Sage K. Naidoo
Chevonne Reynolds- School of Animal, Plant and Environmental Sciences, University 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.
1 Introduction
Rapid urban development is transforming human-nature relationships worldwide, as expanding cities fragment natural habitats, create novel built ecosystems, and alter how urban residents perceive and engage with urban wildlife (Seto et al., 2011; McKinney, 2006; Elmqvist et al., 2015; Belaire et al., 2015). This transformation is particularly pronounced in African cities, where rapid urbanisation intensifies pressures on urban wildlife (Güneralp et al., 2017; Reynolds et al., 2021; Ayeni et al., 2023) while simultaneously reducing residents’ access to diverse natural experiences (Chamberlain et al., 2019; Lee et al., 2021; Awoyemi et al., 2024). Understanding how urban residents perceive and value urban wildlife has become increasingly important, as these relationships form the foundation for cultural ecosystem services, i.e., the “non-material, or intangible, benefits derived from nature” (Millennium Ecosystem Assessment, 2005), which are essential for improving human well-being in urban environments (Plieninger et al., 2013; Cox and Gaston, 2016).
Birds are an ideal taxon for examining human-wildlife relationships in urban contexts due to their visibility, recognisability, and widespread distribution (Belaire et al., 2015; Gaston et al., 2018). African urban environments present particularly complex interaction contexts, shaped by both ecological and cultural factors. Ecologically, residents encounter diverse assemblages ranging from highly localised endemic species to widespread native species and problematic introduced species (van Rensburg et al., 2009; Kopij, 2015; Lee et al., 2017; Lee and Barnard, 2016; Shivambu et al., 2020). For example, South African cities host endemic species such as the Cape Sugarbird (Promerops cafer) alongside urban adapters such as the Hadeda Ibis (Bostrychia hagedash) and invasive species such as the Common Starling (Sturnus vulgaris; Lee and Barnard, 2016; Lee et al., 2017; Shivambu et al., 2020). Additionally, the rich cultural significance of birds in many African societies, where different species may hold traditional, spiritual, or symbolic meaning, further influences contemporary urban perceptions (Gora, 2009; Williams et al., 2012, 2014; Smith et al., 2021). This ecological and cultural diversity provides opportunities to investigate how aesthetic, behavioural, and cultural factors interact to shape urban residents’ perceptions of different species.
Research across multiple contexts demonstrates that while urban birds are generally well-liked, resident perceptions vary substantially among species (Clergeau et al., 2001; Belaire et al., 2015; Cox and Gaston, 2015; Lišková and Frynta, 2013; Tisdell et al., 2005). This species-level variation represents a critical gap in understanding how human-bird relationships function in urban environments. Individual perceptions of birds are influenced by multiple interacting factors, with aesthetic value and behavioural characteristics being primary drivers (Cox and Gaston, 2015; Collins et al., 2021; Andrade et al., 2022; Champness et al., 2023). Birds that are smaller, more colourful, patterned, or uniquely coloured tend to be more positively perceived than larger or more plainly coloured birds (Collins et al., 2021; Andrade et al., 2022). For example, Collins et al. (2021) observed that urban residents preferred songbirds with distinctive colour schemes (e.g., blue, or black) over species with more common or simple colouration (e.g., yellow/orange). Conversely, behavioural characteristics often drive negative perceptions of birds within urban environments (Perry et al., 2020; Champness et al., 2023). Within South African urban contexts, the Common Myna (Acridotheres tristis), is frequently regarded as an undesirable species due to its aggressive behaviour towards other bird species and its non-native status (Perry et al., 2020; Champness et al., 2023). These aesthetic and behavioural traits also affect how recognisable species are, which further shapes individuals’ perceptions (Donázar et al., 2016; Perry et al., 2020; Champness et al., 2023). Highly recognisable species tend to elicit stronger personal and emotional connections, potentially encouraging greater participation in urban birding activities or increased learning about urban ecosystems (Champness et al., 2023; Belaire et al., 2015; Riechers et al., 2016; Hedblom et al., 2017; Hegetschweiler et al., 2022). When positive, these relationships encourage engagement with urban nature and contribute to human well-being (Cox and Gaston, 2015; Belaire et al., 2015; Hedblom et al., 2017).
Despite the importance of understanding these preference patterns for urban conservation and management, quantifying human-wildlife relationships at meaningful scales remains challenging due to the subjective nature of individual human responses (Chan et al., 2012; Boerema et al., 2017; Methorst et al., 2021). Capturing these nuanced and often contradictory perceptions requires analysis of extensive community feedback, yet traditional approaches typically rely on small-scale qualitative studies or limited surveys, constraining our ability to understand human-wildlife relationships at scale. The recent emergence of large language models (LLMs) such as ChatGPT (OpenAI, 2024), Claude (Anthropic, 2025), and similar artificial intelligence systems presents new opportunities for processing extensive community-generated text data. These tools can enable systematic analysis of thousands of qualitative responses while maintaining the nuance needed to capture the complexity of factors that shape preference and appeal (Wang et al., 2023).
Existing research on human-nature relationships is strongly biased towards developed countries in the Global North (Clergeau et al., 2001; Belaire et al., 2015; Cox and Gaston, 2015). This leaves African urban contexts particularly understudied despite the continent’s rich avian diversity, high levels of endemism, rapid urbanisation rates, and unique socio-cultural relationships with wildlife (Aouissi et al., 2021; Collins et al., 2021; McPherson et al., 2021). In this study, we combined quantitative and qualitative data to assess human perceptions of urban bird species across South Africa using large language model analysis. Specifically, we measured how likeable different urban bird species are across four of South Africa’s most urbanised provinces, then employed ChatGPT to analyse approximately 71,000 open-ended responses about 36 urban bird species. Our research questions were threefold. First, we asked how likeability varies among urban bird species in South Africa. Second, we examined which dominant perceptual themes emerge from large-scale text responses and how these themes contribute to understanding species likeability. Third, we investigated how these perceptual themes interact with one another, including potential synergies and trade-offs, and how such interactions relate to broader patterns of species likeability.
While ChatGPT has been used previously as a sentiment analysis tool for extracting information from social media or other text-based sources to discern social and cultural patterns, primarily in psychology (e.g., Wang et al., 2023; Buttrick, 2024), our study represents, to our knowledge, one of the first applications of LLMs in research on human-nature relationships. We identified major themes underlying species perceptions and examined how these factors interact to create complex, multidimensional human-bird relationships. This approach enables processing of individual responses at unprecedented scale providing both methodological innovation for community engagement and practical insights for conservation messaging in urban African contexts.
2 Methods
2.1 Methods overview
We assessed human perceptions of urban bird species using a mixed-methods approach that combined quantitative likeability ratings with large-scale qualitative analysis. Online questionnaires were distributed across four South African provinces, receiving 1,977 responses from urban residents who rated bird species on a 5-point Likert scale and provided open-ended explanations for their ratings. We assessed 13 species per province (36 species total, with 11 species assessed across multiple provinces) and analysed approximately 71,000 open-ended responses using ChatGPT to identify eight thematic categories. We then used factor analysis to examine relationships between themes and to characterise the multidimensional nature of urban residents’ perceptions of birds. Ethical clearance for this research was provided by the University of the Witwatersrand Human Research Ethics Committee (Non-medical; Clearance number: H23/06/29).
2.2 Study area
We conducted this study across four South African provinces (Figure 1) encompassing the country’s major metropolitan areas: Gauteng (including Johannesburg and Tshwane), Western Cape (Cape Town), KwaZulu-Natal (Durban), and Free State (Bloemfontein). South Africa is one of the most diverse countries in the world with a broad ethnic, cultural, linguistic, and religious composition (Mino, 2013). It is home to ~62 million people (~81% Black, ~7% White, and ~3% Indian/Asian, among other demographic groups; SA Census, 2022). Following the end of Apartheid in 1994, rapid urban development has contributed to ~68% of the population now living in cities (Collinson et al., 2007; Swilling, 2010; Todes et al., 2010; O’Neill, 2023; Statistics SA, 2023). South Africa is also recognised as one of the world’s 18 megadiverse countries, due to its exceptional richness of endemic plants, vertebrates, and invertebrates across diverse terrestrial and aquatic ecosystems (Egoh et al., 2009; Maphisa et al., 2016; Van Wilgen et al., 2020).
Figure 1. Map of South Africa showing the four study provinces: Gauteng, Western Cape, KwaZulu-Natal, and Free State. Points indicate locations of survey responses.
The four provinces represent diverse ecological contexts, from the Mediterranean climate of the Western Cape to the subtropical conditions of KwaZulu-Natal and the temperate highveld environments of Gauteng and Free State (Rutherford et al., 2006; Finch and Meadows, 2018). Urban areas across these provinces support varied bird communities including endemic South African species, widespread native species, and introduced species (Kopij, 2015; Lee et al., 2017; Shivambu et al., 2020). For example, Cape Town hosts endemic species such as Cape Sugarbird and Orange-breasted Sunbird (Anthobaphes violacea), alongside introduced species like Common Starling (Lee and Barnard, 2016; Lee et al., 2017; Shivambu et al., 2020; Supplementary Table S1). Johannesburg supports a mix of native species, e.g., African Hoopoe (Upupa africana) and African Harrier-hawk (Polyboroides typus), alongside the introduced Common Myna and Rose-ringed Parakeet (Psittacula krameri), which have established populations in many urban areas (Peacock et al., 2007; Ivanova and Symes, 2019; Shivambu et al., 2020; Supplementary Table S1). This diversity of urban bird assemblages across provinces provided an opportunity to examine human responses to a wide range of species in different ecological and urban contexts.
2.3 Species likeability and perceptions
2.3.1 Questionnaire
We designed online questionnaires using Google Forms to capture both quantitative and qualitative components of urban residents’ perceptions of birds (Supplementary Materials B). Four separate questionnaires were created, one for each province, to account for regional differences in urban bird species composition across South Africa’s diverse ecological contexts. This approach allowed us to capture the unique avifauna of each region, including local endemics and invasive species, which would have been poorly represented using a single national questionnaire. Each questionnaire included species encountered in the respective metropolitan areas and selected along a common-to-rare gradient (see Section 2.3.2; Supplementary Table S2).
2.3.2 Study species selection
We used the South African Bird Atlas Project 2 (SABAP2) database to select a stratified representation of bird species that people may encounter across the four different provinces. SABAP2 is an ongoing (2007-current) citizen science project mapping the distribution of birds across southern Africa. Volunteers contribute to SABAP2 by recording birds detected within 5x5 minute longitudinal and latitudinal grid cells (‘pentads’, approximately 9x9 km; Brooks et al., 2022). Atlassing is conducted over a two-hour to five-day window within a pentad, where observers are required to identify all bird species seen or heard (Brooks et al., 2022). A pentad can only be sampled once every five days by the same observer (Brooks et al., 2022). To date, SABAP2 has generated over 22 million species records across southern Africa, contributed by thousands of citizen scientists.
For this study, we focused species selection on regions surrounding the metropolitan areas of Johannesburg/Tshwane (Gauteng), Cape Town (Western Cape), Durban (KwaZulu-Natal), and Bloemfontein (Free State). Within each metropolitan area, we randomly selected one pentad from the SABAP2 database. We then used SABAP2 reporting rates (i.e., the frequency of how often individual species are recorded across sets of full protocol cards, expressed as percentages) to select bird species along a common-to-rare species observation gradient Supplementary Table S2). Specifically, we randomly selected 13 species for each province that ranged from having high reporting rates (i.e., commonly observed, 80-100%) to low reporting rates (i.e., rarely observed, 1-5%) within urban environments (Supplementary Table S2). Species with the highest and lowest reporting rates for each province are shown in Supplementary Figure S1.
To ensure broad representation of urban avifauna, we included both native and non-native (exotic) species (Supplementary Table S1). Each provincial questionnaire assessed 13 species (Supplementary Table S1), totalling 36 unique species overall. Of these, 11 species occur more widely across South Africa and were therefore evaluated in multiple provinces and questionnaires (Supplementary Table S1).
2.3.3 Questionnaire structure
We first provided prospective participants with background information about the study and requested informed consent to participate (Ethical Clearance Number: H23/06/29). Each questionnaire was divided into three main sections.
In the first section, we collected demographic information including age, gender, race, suburb, and dwelling type (Supplementary Table S3). These data were used to document the demographic composition of our respondent pool. We did not specifically analyse demographic differences in species perception, as our primary focus was on species-level perceptual patterns across the South African urban context. Nevertheless, we acknowledge that demographic bias in our responses Supplementary Table S3) is a limitation.
The second section focussed on participants’ interactions with birds. We asked questions such as “Do you believe birds add value to your environment?” and “Do you partake in any birding activities such as bird ringing or bird watching?” to assess general attitudes toward birds and level of birding experience. Question 3–6 from this section of the questionnaire was not analysed as this data forms the basis for another research paper.
The third section was designed to capture both quantitative and qualitative data on species likeability and perception. For each bird species, participants were presented with the common name and a single colour image (Supplementary Materials B). All images were sourced from the eBird website, selecting.jpeg files with a minimum resolution of 1000x1000 pixels that clearly depicted the full body of each bird with clear display of colours, patterns, and distinguishable traits. We standardised image selection to reduce potential bias in how birds were visually “framed” or presented.
For quantitative assessment, participants rated the appeal of each bird on a 5-point Likert scale (1 = Strongly Dislike, 5 = Strongly Like), drawing on both their response to the bird in the image and any personal experiences with the species. For qualitative assessment, an open-ended question (“What do you find appealing, or not, about [the bird in the image]?”) captured the reasoning behind each rating (Supplementary Materials B). Completion time for questionnaires ranged from 15–30 minutes depending on the detail provided.
2.3.4 Questionnaire distribution
We aimed to reach a diverse sample of urban residents across the four study provinces and designed the questionnaire to be accessible to anyone regardless of birding experience or expertise. Our goal was to capture broad community perspectives on urban bird species rather than restrict responses to experienced birders or ornithologists.
To maximise reach and demographic diversity, questionnaires were distributed across multiple platforms. The questionnaires were shared on various social media platforms, including Instagram, Facebook, X (formerly Twitter), Threads, WhatsApp, and LinkedIn. We also distributed questionnaires via email networks and through social media groups of birding organisations and local birding clubs, including BirdLife South Africa (BLSA) and its provincial subdivisions. Additionally, the study was also promoted through the BLSA online bird identification game Birdle (https://birdle.co.za/), which provided access to engaged birding communities.
Local nature reserves assisted with distribution, including Modderfontein Nature Reserve in Johannesburg and Friends of Franklin Nature Reserve in Bloemfontein, helping us reach visitors to urban green spaces. In Johannesburg, questionnaires were sent to the University of the Witwatersrand students through the institutional email list and were promoted via posters and business cards containing QR codes linking to the online questionnaire. These business cards were also distributed at various medical facilities, veterinary clinics, and animal-related businesses to reach a broader urban audience. Further reach was gained through a guest appearance on The Birding Life podcast.
Questionnaires remained open for responses for three months, from 29 September 2023 to 11 January 2024, after which they were closed due to declining engagement. A total of 1,977 responses were received: Gauteng received the most respondents (1,141 or 57.7%), followed by KwaZulu-Natal (426 or 21.5%), the Western Cape (369 or 18.7%), and the Free State (41 or 2.1%). While demographic information was collected to document respondent composition (Supplementary Table S3), our primary analysis focused on species-level perceptual patterns rather than demographic influences on bird perception.
2.4 Qualitative analysis of species perceptions
We analysed the open-ended responses to identify the key themes influencing species perceptions and human-bird relationships. Given the open-ended, subjective nature of these responses and the large dataset (~71,000 responses across 36 species), we employed ChatGPT 4.0 (OpenAI, 2024; accessed June 2024) to categorise the data. This approach minimised potential researcher bias in classification and enabled analysis of extensive community feedback at a scale not feasible with traditional qualitative methods.
Sentiment analysis is typically used to identify opinions, attitudes, and emotions within text drawn from sources such as social media posts, news articles, or in this case, open-ended survey responses (Medhat et al., 2014; Wang et al., 2023). The capacity of ChatGPT to perform such analyses has been previously evaluated (Wang et al., 2023). Wang et al. (2023) compared ChatGPT to the fine-tuned, small language model BERT (Bidirectional Encoder Representations from Transformers; Devlin et al., 2019), a widely used baseline model for sentiment analyses. Additionally, they tested ChatGPT against state-of-the-art (SOTA) models, which are typically developed for ‘task-specific’ applications (Wang et al., 2023). Their findings indicate that while ChatGPT performed comparably to the BERT model, it underperformed relative to domain-specific SOTA models. Despite these limitations, ChatGPT provided a suitable, general-purpose tool for our study, allowing us to analyse the ~71,000 open-ended survey responses to identify the main factors influencing perceptions of different bird species.
2.4.1 Testing ChatGPT in sentiment analysis
We assessed ChatGPT’s ability to perform sentiment analysis on data comparable to our questionnaire responses by prompting it to generate a test, or “dummy”, dataset of 30 responses. These were designed to resemble the open-ended responses in our questionnaires for three species: the Barn Owl (Tyto alba), White-eared Barbet (Stactolaema leucotis), and Sacred Ibis (Threskiornis aethiopicus). We then prompted ChatGPT to categorise these dummy datasets into major themes. From the first attempt, ChatGPT successfully recognised and classified the data into distinct themes. For example, the Barn Owl dataset was categorised under themes such as Beauty and Aesthetics, Hunting Skills and Predatory Nature, or Silent Flight. While these reflected meaningful traits, we noted these themes to be too species-specific to be applied consistently across all bird species in the questionnaires. To address this, we repeated the exercise with the White-eared Barbet, but instead prompted ChatGPT to generate a “step-by-step guide” for identifying broader themes of perceptual value. This produced more generalisable themes including Appearance, Behaviour, Sound, Cultural Significance, and Personal Experience.
We further refined this process by using actual participant responses to see how ChatGPT would classify them. For example, in responses about the Common Myna, many participants linked their likeability scores to its invasive status. We asked ChatGPT: “what type of category (in terms of a broad theme) could there be if a lot of people were to say they gave a bird a rating because of its ‘invasive nature’?”. ChatGPT generated a new theme of Ecological Impact, defined as “how the bird may be affecting the environment, other species, or human activities”. We also tested ChatGPT’s ability to identify responses that contradicted or opposed the major themes identified in the test datasets, which led to creation of a Negative/Neutral theme.
Finally, we tested ChatGPT with a subset of data from the actual questionnaires, focussing on responses for the Spotted Thick-knee (Burhinus capensis) in the Free State. We provided ChatGPT with a simple prompt: “This is a dataset of the opinions of the Spotted Thick-knee, can you identify the major themes?”. ChatGPT adequately categorised the responses into an initial seven themes: Appearance, Behaviour, Vocalisations, Adaptability, Camouflage, Emotional Impact, and Neutral/Negative. While themes such as Adaptability and Camouflage were more species-specific, the overall outcomes demonstrated that ChatGPT could generate qualitative themes broadly consistent with the training datasets used above.
2.4.2 Thematic categorisation using ChatGPT
We conducted a thematic analysis with ChatGPT to identify, analyse, and report the major themes underlying species perceptions. ChatGPT was instructed to read and contextualise the qualitative sentiment datasets to identify key points and recurring patterns (Supplementary Materials C). To guide this process, we provided specific keywords that could be used to assign thematic labels to different parts of the text. For example, when categorising responses linked to cultural significance, keywords such as ‘culture’, ‘symbol’, or ‘iconic’ were supplied to help group relevant responses under a cultural value theme (Supplementary Materials C). ChatGPT was also instructed to cross-check its categorisation to ensure that responses were accurately assigned to the appropriate themes.
From example datasets and trial runs, we identified the eight major themes most applicable across species, while also gaining insight into how ChatGPT generated them. The eight themes identified for final analysis included Appearance/Aesthetics, Behavioural Characteristics, Ecological Impact, Cultural Significance/Symbolism, Emotional Impact (Personal Connection), Vocalisation, Prevalence/Rarity, and Negative/Neutral (Supplementary Table S4). For the final assessment, ChatGPT was provided with the anonymised sentiment response dataset for each of the 36 species. These datasets contained only open-ended responses, with no identifying information. ChatGPT 4.0 had a maximum context window (token limit) of approximately 128,000 tokens, hence, to ensure the data fit within the model’s memory and did not exceed the token limit, we ran the ChatGPT categorisation process for each full species dataset separately. Each species dataset was accompanied by a detailed prompt to aid categorisation. An example of this prompt is included in Supplementary Materials C for the Hadeda Ibis. The prompt described each theme, listed relevant response types and keywords, and emphasised that a single response could be categorised into more than one theme to ensure complete representation. Prompts for all other species followed the same format (Supplementary Materials C) but were modified to include species-specific examples to clarify how and why responses should be categorised into any of the eight themes.
2.4.3 Refining ChatGPT’s thematic categorisation
Given the limited reported applications of ChatGPT for sentiment analyses, it was not entirely feasible to train the model to recognise and categorise every response with complete accuracy. It is therefore important to acknowledge the limitations of this analysis. Across the 36 species, the average ‘categorisation error rate’ (i.e., responses that were not categorised into any of the eight themes or were only assigned into one of multiple possible themes) was 15.13% (± 5.52). Uncategorised rationales typically fell into two themes: Appearance/Aesthetic and Emotional Impact (Personal Connection). In the case of aesthetics, spelling variation and the use of multiple synonyms or colloquial phrases frequently led to missed classifications. For example, the Grey Go-away-bird (Corythaixoides concolor) and Purple-crested Turaco (Tauraco porphyreolophus) were well-liked for the crest-like feathers on their heads. However, respondents used a wider range of descriptors such as “mohawk”, “mane”, or “kuif”, but also several misspellings of these terms (e.g., “maine” or “mowhawk”). This variation impacted how ChatGPT processed the content and affected ChatGPT’s ability to categorise these responses consistently.
ChatGPT also struggled with recognising emotion-based responses, particularly when the emotional connection was implied rather than explicitly stated. Straightforward expressions such as “I love owls” or “It’s a fish eagle! No need to say more” were readily classified as emotional. However, more complex, or metaphorical responses posed challenges. For example, statements such as “Love the way they drift like leaves in my garden” or references that described character-based traits, such as “Feisty characters and great mimics,” were not always identified as emotional connections. This suggests that ChatGPT may have difficulty interpreting underlying emotional nuance in text. To address these gaps, we manually assessed and categorised them to ensure all responses were appropriately themed, while also identifying reasons for misclassification. To maintain inter-rater consistency, all manual recategorisation of the analysis was conducted by SN.
This process served as a full validation step, as the manual review of all categorised responses allowed us to quantify agreement between ChatGPT’s initial theme assignments and the final corrected dataset. Across all 36 species, an average of 84.86% of responses were correctly categorised by ChatGPT, while 15.13% required manual adjustment (Supplementary Table S5). All corrections were made following a consistent decision protocol applied by a single reviewer (SN), ensuring internal consistency. Although inter-rater reliability metrics could not be calculated due to the use of single reviewer, this systematic review-and-correction procedure provides a transparent measure of classification accuracy.
Once datasets were processed and corrected, the number of responses within each theme was recorded. This enabled us to rank the themes by species and identify which themes were most prominent in shaping perceptions for each species. The ChatGPT classification, including manual categorisation, took approximately 100 hours over a 3-week period.
2.5 Data analysis
All analyses were conducted in R version 4.3.1 (R Core Team, 2024) using a mixed-methods framework to understand species-level variation in urban bird perceptions and the factors shaping human–bird relationships.
We calculated average likeability scores for each species using all available responses per species. For the 25 species assessed in a single province, overall means were computed directly. For the 11 species assessed across multiple provinces, provincial means were first calculated and then combined into overall averages weighted by sample size to account for uneven distribution in responses.
From the ChatGPT analysis, we extracted the number of responses categorised into each of the eight themes for each of the 36 species. We then calculated the proportion of responses in each thematic category per species, then ranked themes by prominence to identify which perceptual factors most strongly influenced species appeal. Heat maps were generated using the ggplot2 package (v3.4.2; Wickham et al., 2016) to visualise thematic rankings across species and highlight variation in how different birds were perceived. To address potential biases arising from selecting study species based on their SABAP2 reporting rates, we conducted a validation analysis using the Gauteng dataset (the largest sample size). We fitted linear models to test whether species’ reporting rates influence their thematic rankings from the ChatGPT analysis. This allowed us to confirm that the prominence of particular perceptual themes was not simply of function of how frequently species were observed (see Results).
We conducted exploratory factor analysis using the ‘factanal’ function in the nFactors R package (v2.4.1.1; Raiche and Magis, 2022) to identify thematic clusters underlying the eight perceptual themes. The number of factors to extract was determined through scree plot and parallel analyses. Factor analysis enabled us to examine how different themes relate to one another and whether they group into meaningful clusters to explain human-bird relationships. Themes with loadings >0.35, a suitable threshold given our large sample size (N = 1, 977; Hair et al., 1998), were interpreted as forming clusters of related responses, allowing us to determine whether species perceptions were shaped by independent themes (e.g., Appearance/Aesthetics or Behavioural Characteristics) or by more complex multivariate clusters of themes. Factor loadings were then used to position each species within multidimensional perception space, allowing us to visualise how species clustered based on the types of human responses they elicited. We assessed whether species showed synergistic patterns (where multiple perceptual clusters influence appeal simultaneously) or associative patterns (where perception was dominated by a single cluster) using a Spearman’s rank correlation. This analysis reveals the complexity of human-bird relationships and whether certain species generate more multifaceted human responses than others. While exploratory factor analyse can have limitations associated with assumptions of normality, sample size, or representativeness (Goudarzian, 2023), our large sample size and pre-data checks for correlation and normality assist in mitigating these effects.
3 Results
From our 1,977 online questionnaires completed across all four South African provinces, Gauteng received the greatest response (1,141 or 57.7%), followed by KwaZulu-Natal (426 or 21.5%), Western Cape (369 or 18.7%), and Free State (41 or 2.1%).
3.1 Demographics and birding knowledge
Respondent demographics varied by age, gender, race, and dwelling type (Supplementary Table S3). The largest age group represented was 61–70 years (20.7%), while the fewest responses came from individuals over 70 years (12.2%). Female respondents comprised 60.0% of the sample, males 38.9%, and agender/non-binary individuals 1.0%. Racial composition showed a strong bias, with 79.8% of respondents identifying as White, followed by Indian (7.9%), Black (6.7%), Other (2.5%), and Asian (0.4%). Housing types were diverse, with the majority (56.9%) living in stand-alone houses and the smallest proportion (0.3%) in informal housing (Supplementary Table S4).
Nearly all participants (98.2%) agreed that birds add value to their environment, and 64.6% reported engaging in some form of birding activity. These results indicate that while there is strong overall interest in birds within the sample, approximately 35.4% of respondents were non-birders from the general public.
3.2 Species-level likeability
We found variation in likeability scores across the 36 urban bird species (Table 1). Most bird species received generally positive ratings, but clear preferences emerged. Highly liked species included the Malachite Kingfisher (Alcedo cristata, Mean ± SE = 4.91 ± 0.02), Orange-breasted Sunbird (4.91 ± 0.02), and African Fish Eagle (Haliaeetus vocifer, 4.88 ± 0.02), all of which received mean scores above 4.80 on the 5-point scale. Conversely, species with notably lower likeability included the Common Myna (2.49 ± 0.03), Common Starling (2.96 ± 0.06), and Pied Crow (Corvus albus, 2.99 ± 0.03).
Table 1. Measures of the average likeability scores and standard errors received by study species across the research questionnaires in Gauteng (GT), Western Cape (WC), Kwa-Zulu Natal (KZN), or Free State (FS) with the respective sample sizes.
3.3 Perceptual themes
Ranking of the number of responses across the eight themes revealed clear patterns in how species were perceived (Figure 2). Appearance/Aesthetics emerged as the most frequent theme for the majority of species, indicating that physical attractiveness strongly influences urban bird appeal (Figure 2). Traits related to colouration, patterns, and distinctive features were consistently highlighted across species (Figure 2). Emotional Impact (Personal Connection) was also common, particularly for raptors, where respondents often described memorable or personal encounters with these less frequently encountered species (Figure 2).
Figure 2. Heat map showing the distribution of responses across perceptual categories for the 36 bird species included in the analysis. The colour scale represents the rank of each category for a given species, with darker shading indicating a higher frequency of responses per category. The first 11 species were assessed in at least two provincial questionnaires.
For some species, notably Hadeda Ibis, Rock Dove (Columba livia), and Common Myna, Negative/Neutral responses dominated (Figure 2). These reflected recognition of problematic traits in urban environments, with common descriptors including “loud”, “messy”, “invasive”, and “bully.” Cultural Significance/Symbolism appeared infrequently across most species, suggesting limited cultural associations with urban birds among our respondents. (Figure 2). An exception was the African Fish Eagle, where symbolic associations (e.g., strength, national significance) contributed substantially to its high appeal. (Figure 2).
The effect of reporting rate on thematic rank showed that reporting rate did not influence the rankings for most perceptual themes (P>0.01; Supplementary Table S6). Two exceptions were identified. Emotional Impact (Personal Connection) showed a negative association with reporting rate (P = 0.003), indicating that species reported more frequently were ranked lower for this theme. In contrast, Negative/Neutral responses showed a positive association with reporting rate (P = 0.007), with more frequently reported species ranked higher. Overall, these results suggest that thematic rankings are generally not driven by observation frequency alone, although certain themes may be shaped by additional factors, including personal experiences with particular species.
3.4 Dimensions of human-bird relationships
Exploratory factor analysis identified three underlying clusters that together explained 75.6% of variance in responses. Five themes loaded strongly (>0.35) onto single factors, while three themes (Behavioural Characteristics, Prevalence/Rarity, Emotional Impact) loaded onto multiple factors (Supplementary Table S7), highlighting their complex role in shaping species perception.
The first cluster, explaining 32.6% of variance, reflected “Perception and Environmental Impact”, comprising Behavioural Characteristics, Ecological Impact, Prevalence/Rarity, and Negative/Neutral responses. This cluster captured how people perceive birds based on their behaviours or influence on their urban environment and interactions with other wildlife. The second cluster (21.6% of variance) represented “Cultural and Emotional Resonance”, encompassing Cultural Significance/Symbolism, Vocalisation, and Emotional Impact. This cluster reflected deeper personal, symbolic, and affective connections to species. The third cluster (21.4% of variance) captured “Aesthetic and Behavioural Appeal”, comprising Behavioural Characteristics, Prevalence/Rarity, Appearance/Aesthetics, and Emotional Impact). This cluster was driven by immediate visual attractiveness and behaviour-based appreciation.
The multidimensional nature of human-bird relationships varied considerably among species (Figure 3). Some species generated complex, multifaceted responses across multiple perceptual clusters, while others were perceived primarily through a single dominant cluster (Figure 3). For example, Pied Crow (PC in Figure 3) showed complex perceptual patterns, with strong negative responses across both “Perception and Environmental Impact” and “Cultural and Emotional Resonance” clusters, indicating that multiple factors contribute to its poor appeal among urban residents (Figures 3a–c). In contrast, the African Fish Eagle (AFE in Figure 3) showed strong single-cluster appeal, being valued primarily for “Cultural and Emotional Resonance” while generating minimal interest across other clusters (Figure 3a). This pattern suggests that some species evoke unified human responses, while others generate more ambivalent or multifaceted reactions.
Figure 3. Multidimensional perceptual space of urban bird species based on questionnaire responses between the clusters (a) "Cultural and Emotional Resonance" and "Perception and Environmental Impact", (b) "Aesthetic and Behavioural Appeal" and "Perception and Environmental Impact", and (c) "Aesthetic and Behavioural Appeal" and "Cultural and Emotional Resonance". Species in blue shading represent synergies, where multiple perceptual clusters contribute simultaneously to species appeal. Species in red shading represent associations dominated more strongly by a single perceptual cluster. Species codes are shown for clarity; full species names are provided in Table 1.
4 Discussion
This study provides a comprehensive quantitative and qualitative assessment of human perceptions of urban bird species across South African urban contexts. Quantitatively, urban birds are generally well-liked, though variation exists among species. Coupling these results with large-scale qualitative analysis of approximately 71,000 open-ended responses shows that human-bird relationships are strongly influenced by aesthetic appeal and emotional connections, whereas species perceived to negatively impact urban environments or other wildlife tend to elicit unfavourable responses. Our analysis demonstrates that perceptions of urban birds are complex and can operate along multiple dimensions, with some species evoking multifaceted responses while others are viewed primarily through a single dominant lens.
4.1 Appeal of urban birds
The vast majority of survey respondents agreed that birds add value to their environment, and this positive attitude extended well beyond dedicated birders to include more than a third of respondents who did not regularly engage in birding activities. This aligns with previous work showing that birds are prominent, visible components of urban nature with generally high public appeal (Clergeau et al., 2001; Belaire et al., 2015). This suggests that charismatic and recognisable bird species have the potential to act as flagship taxa for communication, education, and urban nature initiatives aimed at diverse urban audiences (e.g., Ainsworth et al., 2018; Hegetschweiler et al., 2022). Education and awareness programmes that can leverage everyday encounters (e.g., bird feeders, identification games, neighbourhood biodiversity challenges) could support engagement and well-being benefits (Cox and Gaston, 2016; Hedblom et al., 2017) while broadening participation in citizen science initiatives such as ABAP and eBird (Sullivan et al., 2009; Brooks et al., 2022).
Despite the overall positive sentiment, species-level variation was evident in how urban birds were perceived. Variation in appeal is common among avian taxa (e.g., Lišková and Frynta, 2013; Tisdell et al., 2015), with numerous factors including species visibility, recognisability, physical traits, and behaviours influencing how individual species are perceived and ultimately valued (Cox and Gaston, 2015; Collins et al., 2021; Andrade et al., 2022; Champness et al., 2023). While most species were generally positively perceived, three species emerged as predominantly disliked by respondents, and were the Common Myna, Common Starling, and Pied Crow (discussed below). This species-level variation highlights the importance of recognising individual species when assessing human-wildlife relationships in urban contexts.
4.2 Human-bird relationships
Eight themes were identified to explain species appeal, with Appearance/Aesthetics emerging as the most frequent theme for the majority of species. The highest-rated species from our study i.e., the Malachite Kingfisher, Orange-breasted Sunbird, and African Fish Eagle, were primarily valued for Appearance/Aesthetics, with responses highlighting bright colouration and distinctive features. This supports general patterns in the literature where positive species perception is influenced by trait-based factors such as body size, colouration, and unique patterns, rather than species abundance (Collins et al., 2021; Andrade et al., 2022). However, these aesthetic patterns cannot be applied universally across our study species.
Many highly rated bird species were notably raptors (e.g., Crowned Eagle (Stephanoaetus coronatus), Black Sparrowhawk (Accipiter melanoleucus)), which are characteristically larger and have simpler plumage colours. Sentiment analysis revealed that appeal for raptor species was primarily driven by Emotional Impact (Personal Connection). These emotional connections suggest that charismatic species elicit stronger personal responses, as previously demonstrated for birds of prey (Donázar et al., 2016). The desire to conserve declining raptor populations may further enhance appeal for these species (Perry et al., 2020; Champness et al., 2023).
The strong negative responses to Common Myna, Common Starling, and Pied Crow were primarily associated with aggressive behaviours towards other species. Common Myna and Common Starling are recognised as non-native species with strong competitive abilities in urban landscapes, often thriving at the cost of native species (Martin-Albarracin et al., 2015). Whilst Pied Crow is indigenous to South Africa, its generalist nature and pest-like behaviour likely contribute to similar negative perceptions (Fincham et al., 2015; Cunningham et al., 2016).
However, these negative perception patterns do not apply to all non-native or pest-like species. House Sparrow (Passer domesticus), Rose-ringed Parakeet, and Hadeda Ibis received moderately positive responses, valued for Appearance/Aesthetics and Emotional Impact (Personal Connection) rather than being disliked for problematic behaviours. This reveals that public perception does not neatly align with ecological or conservation categorisations, and factors such as visibility, cultural familiarity, or conservation messaging framing can shape how problematic species are perceived (Verbrugge et al., 2013; Van Eeden et al., 2020; Díaz et al., 2022).
4.3 Multidimensional nature of human-bird relationships
Whilst appeal was most strongly associated with Appearance/Aesthetics, Behavioural Characteristics, and Emotional Impact (Personal Connection), these factors do not operate in isolation and are species-specific. Factor analysis revealed three underlying clusters explaining 75.7% of variance in responses and demonstrates that some species exhibit complex patterns whilst others show simpler relationships. Species positioned in multidimensional appeal zones were perceived through several clusters simultaneously, creating richer but potentially more complex human-bird relationships, whilst species in single-cluster zones elicited more straightforward responses dominated by a particular perceptual factor.
For example, the African Fish Eagle’s appeal is primarily driven by “Cultural and Emotional Resonance”, whilst Pied Crow demonstrated complex patterns with strong negative responses across multiple clusters. This supports that human-bird relationships can be complex, with different species attributes and individual experiences affecting how different birds are perceived. Similar complexity in human-nature relationships has been demonstrated in other contexts (Ament et al., 2017), where landscape effects and social preferences create complex ecosystem service provision patterns.
These distinct perceptual patterns have important implications for conservation messaging and urban wildlife management. Species evoking multifaceted or contradictory responses, e.g., Hadeda Ibis or Egyptian Goose (Alopochen aegyptiaca), scored moderately despite frequent negative comments about noise and disturbance, and may require more nuanced approaches that acknowledge multiple, sometimes conflicting, dimensions of human experience (Fisher et al., 2023). Understanding whether a species occupies a synergistic position (where multiple perceptual clusters influence appeal simultaneously) or an associative position (where perception is dominated by a single cluster) can inform the design of more effective engagement strategies tailored to specific contexts (conflict of interest species; e.g., Shackleton et al., 2019; Reynolds et al., 2020).
4.4 ChatGPT as a qualitative analysis tool
This study represents, to our knowledge, one of the first applications of large language models to human-wildlife perception research, demonstrating that LLMs can enable qualitative analysis at large scales. ChatGPT successfully categorised approximately 71,000 open-ended responses into eight thematic categories, enabling identification of perceptual patterns across 36 species that would be impractical using traditional qualitative methods.
Large language models are increasingly used for sentiment analysis, though their performance varies by application. Wang et al. (2023) found that whilst ChatGPT performs comparably to baseline models like BERT, it underperforms relative to domain-specific models developed for specific applications. Despite these limitations, ChatGPT provided a suitable general-purpose tool for processing the diverse, open-ended responses in this study. The application of ChatGPT to sentiment data extends beyond avian perception research, where similar methodologies could be applied to other urban wildlife taxa, attitudes towards urban green spaces, or perceptions of ecosystem services (e.g., Bjerke and Østdahl, 2004; Baur et al., 2014; Campbell-Arvai, 2019). In rapidly urbanising contexts, particularly in the Global South where research capacity may be limited and community perspectives are underrepresented, LLM-assisted analysis could facilitate large-scale qualitative research. However, careful validation, iterative prompt refinement, and manual quality control remain essential to ensure accuracy and address the limitations inherent in automated sentiment analysis (Hadi et al., 2024; Herrera-Poyatos et al., 2025; Rodriguez et al., 2025).
It is important to recognise that the use of ChatGPT 4.0 to conduct the full sentiment analysis has some limitations. As LLM’s are continually being updated, results generated using a specific version of ChatGPT may not be perfectly reproducible with future iterations or alternative AI systems. Additionally, like all AI-based text classifiers, ChatGPT may introduce biases shaped by the training data and methods used, which could influence how certain phrases or sentiments are interpreted. Our findings should be viewed in light these methodological constraints, although we did attempt to mitigate some of these biases through manual proofing and correction.
4.5 Limitations
Several limitations may affect the interpretation of these results. The demographic composition of the survey respondents presents an important constraint, with 79.8% of respondents identifying as White in a country where 81% of the population is Black African (SA Census, 2022). This bias, combined with generally higher socioeconomic groups (56.9% in stand-alone houses), suggests results do not reflect perceptions across South Africa’s diverse population. Questionnaire distribution through social media and birding networks, despite reaching 35.4% non-birders, likely skewed towards more engaged urban residents. Provincial response imbalance (57.7% from Gauteng vs. 2.1% from Free State) also creates uneven geographic representation.
Importantly, we measured perceptions rather than behaviours, and how positive perceptions translate into conservation support or wildlife-friendly actions remains unclear. There is evidence, however, that perception can motivate change. For example, the expansion of Pied Crow with urbanisation and the generally negative perception of these birds (Fincham et al., 2015; Cunningham et al., 2016) has led to programs aimed at reducing further spread. Furthermore, the urban focus excludes rural contexts where human-bird relationships may differ substantially (Williams et al., 2012, 2014), and the three-month survey period did not capture seasonal variation.
The low ranking of Cultural Significance/Symbolism warrants attention. Given the rich cultural significance of birds in many African societies (e.g., Msimanga, 2000; Gichuli and Terer, 2001; Muiruri and Maundu, 2012; Aticho et al., 2024), this pattern likely reflects the demographic biases in our responses rather than absence of cultural connections. This demographic skew limited the generalisability of our results towards the broader population of South Africa, as well as understanding a holistic view of the perception of birds. Hence, for further study, the traditional ecological knowledge, and indigenous perspectives, that were underrepresented in this study, should be developed, and explored to provide a more culturally inclusive perspectives on human-nature connections in Africa. This can be achieved through in-person surveys and community-based engagements that focus on the diverse communities that hold varied perceptions of nature that we were unable to access with our online questionnaires.
4.6 Conservation and management implications
Species that generate strong positive emotional responses, particularly raptors such as Crowned Eagle and Black Sparrowhawk, have clear potential as flagship species for conservation campaigns (Belaire et al., 2015; Donázar et al., 2016; Perry et al., 2020). As a contemporary illustration of this point, BirdLife South Africa’s public vote for “Bird of the Year 2026” selected the Endangered, near-endemic Black Harrier (Circus maurus) over the competing Botha’s Lark (Spizocorys fringillaris), a Critically Endangered South African grassland Endemic (BirdLife South Africa, 2025). Negative responses to certain species highlight the need for targeted communication. For example, negative perceptions of the Common Myna and Common Starling, which align with their ecological impacts (Lee and Barnard, 2016; Lee et al., 2017; Shivambu et al., 2020), and Mallard Duck (Anas platyhynchos) hybridisation with the native, Yellow-billed Duck (Anas undulata; Stephens et al., 2020), can be reinforced through conservation messaging to garner public support for management interventions.
The multidimensional nature of human-bird relationships suggests that species-specific approaches may be more effective than generic strategies. Importantly, public perception does not always neatly align with ecological priorities (e.g., Verbrugge et al., 2013; Díaz et al., 2022). Species causing genuine ecological harm may receive public support if they possess compensating aesthetic or emotional appeal (Verbrugge et al., 2013; Díaz et al., 2022; Ingaramo et al., 2024), while native species that are becoming more prevalent within urban spaces, e.g., Hadeda Ibis and Egyptian Goose, may be more disliked due to behavioural characteristics that may outweigh ecosystem service provisions (Little and Sutton, 2013; Singh and Downs, 2016). Beyond species-specific strategies, urban birds can support broader nature-connection and well-being initiatives. In African urban contexts, where access to natural spaces is often inequitable and development pressures are intensifying (Güneralp et al., 2017; Chamberlain et al., 2019; Reynolds et al., 2021), promoting positive relationships with accessible urban wildlife may be important for supporting cultural ecosystem services and improving quality of life.
4.7 Future research priorities
Future research must prioritise broader demographic engagement to understand how cultural backgrounds and socioeconomic factors influence human-bird relationships across diverse populations. Studies that are explicitly designed to capture diverse perspectives are urgently needed to capture the full spectrum of cultural values and traditional ecological knowledge associated with urban birds. Furthermore, examining whether positive perceptions translate into tangible conservation behaviours would strengthen connections between perception research and conservation outcomes (e.g., Hughes, 2013; Dean and Wilson, 2023). Finally, cross-cultural comparative research examining how perceptions of the same species may differ or not across different African cities with distinct ecological and socioeconomic contexts could further help us to understand human-wildlife relationships in rapidly urbanising regions.
Data availability statement
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.
Ethics statement
Non-medical human ethics was acquired for the study from the Wits University Non-Medical Research Committee. Our ethical clearance number = H23/06/29. The studies were conducted in accordance with the local legislation and institutional requirements. The participants provided their written informed consent to participate in this study. The animal study was approved by Wits University Non-Medical Research Committee. The study was conducted in accordance with the local legislation and institutional requirements.
Author contributions
SN: Writing – original draft, Writing – review & editing. CR: Writing – review & editing, Writing – original draft.
Funding
The author(s) declared that financial support was received for this work and/or its publication. This research was supported in part by the National Research Foundation (NRF) of South Africa (Reference Number: PMDS22061121329) and the Wits Post-graduate Merit Award (2023)(2024).
Acknowledgments
We extend our gratitude to all the individuals who responded to our research survey, aiding in the creation of our dataset. We are thankful to BirdLife South Africa, and its provincial constituents, as well as numerous individuals, organisations, including the University of the Witwatersrand, and birding clubs who assisted in the distribution and promotion of our research surveys.
Conflict of interest
The author(s) declared that this work was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
Generative AI statement
The author(s) declared that generative AI was used in the creation of this manuscript. ChatGPT 4.0 was used to conduct a full sentiment analysis of open-ended questionnaire data collected for the study. In the methods section, we provide a full, and in-depth, description of the use and training of ChatGPT, to conduct the sentiment analysis of our large (~71,000) open-ended response dataset. We provide an example of our full prompt in the appendix. ChatGPT 5.0 was used to help edit the grammar of the final manuscript but is not listed as an author of the manuscript, the content edited using the Generative AI has been checked for factual accuracy and plagiarism by the authors and the use of ChatGPT has been included in the methods section of the manuscript listing the name, model, version and source of the Generative AI, the initial and final prompts provided to the Generative AI have been included in the Supplementary Files.
Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.
Publisher’s note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
Supplementary material
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fbirs.2025.1726726/full#supplementary-material
References
Ainsworth G. B., Fitzsimons J. A., Weston M. A., and Garnett S. T. (2018). The culture of bird conservation: Australian stakeholder values regarding iconic, flagship and rare birds. Biodivers. Conserv. 27, 345–363. doi: 10.1007/s10531-017-1438-1
Ament J. M., Moore C. A., Herbst M., and Cumming G. S. (2017). Cultural ecosystem services in protected areas: Understanding bundles, trade-offs, and synergies. Conserv. Lett. 10, 440–450. doi: 10.1111/conl.12283
Andrade R., Larson K. L., Franklin J., Lerman S. B., Bateman H. L., and Warren P. S. (2022). Species traits explain public perceptions of human–bird interactions. Ecol. Appl. 32, e2676. doi: 10.1002/eap.2676
Anthropic (2025). Claude Sonnet 4 [Large language model]. Available online at: https://claude.ai.
Aouissi H. A., Petrişor A. I., Ababsa M., Boştenaru-Dan M., Tourki M., and Bouslama Z. (2021). Influence of land use on avian diversity in North African urban environments. Land. 10, 434. doi: 10.3390/land10040434
Aticho A., Beyene A., Zelelew S. A., Puok C., Gosler A. G., Stenseth N. C., et al. (2024). Birds as sacred ‘children of god’, beholders of spirit and symbols of twinhood: Human–bird interaction among the Nuer in western Ethiopia. People Nat. 6, 2418–2430. doi: 10.1002/pan3.10719
Awoyemi A. G., Ibáñez-Rueda N., Guardiola J., and Ibáñez-Álamo J. D. (2024). Human-nature interactions in the Afrotropics: Experiential and cognitive connections among urban residents in southern Nigeria. Ecol. Econ 218, 108105. doi: 10.1016/j.ecolecon.2024.108105
Ayeni A. O., Aborisade A. G., Onuminya T. O., Soneye A. S. O., and Ogundipe O. T. (2023). Urban development in Africa and impact on biodiversity. Curr. Landscape Ecol. Rep. 8, 73–89. doi: 10.1007/s40823-023-00086-x
Baur J. W., Tynon J. F., Ries P., and Rosenberger R. S. (2014). Urban parks and attitudes about ecosystem services: Does park use matter? J. Park Recreation Administration. 32, 19–34.
Belaire J. A., Westphal L. M., Whelan C. J., and Minor E. S. (2015). Urban residents’ perceptions of birds in the neighborhood: Biodiversity, cultural ecosystem services, and disservices. Condor: Ornithol. Appl. 117, 192–202. doi: 10.1650/condor-14-128.1
BirdLife South Africa (2025). Vote for the 2026 Bird of the Year! BirdLife South Africa. Available online at: https://www.birdlife.org.za/vote-boty2026/ (Accessed October 8, 2025).
Bjerke T. and Østdahl T. (2004). Animal-related attitudes and activities in an urban population. Anthrozoös. 17, 109–129. doi: 10.2752/089279304786991783
Boerema A., Rebelo A. J., Bodi M. B., Esler K. J., and Meire P. (2017). Are ecosystem services adequately quantified? J. Appl. Ecol. 54, 358–370. doi: 10.1111/1365-2664.12696
Brooks M., Rose S., Altwegg R., Lee A. T., Nel H., Ottosson U., et al. (2022). The African Bird Atlas Project: A description of the project and BirdMap data-collection protocol. Ostrich. 93, 223–232. doi: 10.2989/00306525.2022.2125097
Buttrick N. (2024). Studying large language models as compression algorithms for human culture. Trends Cogn. Sci. 28, 187–189. doi: 10.1016/j.tics.2024.01.001
Campbell-Arvai V. (2019). Engaging urban nature: Improving our understanding of public perceptions of the role of biodiversity in cities. Urban Ecosyst. 22, 409–423. doi: 10.1007/s11252-018-0821-3
Chamberlain D. E., Henry D. A., Reynolds C., Caprio E., and Amar A. (2019). The relationship between wealth and biodiversity: A test of the Luxury Effect on bird species richness in the developing world. Global Change Biol. 25, 3045–3055. doi: 10.1111/gcb.14682
Champness B. S., Fitzsimons J. A., Kendal D., and Palmer G. C. (2023). Perceptions of birds by urban residents in an Australian regional city and implications for conservation. Birds. 4, 262–276. doi: 10.3390/birds4030022
Chan K. M., Guerry A. D., Balvanera P., Klain S., Satterfield T., Basurto X., et al. (2012). Where are cultural and social in ecosystem services? A framework for constructive engagement. BioScience. 62, 744–756. doi: 10.1525/bio.2012.62.8.7
Clergeau P., Mennechez G., Sauvage A., and Lemoine A. (2001). “Human perception and appreciation of birds: A motivation for wildlife conservation in urban environments of France,” in Avian Ecology and Conservation in an Urbanizing World. Eds. Marzluff J., Bowman R., and Donnelly R. (Springer, USA), 69–88. doi: 10.1007/978-1-4615-1531-9_4
Collins L., Paton G. D., and Gagné S. A. (2021). Testing the likeable, therefore abundant hypothesis: Bird species likeability by urban residents varies significantly with species traits. Land. 10, 487. doi: 10.3390/land10050487
Collinson M. A., Tollman S. M., and Kahn K. (2007). Migration, settlement change and health in post-apartheid South Africa: Triangulating health and demographic surveillance with national census data. Scand. J. Public Health 35, 77–84. doi: 10.1080/14034950701356401
Cox D. T. and Gaston K. J. (2015). Likeability of garden birds: Importance of species knowledge & richness in connecting people to nature. PloS One 10, e0141505. doi: 10.1371/journal.pone.0141505
Cox D. T. and Gaston K. J. (2016). Urban bird feeding: Connecting people with nature. PloS One. 11, e0158717. doi: 10.1371/journal.pone.0158717
Cunningham S. J., Madden C. F., Barnard P., and Amar A. (2016). Electric crows: Powerlines, climate change and the emergence of a native invader. Diversity Distrib. 22, 17–29. doi: 10.1111/ddi.12381
Dean A. J. and Wilson K. A. (2023). Relationships between hope, optimism, and conservation engagement. Conserv. Biol. 37, e14009. doi: 10.1111/cobi.14020
Devlin J., Chang M. W., Lee K., and Toutanova L. K. (2019). “BERT: Pre-training of deep bidirectional transformers for language understanding,” in Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Vol. 1. 4171–4182. doi: 10.18653/v1/N19-1423
Díaz R. A., Sevillano V., and Cassini M. H. (2022). Do people care about the origin of wildlife? The role of social stereotypes on public preference for exotic animals. Animals 12, 2160. doi: 10.3390/ani12172160
Donázar J. A., Cortés-Avizanda A., Fargallo J. A., Margalida A., Moleón M., Morales-Reyes Z., et al. (2016). Roles of raptors in a changing world: From flagships to providers of key ecosystem services. Ardeola. 63, 181–234. doi: 10.13157/arla.63.1.2016.rp8
Egoh B., Reyers B., Rouget M., Bode M., and Richardson D. M. (2009). Spatial congruence between biodiversity and ecosystem services in South Africa. Biol. Conserv. 142, 553–562. doi: 10.1016/j.biocon.2008.11.009
Elmqvist T., Zipperer W. C., and Güneralp B. (2015). “Urbanization, habitat loss and biodiversity decline: Solution pathways to break the cycle,” in The Routledge Handbook of Urbanization and Global Environmental Change. Eds. Seto K., Solecki W., and Griffith C. (Routledge, London), 163–175. doi: 10.4324/9781315849256
Finch J. M. and Meadows M. E. (2018). “South African biomes and their changes over time,” in The geography of South Africa: Contemporary changes and new directions. Eds. Knight J. and Rogerson C. (Springer International Publishing, Cham), 57–69. doi: 10.1007/978-3-319-94974-1_7
Fincham J. E., Visagie R., Underhill L. G., Brooks M., and Markus M. B. (2015). The impacts of the Pied Crow Corvus albus on other species need to be determined. Biodivers. Observ. 5, 232–239.
Fisher J. C., Dallimer M., Irvine K. N., Aizlewood S. G., Austen G. E., Fish R. D., et al. (2023). Human well-being responses to species’ traits. Nat. Sustain. 6, 1219–1227. doi: 10.1038/s41893-023-01151-3
Gaston K. J., Cox D. T., Canavelli S. B., García D., Hughes B., Maas B., et al. (2018). Population abundance and ecosystem service provision: The case of birds. BioScience 68, 264–272. doi: 10.1093/biosci/biy005
Gichuli N. and Terer T. (2001). Significance of indigenous knowledge and values of birds in promoting biodiversity conservation in Kenya. Significance. 9, 153–157.
Goudarzian A. H. (2023). Challenges and recommendations of exploratory and confirmatory factor analysis: A narrative review from a nursing perspective. J. Nurs. Rep. Clin. Pract. 1, 133–137. doi: 10.32598/JNRCP.23.44
Güneralp B., Lwasa S., Masundire H., Parnell S., and Seto K. C. (2017). Urbanization in Africa: Challenges and opportunities for conservation. Environ. Res. Lett. 13, 015002. doi: 10.1088/1748-9326/aa94fe
Hadi M. U., Al-Tashi Q., Qureshi R., Shah A., Muneer A., Irfan M., et al. (2024). LLMs: A comprehensive survey of applications, challenges, datasets, models, limitations, and future prospects. TechRxiv, 439–464. doi: 10.4018/979-8-3693-9311-6.ch012
Hair J. F. Jr., Black W. C., Babin B. J., and Anderson R. E. (1998). Multivariate Data Analysis with Readings (Englewood Cliffs, NJ: Prentice-Hall).
Hedblom M., Knez I., and Gunnarsson B. (2017). Bird diversity improves the well-being of city residents. Ecol. Conserv. Birds Urban Environ., 287–306. doi: 10.1007/978-3-319-43314-1_15
Hegetschweiler K. T., Wartmann F. M., Dubernet I., Fischer C., and Hunziker M. (2022). Urban forest usage and perception of ecosystem services: A comparison between teenagers and adults. Urban Forest. Urban Green. 74, 127624. doi: 10.1016/j.ufug.2022.127624
Herrera-Poyatos D., Peláez-González C., Zuheros C., Herrera-Poyatos A., Tejedor V., Herrera F., et al. (2025). An overview of model uncertainty and variability in LLM-based sentiment analysis. Challenges, mitigation strategies and the role of explainability. arXiv preprint. doi: 10.48550/arXiv.2504.04462
Hughes K. (2013). Measuring the impact of viewing wildlife: Do positive intentions equate to long-term changes in conservation behaviour? J. Sustain. Tourism. 21, 42–59. doi: 10.1080/09669582.2012.681788
Ingaramo M., Cammerino A. R. B., Rizzi V., Gioiosa M., and Monteleone M. (2024). Birds as cultural ambassadors: Bridging ecosystem services and biodiversity conservation in wetland planning. Sustainability 16, 10286. doi: 10.3390/su162310286
Ivanova I. M. and Symes C. T. (2019). Invasion of Psittacula krameri in Gauteng, South Africa: Are other birds impacted? Biodivers. Conserv. 28, 3633–3656. doi: 10.1007/s10531-019-01841-5
Kopij G. (2015). Avian diversity in an urbanized South African grassland. Zool. Ecol. 25, 87–100. doi: 10.1080/21658005.2015.1013744
Lee A. T., Altwegg R. E. S., and Barnard P. (2017). Estimating conservation metrics from atlas data: The case of southern African endemic birds. Bird Conserv. Int. 27, 323–336. doi: 10.1017/S0959270916000307
Lee A. T. and Barnard P. (2016). Endemic birds of the Fynbos biome: A conservation assessment and impacts of climate change. Bird Conserv. Int. 26, 52–68. doi: 10.1017/S0959270914000537
Lee A. T. K., Ottosson U., Jackson C., Shema S., and Reynolds C. (2021). Urban areas have lower species richness, but maintain functional diversity: Insights from the African Bird Atlas Project. Ostrich. 92, 1–15. doi: 10.2989/00306525.2021.1902876
Lišková S. and Frynta D. (2013). What determines bird beauty in human eyes? Anthrozoös 26, 27–41. doi: 10.2752/175303713x13534238631399
Little R. M. and Sutton J. L. (2013). Perceptions towards Egyptian geese at the Steenberg Golf Estate, Cape Town, South Africa. Ostrich 84, 85–87. doi: 10.2989/00306525.2013.772079
Maphisa D. H., Smit-Robinson H., Underhill L. G., and Altwegg R. (2016). Drivers of bird species richness within moist high-altitude Grasslands in Eastern South Africa. PloS One. 11, e0162609. doi: 10.1371/journal.pone.0162609
Martin-Albarracin V. L., Amico G. C., Simberloff D., and Nuñez M. A. (2015). Impact of non-native birds on native ecosystems: A global analysis. PloS One 10, e0143070. doi: 10.1371/journal.pone.0143070
McKinney M. L. (2006). Urbanization as a major cause of biotic homogenization. Biol. Conserv. 127, 247–260. doi: 10.1016/j.biocon.2005.09.005
McPherson S. C., Sumasgutner P., and Downs C. T. (2021). South African raptors in urban landscapes: A review. Ostrich 92, 41–57. doi: 10.2989/00306525.2021.1900942
Medhat W., Hassan A., and Korashy H. (2014). Sentiment analysis algorithms and applications: A survey. Ain Shams Eng. J. 5, 1093–1113. doi: 10.1016/j.asej.2014.04.011
Methorst J., Rehdanz K., Mueller T., Hansjürgens B., Bonn A., and Böhning-Gaese K. (2021). The importance of species diversity for human well-being in Europe. Ecol. Econ. 181, 106917. doi: 10.1016/j.ecolecon.2020.106917
Millennium Ecosystem Assessment (MEA) (2005). Ecosystems and human well-being: synthesis (Washington, DC: Island Press).
Mino T. (2013). “Intergroup peace and interpersonal violence in South Africa,” in The Economic Roots of Conflict and Cooperation in Africa. Eds. Ascher W. and Mirovitskaya N. (Palgrave Macmillan US, New York), 221–242. doi: 10.1057/9781137356796.0013
Msimanga A. (2000). The role of birds in the culture of the Ndebele people of Zimbabwe. Ostrich. 71, 22–24. doi: 10.1080/00306525.2000.9639858
Muiruri M. N. and Maundu P. (2012). “Birds, people and conservation in Kenya,” in Ethno-ornithology: birds, indigenous people, culture and society. Eds. Tidemann S. and Gosler A. (Earthscan/James & James, London), 279–289. doi: 10.4324/9781849774758
O’Neill A. (2023). South Africa - urbanization 2011-2021 (Statista). Available online at: https://www.statista.com/statistics/455931/urbanization-in-south-africa/:~:text=Urbanization%20in%20South%20Africa%202021&text=In%202021%2C%20over%2067.85%20percent,total%20population%20of%20a%20country (Accessed February 23, 2023).
OpenAI (2024). ChatGPT (June 2024 version). Available online at: https://chat.openai.com/ (Accessed June 2024).
Peacock D. S., van Rensburg J., and Robertson M. P. (2007). The distribution and spread of the invasive alien common myna, Acridotheres tristis L.(Aves: Sturnidae), in southern Africa. South Afr. J. Sci. 103, 465–473.
Perry G., Boal C., Verble R., and Wallace M. (2020). ““Good” and “bad” urban wildlife,” in Problematic wildlife II: New Conservation and Management Challenges in the Human-wildlife Interactions. Eds. Angelici F. and Rossi L. (Springer, Cham), 141–170. doi: 10.1007/978-3-030-42335-3_5
Plieninger T., Dijks S., Oteros-Rozas E., and Bieling C. (2013). Assessing, mapping, and quantifying cultural ecosystem services at community level. Land Use Policy. 33, 118–129. doi: 10.1016/j.landusepol.2012.12.013
Raiche G. and Magis D. (2022). nFactors: Parallel analysis and other non-graphical solutions to the Cattell scree test (2.4. 1.1). (Repository CRAN) 2, 1–58.
R Core Team (2024). RStudio: Integrated Development Environment for R (Boston, MA: Posit Software, PBC). Available online at: http://www.posit.co/ (Accessed May 01, 2024).
Reynolds C., Byrne M. J., Chamberlain D. E., Howes C. G., Seymour C. L., Sumasgutner P., et al. (2021). “Urban animal diversity in the Global South,” in Urban ecology in the global south. Eds. Shackleton C. M., Cilliers S. S., Davoren E., and du Toit M. J. (Springer International Publishing, Cham), 169–202. doi: 10.1007/978-3-030-67650-6_7
Reynolds C., Venter N., Cowie B. W., Marlin D., Mayonde S., Tocco C., et al. (2020). Mapping the socio-ecological impacts of invasive plants in South Africa: Are poorer households with high ecosystem service use most at risk? Ecosyst Serv. 42, 101075. doi: 10.1016/j.ecoser.2020.101075
Riechers M., Barkmann J., and Tscharntke T. (2016). Perceptions of cultural ecosystem services from urban green. Ecosyst Serv. 17, 33–39. doi: 10.1016/j.ecoser.2015.11.007
Rodriguez J., Sanan O., Vizarreta-Luna G., and Conrad S. A. (2025). Text chunking for document classification for urban system management using large language models. arXiv preprint. doi: 10.48550/arXiv.2504.00274
Rutherford M. C., Mucina L., and Powrie L. W. (2006). “Biomes and bioregions of southern Africa,” in The Vegetation of South Africa, Lesotho, and Swaziland, vol. 19. (South African National Biodiversity Institute, Pretoria), 30–51.
SA Census (2022). Statistics South Africa. Available online at: https://census.statssa.gov.za//statsbytheme (Accessed January-February 2023).
Seto K. C., Fragkias M., Güneralp B., and Reilly M. K. (2011). A meta-analysis of global urban land expansion. PloS One. 6, e23777. doi: 10.1371/journal.pone.0023777
Shackleton R. T., Richardson D. M., Shackleton C. M., Bennett B., Crowley S. L., Dehnen-Schmutz K., et al. (2019). Explaining people’s perceptions of invasive alien species: A conceptual framework. J. Environ. Manage. 229, 10–26. doi: 10.1016/j.jenvman.2018.04.045
Shivambu T. C., Shivambu N., and Downs C. T. (2020). Impact assessment of seven alien invasive bird species already introduced to South Africa. Biol. Invasions. 22, 1829–1847. doi: 10.1007/s10530-020-02221-9
Singh P. and Downs C. T. (2016). Hadedas in the hood: Hadeda Ibis activity in suburban neighbourhoods of Pietermaritzburg, KwaZulu-Natal, South Africa. Urban Ecosyst. 19, 1283–1293. doi: 10.1007/s11252-016-0540-6
Smith Y. C. E., Maseko M. S., Sosibo M., Dlamini P. V., Gumede S. T., Ngcobo S. P., et al. (2021). Indigenous knowledge of South African bird and rangeland ecology is effective for informing conservation science. J. Environ. Manage. 284, 112041. doi: 10.1016/j.jenvman.2021.112041
Statistics SA (2023). Statistics South Africa. Available online at: http://www.statssa.gov.za/ (Accessed January 31, 2023).
Stephens K., Measey J., Reynolds C., and Le Roux J. J. (2020). Occurrence and extent of hybridisation between the invasive Mallard Duck and native Yellow-billed Duck in South Africa. Biol. Invasions. 22, 693–707. doi: 10.1007/s10530-019-02122-6
Sullivan B. L., Wood C. L., Iliff M. J., Bonney R. E., Fink D., and Kelling S. (2009). eBird: A citizen-based bird observation network in the biological sciences. Biol. Conserv. 142, 2282–2292. doi: 10.1016/j.biocon.2009.05.006
Swilling M. (2010). Sustainability, poverty and municipal services: The case of Cape Town, South Africa. Sustain. Dev. 18, 194–201. doi: 10.1002/sd.489
Tisdell C., Wilson C., and Nantha H. S. (2005). Association of public support for survival of wildlife species with their likeability. Anthrozoös. 18, 160–174. doi: 10.2752/089279305785594216
Todes A., Kok P., Wentzel M., Van Zyl J., and Cross C. (2010). Contemporary South African urbanization dynamics. Urban Forum. 21, 331–348. doi: 10.1007/s12132-010-9094-5
Van Eeden L. M., Newsome T. M., Crowther M. S., Dickman C. R., and Bruskotter J. (2020). Diverse public perceptions of species’ status and management align with conflicting conservation frameworks. Biol. Conserv. 242, 108416. doi: 10.1016/j.biocon.2020.108416
van Rensburg B. J., Peacock D. S., and Robertson M. P. (2009). Biotic homogenization and alien bird species along an urban gradient in South Africa. Landscape Urban Plann. 92, 233–241. doi: 10.1016/j.landurbplan.2009.05.002
Van Wilgen B. W., Measey J., Richardson D. M., Wilson J. R., and Zengeya T. A. (2020). Biological invasions in South Africa (Heidelburg, South Africa: Springer Nature). doi: 10.1007/978-3-030-32394-3
Verbrugge L. N., Van den Born R. J., and Lenders H. R. (2013). Exploring public perception of non-native species from a visions of nature perspective. Environ. Manage. 52, 1562–1573. doi: 10.1007/s00267-013-0170-1
Wang Z., Xie Q., Ding Z., Feng Y., and Xia R. (2023). Is ChatGPT a good sentiment analyzer? A preliminary study. arXiv preprint. doi: 10.48550/arXiv.2304.04339
Wickham H., Chang W., and Wickham M. H. (2016). Package ‘ggplot2’. Create elegant data visualisations using the grammar of graphics. Version, (Cham: Springer International Publishing) Vol. 2. 1–189.
Williams V. L., Cunningham A. B., Bruyns R. K., and Kemp A. C. (2012). “Birds of a feather: Quantitative assessments of the diversity and levels of threat to birds used in African traditional medicine,” in Animals in Traditional Folk Medicine: Implications for Conservation. Eds. Alves R. and Rosa I. (Springer, Berlin, Heidelberg), 383–420. doi: 10.1007/978-3-642-29026-8_18
Keywords: artificial intelligence (AI), ChatGPT, citizen science, Global South, human-wildlife relationships, urban biodiversity, urbanisation
Citation: Naidoo SK and Reynolds C (2026) Large language models enable large-scale analysis of human-bird relationships in South African cities. Front. Bird Sci. 4:1726726. doi: 10.3389/fbirs.2025.1726726
Received: 16 October 2025; Accepted: 03 December 2025; Revised: 27 November 2025;
Published: 15 January 2026.
Edited by:
Ulf Ottosson, Halmstad University, SwedenReviewed by:
João Carlos Pena, São Paulo State University, BrazilBello Danmallam, A.P. Leventis Ornithological Research Institute, Nigeria
Copyright © 2026 Naidoo and Reynolds. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
*Correspondence: Sage K. Naidoo, c2FnZS5rLm5haWRvb0BnbWFpbC5jb20=