JMIR Publications

ABSTRACT

Background:

Conversational agents (AI-based chatbots) offer a novel approach to health interventions by providing personalized, adaptive interactions that improve over time based on user engagement. In nutrition education, given the wide variation in knowledge, skills and aptitude across participants, AI-based chatbots have the potential to enhance accessibility, engagement, and behavior change. Food is Medicine (FIM) interventions, which aim to improve food security and diet quality among multicultural, at-risk populations, often encounter challenges in engagement and utilization.

Objective:

This manuscript presents the design and development of an AI-chatbot designed to support a FIM intervention.

Methods:

We used an iterative development process informed by behavioral theory, human-centered design (HCD), and Plan-Do-Study-Act (PDSA) quality improvement cycles. The conversational agent was integrated into a randomized controlled trial (RCT) with two study arms: (1) PRx plus standard nutrition education, and (2) PRx plus AI-driven personalized nutrition chatbot support. Architecture design, usability, refinements, and workflow integration were documented across two PDSA cycles. Qualitative feedback was collected from participants and community advisory group (CAG) members to identify facilitators and barriers to chatbot use. This trial is registered at ClinicalTrials.gov (NCT07165990).

Results:

The chatbot was developed using the ChatGPT-3.5 Turbo API. An initial prototype, built in Python with Gradio, enabled rapid testing but lacked flexibility for modifications. To improve scalability and logging capabilities, the system was rebuilt using PHP, HTML, JavaScript, and SQL. Following deployment, two PDSA cycles guided iterative refinements. The first cycle addressed low initial engagement, while the second focused on improving content relevance and aligning responses with user expectations. Informational Q&A sessions identified behavioral theory themes influencing engagement including high cooking self-efficacy, perceived lack of necessity, and low urgency due to competing priorities. These findings informed targeted improvements, such as culturally tailored content and reminder prompts to nudge usage.

Conclusions:

This work addresses critical gaps in nutrition education research by embedding a bilingual, customizable AI conversational agent into a rigorous RCT. The findings demonstrate feasibility and provide insight for scaling FIM programs with inclusive, technology-enhanced nutrition education. Future research should assess clinical outcomes, long-term engagement, and policy implications for integrating nutrition AI-driven support to vulnerable populations. Clinical Trial: This trial is registered at ClinicalTrials.gov (NCT07165990).