"I Felt Listened to": Evaluating an AI-Powered Reflection Tool for Care Partners
Jazette Johnson, Hira Jamshed, Rachael Zuppke, Amanda Leggett, Emily Mower Provost, Robin N. Brewer · 2025 · ACM Transactions on Accessible Computing · doi:10.1145/3731562
Summary
This study evaluates CareJournal, an Amazon Alexa skill designed to support communication between older adult care receivers and their family caregivers through AI-assisted daily reflection. The research addresses the challenge of "articulation work"—the often invisible effort of putting care needs and feelings into words—which is difficult for both caregivers managing multiple responsibilities and older adults struggling with autonomy and asking for help. CareJournal prompts care partners to complete three daily reflection questions via voice, then generates summaries (one AI-generated using ChatGPT 3.5, one human-written) that partners can choose to share with each other. The researchers conducted a 4-week pilot study with 14 care partner pairs followed by a 4-week field study with 16 pairs, refining the system between phases. Participants included diverse relationship dynamics (spouses, adult children, parents, siblings, grandparents) with care receivers aged 19-97 and caregivers aged 38-85. The Echo Show 8 was chosen for its visual display supporting users with hearing disabilities. Reflection questions evolved from the pilot to the field study—changing from "Is there anything you wish you or your care partner had done differently?" to "What would you have wanted to do differently?"—to elicit more descriptive responses rather than one-word answers. The study examined two research questions: how AI tools can support articulation work in care relationships, and what effects AI-based articulation tools have on care partners' relationships.
Key findings
In the field study, participants logged 1,413 reflections (660 from care receivers, 753 from caregivers), with non-descriptive responses dropping from 31% in the pilot to 12.46% after question refinement. Participants showed nearly equal preference between AI-generated (N=62) and human-written (N=60) summaries, challenging prior skepticism about AI-generated content quality. Participants valued summaries that contained detailed content relevant to shared experiences, had natural conversational flow, accurately represented their words and meanings, and matched their emotional tone. CareJournal strengthened care partner connections by enabling more focused and intentional communication. One care receiver noted it helped balance their relationship: "I am the talker and she's the listener. So, it balanced it out a little bit for me." The tool improved awareness of needs—caregivers learned about tasks they commonly overlooked, while care receivers felt validated when summaries acknowledged their contributions. However, AI hallucinations occurred, with one participant reporting the summary mentioned a "wound nurse" that was never discussed, highlighting risks of persuasive but inaccurate AI-generated content. Technical constraints frustrated participants: Alexa's speech timeout cut users off mid-thought, forcing some to pre-plan responses or respond with "nothing" rather than risk being interrupted. Speech recognition struggled with non-American accents and names, requiring some participants to use text input via the Alexa mobile app instead.
Relevance
This research demonstrates that AI can enhance rather than replace human-to-human relationships in care contexts—a significant finding given concerns about technology depersonalizing care. The concept of "articulation-driven AI" offers an alternative to deficit-focused AI that detects problems; instead, AI supports people in expressing their own needs and experiences. For accessibility practitioners, the study highlights both opportunities and challenges in voice-based interfaces for older adults and diverse users. Key design implications include: prompts must be carefully engineered to elicit descriptive rather than yes/no responses; voice systems need longer timeout periods and better accommodation of natural speech pauses, accents, and speech impairments; users should be able to review and edit reflections before sharing; and AI summaries require human oversight to catch hallucinations. The potential for CareJournal to serve multiple caregivers (including paid professionals) suggests broader applications for care coordination. However, the study is limited by its use of a single voice assistant platform, relatively small sample, and the emotional complexity revealed when daily reflection highlighted limitations in some participants' lives rather than positive experiences.
Tags: voice assistants · generative AI · caregiving · older adults · reflection · communication · Amazon Alexa · articulation work · care relationships