Having a ChatGPT-based counselling tool in a local specialist HIV and sexual health clinic setting has been shown to be feasible and acceptable to both patients and healthcare professionals (HCPs) in a pilot implementation study from Singapore.
“In this study, which incorporated the chatbot within the clinic workflow, prior to physician or pharmacist consultation, the chatbot allowed patients to clarify concerns and may have facilitated what was perceived to be more focused and efficient discussions with clinicians,” according to first author Matthew Koh from the National University Hospital, Singapore, and colleagues.
Patients used the chatbot mostly for counselling related to initiation or switching of antiretroviral therapy (48.6 percent for both), followed by initiation of other preventive or treatment strategies such as doxycycline post-exposure prophylaxis (28.6 percent). [HIV Med 2026;doi:10.1111/hiv.70323]
According to feedback from both patients and HCPs, the chatbot helped improve patient understanding and enabled more focused consultations. However, clinically relevant inaccuracies were also identified, including incorrect advice regarding specific drug–drug (dolutegravir-statin) interactions and inaccurate medication pricing information, as well as limitations in contextual specificity, Koh and colleagues noted.
Patients with limited digital literacy also appeared to have some difficulty using the chatbot, with some HCPs noting that the quality and depth of the counselling provided by the chatbot required a degree of initial proactive interaction and input from the patients, they added.
Attitudes towards the chatbot, assessed using domains adapted from the Technology Acceptance Model Edited to Assess ChatGPT Adoption (TAME-ChatGPT), where higher scores indicated a positive attitude towards ChatGPT, were generally favourable. Scores were high and similar between patients and HCPs for anxiety (median, 10 vs 11; possible range 3–15), technology/social influence (median, 20 vs 20; possible range 5–25), and willingness for future use (median, 4; possible range 1–5). Perceived risk scores were lower for patients vs HCPs (median, 13 vs 18; possible range 5–25).
Adding the chatbot into the workflow
The pilot study included 35 patients and 35 HCPs, including 11 physicians and six pharmacists. Most patients were between 30 and 59 years of age, male (97.1 percent), of Chinese ethnicity (77.1 percent), and holding a university or postgraduate degree (54.2 percent).
In a specialist HIV and sexual health clinic, patients were invited to interact with the chatbot on a tablet computer during waiting times prior to consultation. Patients were directed to input any queries they might have related to their condition, treatment, or preventive strategies and were instructed not to enter names, identifiers, contact details, or other personally identifiable information.
Following the chatbot use, a physician or pharmacist reviewed the responses with patients during standard counselling to ensure accuracy and to help put the advice given into the appropriate clinical context.
Koh and colleagues noted that the findings from the pilot study, while exploratory, point to the potential value of large language models as adjunctive tools to enhance patient education and engagement, particularly in settings where counselling is complex, sensitive, or time-limited.
“This ‘pre-consultation priming’ effect mirrors observations from other clinical settings, where artificial intelligence-assisted tools improved patient understanding without replacing clinician input,” they continued. [J Travel Med 2025;32:taae099]
“Importantly, the supervised use of the chatbot ensured that responses could be contextualized and corrected where necessary, reinforcing its role as a supportive, rather than standalone, tool. Additionally, while there are concerns pertaining to the accuracy of the information provided by ChatGPT, having a physician or HCW review the information with the patient afterwards mitigates this risk,” Koh and colleagues said.
The authors emphasized the need for additional work to evaluate the chatbot’s factual accuracy, safety, and long-term clinical effectiveness before considering broader clinical rollout.