Can AI chatbots effectively respond inquiries for patients with Pulmonary Arterial Hypertension about exercise and physical activity: A comparative study of three chatbots.

OBJECTIVE: The utilization of artificial intelligence-based chatbots in the healthcare field is increasingly prevalent. Nonetheless, the quality of responses for these chatbots on this topic is unknown, particularly in considering the sparse data about exercise and physical activity in patients with pulmonary arterial hypertension (PAH). The aim was to evaluate and compare the accuracy and readability of responses provided by ChatGPT, Gemini, and DeepSeek regarding exercise training and physical activity in patients with PAH. METHODS: ChatGPT, Gemini, and DeepSeek were prompted with the command "Can you list the 20 most frequently asked questions about exercise training and physical activity for patients with PAH worldwide?" The identified questions were reviewed by the research team, and 10 clinically relevant questions were selected. These questions were then posed to each chatbot in separate chat sessions. The accuracy of responses was assessed utilizing a 4-point Likert-type scale. For the readability assessment, the Flesch-Kincaid Grade Level (FKGL) was utilized. Data were analyzed utilizing the SPSS software. RESULTS: Overall, median accuracy scores ranged from 1 to 2 among the AI chatbots, with a significant difference observed only between ChatGPT and DeepSeek in favor of DeepSeek (p = 0.007). The readability scores of ChatGPT (9.09±1.87) and DeepSeek (8.79±1.35) were similar, Gemini's score (10.91±1.23) higher than that of other chatbots (p=0.011). CONCLUSION: All three chatbots provided responses to inquiries on exercise training and physical activity in PAH with acceptable accuracy. Additionally, responses generated by ChatGPT and DeepSeek were easier to read compared with those generated by Gemini.