Data collection is never a one-size-fits-all activity. Nowhere is this more evident than in low-literacy markets, particularly across parts of Africa, where traditional assumptions about reading, comprehension, and survey behavior quickly fall apart.

The challenge isn’t just what questions are asked, but how people experience the act of answering them. In addition, a major barrier to conducting in-person research in sub-Saharan Africa is population density. With roughly 53 people per square mile, communities are often dispersed, making face-to-face interviews or focus groups time-consuming and logistically challenging. 

Even with sales agents or local enumerators, a single session can last 1–3 hours per community, and scaling across multiple regions becomes resource-intensive. Literacy levels, language fluency, cultural norms, and access to technology all shape how data is shared, interpreted, and trusted. When these realities are ignored, research may still produce clean-looking datasets, but the insights behind them can be deeply misleading. So what is it about low-literacy markets that actually requires a different approach?

The Difference

In many low-literacy settings, participants may:

• Be more comfortable speaking than reading or typing
• Rely on oral explanations, examples, and tone to understand meaning
• Feel intimidated by written surveys or formal research language
• Interpret questions contextually rather than literally

This  means the research interface needs to meet the respondents where they are. When data collection tools don’t align with lived realities, respondents adapt in ways researchers don’t always notice, such as guessing, speeding through questions, or deferring to what they think the “right” answer is.

Start with people, not tools

In low-literacy markets, the most important decision is whether the method fits how people actually communicate. Before choosing a tool, teams need to ask ourselves,‘ Who are we speaking to? How do they naturally share information? What level of depth is required? And where does consistency matter more than flexibility?‘. 

When these questions are answered honestly, methodology stops being purely technical and becomes a design decision. One that directly determines who is included, who is excluded, and how reliable the resulting insights will be.

So is it Text vs Voice Surveys?

Voice and text surveys have long played complementary roles in low-literacy contexts. Voice-based methods often feel more natural and inclusive, especially for rural, older, or less formally educated populations. Hearing questions in familiar languages or accents reduces cognitive load. This allows meaning to be conveyed through tone, and makes participation possible without reading or writing. This often leads to richer, more nuanced responses.

Text surveys still have value, particularly in urban or semi-literate settings, among younger respondents, or where privacy, speed, and scale are priorities. They are easier to standardize and deploy, but only when language is simplified, culturally adapted, and carefully designed. The real risk is in using the wrong method for the wrong audience and mistaking clean-looking data for good data.

Where nuance and AI influence the choice

The real complexity shows up in execution. Voice surveys can introduce interviewer or moderation effects, even when automated. Text surveys can mask misunderstanding behind seemingly complete answers. Translation quality, question length, pacing, and examples matter far more in low-literacy markets than many teams anticipate. Small design choices can significantly alter how responses are interpreted.

AI is increasingly helping researchers navigate these trade-offs. AI-moderated voice interviews can bring consistency to spoken methods, while AI-assisted analysis makes it possible to process open-ended responses at scale. But AI is not a shortcut. Its effectiveness depends on language coverage, accent recognition, cultural context, and thoughtful human oversight. In some cases, AI delivers scale and standardization. In others, especially where meaning is highly localized, human judgment remains essential. The future is about choosing AI and people/voice and text, combining them deliberately to match context, audience, and research intent.

A note for practitioners

If you’re already collecting data in low-literacy markets or preparing to, this conversation matters. The goal is to be more intentional about how tools, technology, and methodology interact with real people’s lives. Better data doesn’t start with smarter software. It starts with understanding the context you’re working in.