Diabetes mellitus (DM) is an escalating global health problem that affects millions of children and adolescents in the 21st century. Diabetes is a chronic medical condition marked by elevated levels of blood glucose (or blood sugar), which leads over time to serious damage to the heart, blood vessels, eyes, kidneys and nerves.¹ It is caused by an absolute insulin deficiency, such as type 1, or by a relative deficiency, such as type 2. Among the different types of diabetes, type 1 diabetes (T1D) in children has become one of the most significant global public health concerns, imposing a growing burden on families and countries. Xie et al.² affirm that the onset of T1D is an escalating global health concern among children, adolescents and young adults, particularly in countries with a low-to-middle and middle Social Development Index. Indeed, the prevalence of T1D in children is particularly significant in low-income African countries, where the condition is often under-recognised, undetected and inadequately treated.³ It is estimated that approximately 1.2 million children and adolescents around the world suffer from type 1 and type 2 diabetes, with the majority living in developing countries.^4,5 Despite this, a growing number of diabetes mellitus cases remain undiagnosed, particularly in resource-constrained healthcare settings with limited access to medical services, resources and early detection services.

Given the context, children and adolescents are regarded as individuals aged between 0 years and 19 years, in accordance with global and South African paediatric classification.⁶ There is, however, a wide range of cognitive maturity, health literacy and decision-making capability among this age group. It is unlikely that young children can engage with generative artificial intelligence (AI) tools independently, so they will need assistance or guidance from parents, caregivers or healthcare professionals. Despite their variable ability to evaluate health information, older adolescents may interact with these tools more independently. Nonetheless, ChatGPT can directly support adolescents and indirectly support younger children through parents and caregivers. Tran et al.⁷ emphasised that although ChatGPT has the potential to provide accessible health information, its application does not guarantee age-appropriate content without the aid of guided input or system design. This underscores the need for age-sensitive adaptations such as the use of simplified language and the implementation of safeguards. Furthermore, there is a need for legal and ethical considerations regarding minors’ access to digital platforms in South Africa, particularly in relation to consent and data privacy.

In the African context, the incidence and prevalence of T1D are rising, with 1 211 900 children aged 0–19 years diagnosed with the condition, and an additional 149 500 new cases identified annually.⁸ Prior research has highlighted geographic variations in the incidence of T1D among children across different regions. For example, the incidence of T1D among children is reported as 2.4 per 100 000 in Ethiopia⁹ and 1.5 per 100 000 in Tanzania,¹⁰ while higher rates have been observed in countries such as Algeria (4.4 per 100 000) and Morocco (20 per 100 000).⁴ Similarly, incidence rates vary in other regions, including 25 per 100 000 in Saudi Arabia and 3.75 per 100 000 in Egypt,¹¹ as well as 2.7 per 100 000 in Rwanda and 11.2 per 100 000 in Eritrea.⁸ These variations reflect differences in healthcare systems, diagnostic capacity and socioeconomic conditions across countries.

In South Africa, the prevalence of diabetes is on the rise among children and adolescents, particularly T2D, which has been linked to lifestyle and socioeconomic factors.^12,13,14 In contrast, T1D is reported at a lower incidence rate, with Dhada and Blackbeard⁴ estimating that the incidence of T1D among children in South Africa aged 0–14 years is 0.8 per 100 000. These incidence rates of type T1D appear lower than those reported in other African countries. However, this may be because of an underdiagnosis of T1D, limited surveillance systems, and disparities in healthcare access. As a result of this, caution should be exercised when making direct comparisons between countries, as differences in the methods for collecting and reporting data may affect the results. Despite these limitations, Lesage et al.¹⁵ stress that there have been growing concerns about the rising prevalence of T1D and T2D in South Africa since the onset of the coronavirus disease 2019 (COVID-19) outbreak. As a result of these trends, there is still a need to improve early detection, disease management and support systems for children and adolescents with diabetes. According to Thomas et al.,¹⁶ South African children and adolescents living with diabetes face several challenges that make it difficult for them to manage their disease effectively. These challenges include a lack of access to healthcare services, a lack of social and family support, difficulty monitoring insulin levels and adhering to treatments, and psychological difficulties like anxiety and frustration. As a result, this affects long-term glycaemic control and health outcomes.

Abrahams et al.¹⁷ argue that the South African public healthcare system is under severe strain and is unable to provide support and necessary care for young people living with diabetes. Public sector hospitals are operating under substantial pressure with outdated infrastructure, limited resources, legacy systems and a shortage of paediatric endocrinologists, all of which hinder the provision of quality care, personalised education and social support for young patients living with diabetes. Moore et al.¹⁸ ascertain that children and young adolescents with T1D from disadvantaged backgrounds or marginalised communities of South Africa struggle with managing their diabetes because of inadequate health services, inaccessible medication and poor digital literacy. Nonetheless, these contextual barriers result in higher risks of complications, hospital readmissions and poor health and treatment outcomes among young people living with T1D. In response to these challenges, AI in healthcare emerged as a solution that has the potential to reshape how patients with diabetes are diagnosed, treated and monitored. Tran et al.⁷ underscore that AI can support clinical consultations by improving patient education, simplifying complex medical information and improving clinician communication with patients and caregivers. Mondal and Naskar¹⁹ argue that generative AI tools such as ChatGPT present a promising solution to supplement overburdened and vulnerable healthcare services by providing accessible information, diabetes education and self-management support for children and adolescents. Daungsupawong and Wiwanitkit²⁰ and Jairoun et al.²¹ attest that ChatGPT offers a promising path towards enhancing diabetes care through early detection, personalised treatment and medical triage support for children and adolescents with diabetes, aiding healthcare providers and patients in making well-informed decisions.

Despite the growing presence of AI in healthcare, there is limited understanding of how healthcare professionals perceive the applicability of tools such as ChatGPT in the management of chronic conditions such as diabetes among children and adolescents in South Africa. Although several studies in South Africa have explored AI in medical diagnosis²² and AI-based diabetes prediction,²³ there is a noticeable gap in the literature on ChatGPT for diabetes management. Therefore, this qualitative study sought to explore the perceptions of healthcare professionals and academics on the use of ChatGPT in supporting diabetes management among children and adolescents. The study contributes to the body of knowledge by providing insight into how ChatGPT, as a representative AI tool, may support clinical interactions and patient care, rather than examining direct use by patients themselves. This article aims to answer the following research questions:

The study comprised 16 participants representing a diverse group of health professionals and academics with knowledge and experience in paediatric diabetes and health technology. The participants comprised nine females and seven males, holding postgraduate diplomas, master’s or doctoral degrees, with professional experience ranging from 10 years to 25 years. They were affiliated with various academic institutions and hospitals across South Africa. Table 1 depicts demographic characteristics of the participants.

The findings of the thematic analysis resulted in four major themes and 13 categories as illustrated in Table 2.

The results of this study are arranged according to four themes.

The findings revealed the potential benefits of ChatGPT as a means for enhancing communication during consultations and as a source of supplementary information based on the perspectives of clinicians and academics. These findings align with Dey²⁴ who emphasised the ability of ChatGPT to offer continuous health education. Participants believed that ChatGPT could help families better understand diabetes, assist with blood sugar control and reduce the burden on healthcare providers. Moreover, there was consensus that ChatGPT could advance diabetes education by making it more engaging and accessible, provided that government support facilitates its integration.

However, participants expressed concern about the reliability and implications of AI-generated information. This concern is supported by Al-Anezi²⁵ who cautioned against overreliance on AI because of its inability to offer human-like emotional understanding. Accuracy and reliability were also major concerns. Some participants feared that incorrect or misleading information could lead to harmful health decisions, particularly for chronic conditions like diabetes. These concerns align with Chatelan et al.²⁶ who stressed that ChatGPT may generate inaccurate and unreliable treatment recommendations. However, other scholars, such as Huang et al.²⁷ and Sng et al.²⁸, argue that ChatGPT can provide reliable health information when properly validated. In this digital era, many AI-driven healthcare solutions have addressed these concerns through incorporating a variety of expertise, including medical consultants, patient educators and health counsellors to monitor and validate that the health information provided to patients is accurate. It is evident from dedicated medical applications that their content has been clinically reviewed and aligned with established guidelines. Nevertheless, these structured validation processes may not be integrated into large language models such as ChatGPT, which are based on probabilistic text generation rather than real-time clinical oversight. Therefore, using large language models in healthcare requires reliable safeguards, validation mechanisms and professional oversight.

Linguistic accessibility emerged as another challenge. Many participants expressed concerns about the reliance of ChatGPT on the English language, which could exclude non-English-speaking users, particularly in rural South African communities. This linguistic barrier could hinder equitable access to healthcare support. Security and data privacy were also critical concerns. Although data privacy concerns are often associated with AI technologies, the risks in this context are related to how clinicians interact with tools such as ChatGPT. For instance, the entry of identifiable patient information into AI platforms could raise concerns regarding confidentiality. However, such risks can be minimised when used appropriately as a tool for general information without sharing sensitive patient information. Participants further stressed the importance of adhering to ethical guidelines and professional standards when integrating AI into clinical practice. Beyond the South African context, international literature has demonstrated the ethical concerns surrounding the use of ChatGPT to support diabetes care and other chronic diseases. A series of studies conducted in Asia, North America and Europe indicated several ethical and legal concerns associated with generative AI tools in DM, including data privacy and security risks, deficiencies in transparency and interpretability, educational and language disparities, inaccurate responses, responsibility and copyright concerns and the risk of misinformation.^29,30,31

The ethical concerns mentioned by participants in this study are consistent with those raised by other scholars, suggesting that ethical issues related to trust, transparency and responsible use of ChatGPT are not unique to South Africa, but are shared across diverse healthcare systems. Nevertheless, in resource-constrained settings such as South Africa, these ethical risks can be exacerbated by infrastructural limitations, linguistic diversity and disparities in digital literacy, underscoring the need for regulatory approaches that are sensitive and internationally informed. Participants expressed scepticism about the ability of ChatGPT to safeguard patient data, fearing potential cybersecurity breaches and a lack of transparency in data management. These concerns align with Alawida et al.,³² who warned that AI-driven health platforms could expose sensitive medical data to cyber threats and fraud.

Several key factors were identified as influencing the acceptance of ChatGPT in diabetes management. Participants emphasised that user-friendliness and ease of navigation are crucial, particularly for individuals managing chronic conditions. This aligns with Zhang and Zhao,³³ who found that perceived usefulness and ease of use significantly influence AI adoption in healthcare. Trust in ChatGPT was seen as contingent on its ability to provide accurate, verified and cross-referenced information. These findings are consistent with Ittarat et al.³⁴ and Triplett,³⁵ who found that familiarity with AI tools enhances trust and acceptance. Furthermore, participants indicated that the more users, particularly parents of children with diabetes, understand how to use ChatGPT for tasks like blood sugar monitoring and healthy habit formation, the more likely they are to adopt it.

Key obstacles to integrating ChatGPT in South African healthcare for children and adolescents with diabetes were also identified. Outdated infrastructure, reliance on legacy systems, and limited internet access, particularly in rural areas, were major barriers. These findings align with Yi et al.³⁶ who highlighted inadequate technological infrastructure as a key challenge in AI adoption. Resistance from healthcare professionals was another concern, with participants emphasising that a lack of AI training could lead to scepticism and reluctance to adopt the technology. Moreover, financial constraints were identified as a significant barrier, with participants noting the high cost of integration and limited government support. While previous studies have addressed technological obstacles, this study highlights the critical role of financial constraints and government intervention factors that remain underexplored in existing literature. One of the most important aspects to consider is the alignment of AI-generated recommendations with local clinical protocols. Public healthcare policies in South Africa govern chronic disease management by regulating prescribing practices and clinical decisions based on the Essential Medicines List and code list restrictions. In resource-constrained public sector settings, AI tools such as ChatGPT may generate recommendations that are based on international guidelines or higher-income country standards. In light of this potential mismatch, clinicians must evaluate AI-generated information and ensure that it is aligned with South African policies and regulations, resource availability and situations before applying it in healthcare institutions.

Findings revealed the potential of ChatGPT to provide personalised, immediate support but also revealed concerns regarding language barriers, information reliability, data security, privacy, transparency and the risk of overreliance on technology. Key obstacles include outdated infrastructure, legacy health systems, resistance to change among healthcare professionals and insufficient financial resources, which could hinder its successful implementation. This study contributes to the development of policy on AI integration by identifying ethical, infrastructural and regulatory considerations that must be addressed before tools like ChatGPT can be effectively deployed.

Despite the potential of ChatGPT to improve social and educational support, clinical decision-making and medical advice should remain the responsibility of qualified healthcare professionals and practitioners. The study underscores the need for collaboration among healthcare professionals, AI developers, and policymakers to optimise the capabilities of ChatGPT in diabetes management. As ChatGPT advances, its integration into diabetes management may contribute to a more tailored, effective and accessible approach to managing this chronic illness.