely to trust it. The framework also lays a foundation for more inclusive AI technologies more broadly. It is not a one-off solution, but a starting point. It proves that African languages are not obstacles to AI development; they are opportunities for innovation.
The research has already earned international recognition. Rananga received the Google PhD Fellowship, one of the most prestigious awards for doctoral research in artificial intelligence. The fellowship is given to promising researchers working on bold and important problems, and it placed Rananga among a global community of scholars pushing the boundaries of AI. She also received the Best Poster Award at the 2025 Deep Learning IndabaX South Africa, which earned her funding to present her work at the 2026 Deep Learning Indaba at Pan-Atlantic University in Lagos, Nigeria. The Deep Learning Indaba is a major gathering of African AI researchers, and presenting there is a significant honour. For Rananga, the recognition is both personal and symbolic. “Receiving the Google PhD Fellowship was a tremendous honour and a strong validation of the importance of conducting AI research for African languages,” she said. “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.” That last point matters. International awards are not just accolades. They open doors. They allow Rananga to connect with other researchers, to share her findings with a wider audience, and to attract attention and resources to a field that needs more support.
Looking ahead, Rananga has ambitious plans. She intends to expand the research to include more South African languages, so that the tools can serve even more communities. Future work will focus on misinformation shared on social media during elections, public emergencies, and other high-impact moments. These are periods when false information can do the most damage, and when having reliable, multilingual detection tools is essential. She also wants to address emerging challenges like deepfakes, the AI-generated videos and images that look real but are not. Deepfakes are a growing threat in every language, but they are especially dangerous in societies where trust in information is already fragile. Rananga also plans to extend her research to hate speech and other harmful online content. Beyond detection, her long-term vision is to build trustworthy multilingual AI systems. That includes large language models, which can understand and generate human language; retrieval systems that can look up and verify information before giving an answer; and knowledge graphs that map how pieces of information connect to each other. These tools could support healthcare, education, agriculture, governance, and public services. They could help a nurse in a rural clinic find accurate guidance, or help a teacher explain a complex topic to students in their mother tongue. For Rananga, the goal is not simply to catch false information after it spreads. It is to create a digital environment where accurate information is accessible, reliable, and inclusive. “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. That vision is not only about technology. It is about people, and about making sure that no language, and no community, is left behind in the digital future.