Large language models enable large-scale analysis of human-bird relationships in South African cities

Sage K. Naidoo^1*Chevonne Reynolds²

• ¹University of the Witwatersrand Johannesburg, Johannesburg, South Africa
• ²University of the Witwatersrand, Johannesburg, South Africa

Human perceptions of urban wildlife can shape conservation priorities and public support for biodiversity initiatives, however research on human-bird relationships remains spatially biased towards the Global North. Here we assessed the perceptions of 36 urban bird species across four South African urban contexts using a mixed-methods approach. Survey respondents (n = 1,977) rated species likeability on a 5-point Likert scale and provided open-ended explanations for their ratings. Quantitatively, South African urban birds were generally well-liked, with notable variation among species: the Malachite Kingfisher (Corythornis cristatus, mean ± SE = 4.91±0.02) and Orange-breasted Sunbird (Anthobaphes violacea, 4.91±0.02) scored highest, and the Common Myna (Acridotheres tristis, 2.50±0.03) scored lowest. To analyse the approximately 71,000 open-ended responses, we employed ChatGPT, a generative AI large language model, to identify eight themes underlying species appeal. The highest-rated species were primarily valued for aesthetic appeal and emotional connections, while the lowest-rated species were associated with aggressive behaviours and negative ecological impacts. Factor analysis revealed three perceptual clusters, demonstrating that some species evoke multidimensional responses whilst others are viewed through a single dominant lens. Notably, aesthetic patterns did not universally predict appeal and many highly rated raptor species were valued for emotional connections rather than physical traits. Additionally, negative perceptions did not apply uniformly to all non-native or problematic species, with some receiving moderately positive responses despite ecological concerns. These findings highlight the complexity of human-bird relationships in urban contexts and demonstrate that large language models can enable qualitative analysis at large scales. By offering an African perspective, this study contributes to a more inclusive understanding of how urban residents perceive and value birds.