An older adult from one of the community collaborators interacting with the AI chatbot, Ami Chat. Credit: SUTD
Seniors Prefer AI Companions That Feel Caring Over Ones That Sound Human
In A Nutshell
- A 140-person study in Singapore found that confidence and a genuine-feeling connection predicted older adults’ interest in using an AI chatbot, more than how human it sounded.
- Authenticity, the sense that the chatbot felt genuine rather than scripted, predicted interest even after accounting for every other factor measured.
- Once closeness with the chatbot was factored in, sounding more humanlike actually predicted less interest in using it, not more.
- Participants in poorer health and with lower life satisfaction showed more interest in the chatbot, possibly reflecting unmet needs it could help fill.
When an AI companion feels genuinely caring, older adults connect with it, regardless of how humanlike its voice sounds. That is the core finding from a new study of 140 older adults in Singapore, published in the journal Innovation in Aging.
Researchers built a voice-based AI companion called Ami Chat, designed specifically for seniors, and had participants spend about 20 minutes talking with it. Authenticity, the sense that the chatbot felt genuine rather than scripted, remained a significant predictor of continued interest even after every other factor was accounted for. Sounding humanlike told a different story: once closeness with the chatbot was factored in, a more humanlike voice tracked with less interest, cutting against a common assumption in tech design.
Confidence in using the technology, along with genuine connection, tracked most closely with interest in continued use, more so than any general enthusiasm for the technology itself or how impressive it seemed at first glance. As more seniors turn to AI for company amid rising loneliness, the findings point toward what tends to work, and what may not.
140 Older Adults Tested an AI Companion Chatbot for 20 Minutes
Scientists recruited 140 community-dwelling older adults in Singapore, averaging about 73 years old. Most were women, with a range of educational backgrounds. Most reported household incomes below 2,000 Singapore dollars a month. None had a formal clinical diagnosis of a neurological, psychiatric, or cognitive disorder, and each completed a standard cognitive screening test used to describe the sample rather than to screen anyone out.
Each participant then spent time with Ami Chat, a voice-based chatbot built using OpenAI’s technology and designed with a warm, feminine persona. Talking replaced typing, buttons were large and simple to tap, and the bot tolerated mistakes and offered help along the way. It also asked gentle check-in questions about memory and mood, based on international guidelines for healthy aging. Afterward, participants rated how easy the chatbot was to use, how humanlike and genuine it felt, how close the interaction felt compared with a real friendship, and how likely they would be to use it at home.
Authenticity Mattered More Than Humanlike Design in AI Chatbots for Older Adults
Two things stood out as the strongest reasons participants wanted to keep using Ami Chat. First was confidence: feeling capable of using the technology and having the support to do so. Participants who felt confident and well supported were far more likely to want to keep talking with it.
Second was authenticity, whether the chatbot felt genuine rather than robotic, which still predicted interest after accounting for every other factor measured in the study. Closely tied to authenticity was how much a conversation with the chatbot resembled talking to a real, trusted friend, complete with back-and-forth exchange and a sense of connection. A stronger sense of closeness corresponded with greater interest in using the chatbot again.
Once that sense of closeness was factored in, participants who found the chatbot more humanlike tended to report less interest in using it, not more. Sounding more human did not help adoption in this sample and may have worked against it. One possible explanation researchers raised: overly humanlike AI can feel unsettling, similar to the “uncanny valley” effect, where something almost, but not quite, human unsettles people instead of putting them at ease.
Participants in poorer health, particularly those with lower cognitive sharpness and life satisfaction, actually reported more interest in the chatbot. Researchers suggest this may reflect greater unmet needs among people facing more daily challenges, needs a judgment-free AI companion might help fill. A similar dynamic has surfaced in research on other AI platforms; in a study of a different companion app cited by the researchers, one participant described the experience of chatting with it as feeling like “having a companion who truly cares about [the participant] and share [his] interests.”
A generally positive view of the chatbot, including whether participants thought it was useful, did not significantly predict interest in using it, contrary to older theories of tech adoption that treat usefulness as the main driver of acceptance. Confidence and connection carried the weight instead.
AI Chatbots for Older Adults Work Best When They Feel Real, Not Human
Loneliness among older adults is a serious and well-documented public health problem, and AI companions are increasingly marketed as a partial fix, with one state agency already distributing hundreds of AI companion devices to combat isolation. This study suggests that making these tools more humanlike is not necessarily the winning strategy some developers assume. What appears to matter more is whether an interaction feels genuine and personally attentive, paired with a design older users can handle with confidence. For anyone choosing or building AI companionship tools for aging parents or grandparents, the study suggests authenticity and usability may matter more than a convincing human disguise.
Disclaimer: This article is based on a peer-reviewed study. The findings describe statistical associations from a single 20-minute session with one chatbot prototype rather than proven cause-and-effect relationships, and may not generalize to other AI companions or to everyday, long-term use.
Paper Notes
Limitations
Several important caveats accompany this study, according to its authors. Participants interacted with the chatbot for only a brief, roughly 20-minute session in a controlled setting, which may not reflect how people would experience the chatbot during everyday, long-term use. The sample was limited to older adults without diagnosed cognitive impairment or disability, so the findings may not extend to those populations, who might prioritize different features such as memory reminders. Some participants also experienced technical issues like internet connectivity problems or speech-recognition failures during their sessions, which were not excluded from analysis since they were considered part of the real experience of ease of use. Because researchers were present in the room during testing, participant responses could have been influenced by social desirability bias, meaning people may have held back from sharing very personal thoughts with the chatbot or may have leaned more heavily on the researcher for help than they otherwise would have. Additionally, the study used a cross-sectional, one-time-point design, meaning it cannot establish cause and effect, only associations. Some subcomponents of the measures used, such as certain health and control-belief factors, were measured with only single survey items, which may limit their reliability.
Funding and Disclosures
This research was supported by the National Research Foundation Singapore under the AI Singapore Programme, and by the Ministry of Education Academic Research Fund Tier 1, Singapore, under the Singapore University of Technology and Design’s Kickstarter Initiative. The authors reported no known conflicts of interest.
Publication Details
This study, titled “Factors influencing older adults’ intention to use social chatbots: examining technology acceptance, anthropomorphism, authenticity, and artificial intelligence social interaction,” was authored by Ka Lon Sou, PhD, Wei Quin Yow, PhD, FGSA, Matthew Zhi Yuan Lau, BEng, and Fuxi Ouyang, MSc, all affiliated with the Singapore University of Technology and Design. It was published in Innovation in Aging, 2026, Volume 10, Issue 9 (igag073), by Oxford University Press on behalf of the Gerontological Society of America. DOI: https://doi.org/10.1093/geroni/igag073.
