New study reveals how AI usage differs by country income and language

A new paper on arXiv from researchers Madeleine Daepp and Isaac Slaughter examines how early adopters worldwide used a widely available free AI chatbot, analyzing a large-scale dataset of anonymized interactions. The key finding: in most countries, especially low-income ones, schooling is the most common use case for generative AI. There is a strong inverse association between schooling-related usage and a country's GDP per capita. Conversely, leisure-related use (e.g., entertainment, casual conversation) is positively correlated with national income. This suggests that AI is being adapted to local needs – with educational applications dominating where formal education resources may be scarce.

The study also highlights the role of language. English-language interactions were overrepresented even in regions where English is not the primary language, indicating that current AI models serve English speakers better. The authors argue that improving performance across languages could be critical in determining whether generative AI widens or narrows digital divides. If models become more inclusive linguistically, they could enable leapfrogging in education and economic opportunities for lower-income countries. The paper contributes empirical evidence to ongoing debates about AI equity and global adoption patterns.