Acceptability of Using Large Language Models to Support Smoking Cessation Attempts: a Qualitative Study Among African Americans Who Smoke

Clinical Snapshot

Key Findings

• P Value: Not applicable for qualitative research

• Effect Size: Not applicable for qualitative research

• Primary Outcome: Three themes: limited LLM familiarity, willingness to use LLM-based apps, and perceptions of features impacting acceptability

• Nnt Or Sensitivity: Not applicable for qualitative research

• Confidence Interval: Not applicable for qualitative research

Clinical Application

Findings suggest moderate feasibility with need for community engagement and cultural tailoring Limited direct applicability to Australian Aboriginal and Torres Strait Islander populations, though principles of community engagement and cultural adaptation remain relevant for Indigenous health initiatives African American adults who smoke, particularly those considering digital smoking cessation interventions

Abstract

INTRODUCTION: African Americans face significant disparities in tobacco-related diseases and access to smoking cessation resources. Digital health interventions using large language models (LLMs) show promise in increasing engagement and successful quit attempts. However, impressions regarding acceptability of LLM-based interventions among African Americans are unknown and evidence suggests such interventions require tailoring for underserved communities. Our goal was to evaluate perceptions of LLM-based interventions among African Americans to inform development of an LLM-based chatbot to support quit attempts. METHODS: We conducted remote focus groups with adult African Americans residing in Minnesota who smoke cigarettes to determine acceptability of LLM-based health interventions and inform the functionality of an LLM-based chatbot supporting quit attempts. Discussions followed a structured guide and were recorded to facilitate analysis. Transcripts were coded independently by two analysts with a shared codebook. RESULTS: 21 adults participated in focus group discussions. 57.1% of participants were female and the mean age was 45.6 (SD, 10.5). Three themes united participants' feedback: (1) participants limited familiarity with LLM and related AI technology, (2) willingness to use an LLM-based app to support quit attempts, and (3) perceptions of features that impact the acceptability of an LLM-based chatbot to support quit attempts. CONCLUSION: African Americans have complex perceptions and expectations regarding the use of LLMs to deliver health interventions. While many recognize the potential benefits, significant concerns and knowledge gaps remain. To build trust and acceptability, intervention developers should prioritize community engagement and ethical guidelines to ensure that LLM-driven solutions are inclusive and effective for multiple communities. IMPLICATIONS: Large language models and related AI technologies have shown promise for improving access and engagement when deployed through digital health interventions. Focus groups with African Americans who smoke combustible cigarettes revealed factors impacting the acceptability in using LLM-based interventions to support quit attempts. Efforts to build digital health interventions should prioritize community engagement to reduce apprehensions and improve alignment with community needs and values.