One Question, Four Voices: How Advice for Alzheimer's Caregiving Differs Between Caregivers, Clinicians, and Large Language Models

Key Findings

• — A consistent division of communicative labour: peers give time-anchored narratives and community grounding, clinicians give concise, advice-forward, clinically bounded guidance, and LLMs produce a formal, verbose, highly structured 'consultation' voice.
• — CareGPT (retrieval plus prompt injection of peer discussions) shows minimal measurable separation from plain ChatGPT.
• — LLM outputs score higher than peers on politeness and emotion-support markers and are comparable to clinicians on informativeness and physician-alignment measures, but are less narrative and less contextually situated.
• — Topic modelling shows LLM responses concentrate on advice-oriented themes while peers cover emotional processing, legal and financial concerns and daily care.

Methodology Notes

85 forum questions; four responder groups; 28 measures; LLM-based scoring calibrated against double-coded human ratings; single caregiving domain (ADRD), authors note other caregiving domains remain to be tested. Funded by the HRSA Geriatric Workforce Enhancement Program. CC BY 4.0. Abstract from the Crossref record; dl.acm.org walled; no arXiv preprint located (the arXiv title API returned zero entries for every query this session, including known titles).

Sources