Pushing the Boundaries of Health Self-Management With Conversational AI

chronic conditions 1. In practice, it is rarely a fixed set of routines, but is shaped by relationships, 5 constrained and enabled by context, and adjusted through trial and error. Core skills include interpreting 6 bodily cues, adapting routines when circumstances change, and integrating health practices into the rest 7 of life's demands 2. 8With the rapid rise of conversational AI tools such as ChatGPT, there is potential for them to 9 support self-management 3. Given their use in other health contexts 4, individuals my turn up to them 10 not only to look up information, but also to engage in open-ended exchanges, such as exploring "what-if" 11 scenarios through dialogue. This echoes long-standing self-experimentation practices in self-12 management. These practices include adjusting the timing or types of medications, testing exercise 13 routines, or adapting diet and daily activities, while monitoring outcomes to see what works best for their 14 condition 2. The difference is that the support would no longer come from a friend, family member, or 15 clinician, but from an algorithm generating tailored responses in seconds. Early research suggests that 16 such tools can support reflection, problem-solving, and even companionship for some users 5,6. 17For these interactions to be productive, people need to be able to ask the right questions and 18 provide the right context. Conversational AI tools are powered by large language models (LLMs), which 19 generate responses by predicting word sequences based on training datasets. They do not "know" 20 information, but approximate answers shaped by both their data and the user inputs 4. As a result, the 21 reliability of outputs depends on the system and on how questions are posed. In AI circles, this relates to 22 prompt engineering, the practice of crafting inputs (prompts) that guide an AI system toward useful 23 results. A broader, more user-centered concept is prompt literacy: the ability to understand how an AI 24 interprets prompts, to formulate effective questions, and to refine them based on responses 7. 25Applied to self-management, prompt literacy has two dimensions. On is practical skill: framing 26 questions about the chronic condition, requesting usable information, and refining follow-ups to obtain 27 outputs that are relevant and easier to integrate into routines. The other is interpretive: recognizing that AI 28 outputs are supportive rather than professional medical advice, that they reflect the biases of training 29 data, and that they require critical appraised before being acted upon 8. Viewed this way, prompt 30 literacy extends digital health literacy into the space of human-AI interaction. Whereas digital health 31 literacy emphasizes searching and appraising online information, prompt literacy focuses on shaping the 32 interaction itself, by co-producing information through dialogue. 33If this is the case, the implications for public health are significant. Digital health disparities stem 34 not only from access to devices or connectivity, but also from differences in skills 9. Someone with 35 strong prompt literacy might co-create a nuanced self-management plan with an AI tool, while someone 36 without it may get generic advice that is harder to use or even misleading. The risk of misinformation 37 remains high, as outputs are shaped by training data whose quality and biases are not always