Structured Exploration vs. Generative Flexibility: A Field Study Comparing Bandit and LLM Architectures for Personalised Health Behaviour Interventions

A 4-week study comparing bandit algorithms and LLM architectures for personalized health behavior interventions found that LLM-based messaging approaches were rated more helpful than templates, but contextual bandit optimization provided no additional benefit over LLM-only methods. The research reveals a trade-off between structured exploration of behavior change techniques and generative flexibility in AI health systems.

• →LLM-based messaging approaches were rated substantially more helpful than template-based methods in health behavior interventions.
• →Contextual bandit optimization for behavior change technique selection showed no additional perceived helpfulness compared to LLM-only approaches.
• →Unconstrained LLMs focused heavily on single behavior change techniques while bandit systems enforced systematic exploration across multiple techniques.
• →Contextual acknowledgement of user input was identified as a key driver of perceived helpfulness in AI health interventions.
• →The study contributes design suggestions for balancing structured exploration and generative autonomy in reflective AI health systems.