Abstract:Adherence to healthy diets reduces chronic illness risk, yet rates remain low. Large Language Models (LLMs) are increasingly used for health communication but often struggle to engage individuals with ambivalent intentions at a pivotal stage of the Transtheoretical Model (TTM). We developed CounselLLM, an open-source model enhanced through persona design and few-shot, domain-specific prompts grounded in TTM and Motivational Interviewing (MI). In controlled evaluations, CounselLLM showed stronger use of TTM subprocesses and MI affirmations than human counselors, with comparable linguistic robustness but expressed in more concrete terms. A user study then tested CounselLLM in an interactive counseling setting against a baseli…
Abstract:Adherence to healthy diets reduces chronic illness risk, yet rates remain low. Large Language Models (LLMs) are increasingly used for health communication but often struggle to engage individuals with ambivalent intentions at a pivotal stage of the Transtheoretical Model (TTM). We developed CounselLLM, an open-source model enhanced through persona design and few-shot, domain-specific prompts grounded in TTM and Motivational Interviewing (MI). In controlled evaluations, CounselLLM showed stronger use of TTM subprocesses and MI affirmations than human counselors, with comparable linguistic robustness but expressed in more concrete terms. A user study then tested CounselLLM in an interactive counseling setting against a baseline system. While knowledge and perceptions did not change, participants’ intentions for immediate dietary change increased significantly after interacting with CounselLLM. Participants also rated it as easy to use, understandable, and supportive. These findings suggest theory-driven LLMs can effectively engage ambivalent individuals and provide a scalable approach to digital counseling.
| Subjects: | Human-Computer Interaction (cs.HC) |
| Cite as: | arXiv:2511.02428 [cs.HC] |
| (or arXiv:2511.02428v1 [cs.HC] for this version) | |
| https://doi.org/10.48550/arXiv.2511.02428 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Chaewon Bak [view email] [v1] Tue, 4 Nov 2025 09:58:45 UTC (2,217 KB)