Most people shopping for an AI companion for an older parent get stuck on the same question: does it sound human? A 2026 study suggests that is exactly the wrong thing to lead with.
What the Study Found: The More Human It Sounds, the More They Want It Off
Researchers at the Singapore University of Technology and Design published an experiment on 9 September 2026 in the journal Innovation in Aging. One hundred and forty community-dwelling older adults, average age around 73, each spent about twenty minutes with a voice chatbot and were then asked whether they would want to use it again.
Two factors predicted that answer best: whether the thing felt like it was being straight with them, which the study calls authenticity, and how confident they were about using the technology. The surprising part comes next. Once the researchers accounted for how close participants felt to the chatbot, a more humanlike voice predicted less interest in coming back. Perceived usefulness did not significantly predict the decision. Being useful was not what convinced them. Being real about what it was did.
One more detail is worth keeping. The participants most interested in the chatbot were the ones in poorer health and with lower life satisfaction.
Four Things That Matter More Than Sounding Human
The same speech recognition and emotion modelling goes into a children's AI toy and an elder companion, but the requirements diverge sharply. With children the question is content grading and time limits; with older adults it is tolerance for accents and getting it right the first time. Here is what deserves more attention than realism.
- Does it admit it is a machine? Authenticity was the top predictor in the study, and in product terms that is a small design decision: on first use, does it say it is an AI, or does it work at passing as a person? Some makers already build this in. inTouch's Mary states up front, on the first call, that it is an AI, rather than letting the user discover it.
- Can it follow dialects and accents? Mía, the robotic rabbit from the Social Robotics Group at Carlos III University of Madrid, hides a microphone in its fur and identifies people by voice alone, reported at over 70% accuracy for returning speakers, and it can learn new voices offline. It deliberately carries no camera, to sidestep older adults' unease with lenses. Feedback from care facilities in China lands in the same place: some robots cannot follow local dialects, their on-screen prompts read like engineering documentation, and after a few failed attempts residents simply stop touching them. Research by Jin Yongai's team at Renmin University's School of Population and Health found that 82.9% of middle-aged and older internet users face barriers using the internet, rising to 87.9% among those aged 70 and above.
- Will it call someone when something is wrong? AARP reported a 2025 case in which an ElliQ user repeatedly told the device she felt unwell. The robot triggered an alert to her emergency contact, and she was subsequently treated for sepsis caused by a urinary tract infection. The same report cites an MIT Media Lab study of 70 older Americans tracked over eight weeks, in which companion robots improved personal growth, self-acceptance and overall wellbeing. Several teams, meanwhile, have found that joke-telling features fall flat in testing. Raising an alarm matters far more than landing a punchline.
- Does it actually help with loneliness? Five studies published in the Journal of Consumer Research in 2026 by Julian De Freitas and colleagues found that using an AI companion eased loneliness about as much as interacting with another person, and more than activities such as watching videos online. The mechanism was not distraction; what predicted the drop was whether the person felt heard. The same work found that users consistently underestimate how much these tools help them. One caveat: none of the studies measured whether the effect lasts.
Cost needs to be settled early as well. A private ElliQ lease runs about $60 a month plus a one-time $250 enrollment fee, and in New York, Florida, Michigan, Wisconsin and Washington it is available free through state aging agencies. Whether that route is open to you depends on local coverage.
When you pick one of these devices for an older parent, put realism last. Ask four things first: does it say it is an AI, can it follow your parent's dialect, will it notify family when something looks wrong, and what will the monthly fee keep charging for afterwards. A device with a slightly mechanical voice that never pretends to be a person tends to stay in the room longer than one that performs being human.
© SWPO. All rights reserved.