Why do older adults dislike an AI voice that keeps saying 'mm-hm' while listening?

This explores which perceptual and physiological factors shape how older adults experience voice agents. The short answer: the corpus points mainly to one factor, timing. It's less about how loud or clear a voice is and more about how the brain handles conversational rhythm. One caveat up front: the collection doesn't cover vision, cognitive load, fatigue or hearing aids for voice interaction, so this is a narrow view rather than a full survey.

The most direct finding comes from a CHI 2025 study. When an LLM voice agent made frequent small listening sounds ('mm-hm', 'right'), older adults rated it lower. The proposed reason is physiological. With age, the brain processes rapid sounds more slowly, so quick-fire listening cues stop sounding attentive and start sounding like interruptions that are hard to separate from the conversation itself Does frequent backchannel from voice agents help or hurt older adults?. What's striking is that this is a hearing problem located in the brain, not the ear. A louder speaker wouldn't fix it.

Another study seems to point the other way. Older adults rated an agent that could interrupt and be interrupted, and that also made listening sounds, as more natural and engaging than a strict take-turns baseline. They didn't resent it when the agent changed the subject to steer toward their interests Do voice agents that interrupt feel more natural to older adults?. Read together, the two studies suggest older listeners don't dislike responsive, overlapping talk. What they dislike is density. The amount and spacing of the cues matters more than whether they're there at all. Work on text-based support chat points the same way: agents that adjust their pacing to the context beat agents with fixed pacing on how well they seem to listen and on how much users open up Can pacing strategies make conversational agents feel more human?.

Looking further afield, a review of how conversation partners 'match' each other separates word-level matching, which helps people understand, from matching emotion and tone of voice, which builds warmth and trust Do different types of alignment serve different conversational goals?. For older users, that means slowing down the rhythm and keeping the warmth are separate adjustments, and a design can get one right and the other wrong. Silence is also something a designer can use. One system fetches memories during the natural pauses in speech, so the agent never adds lag Can memory retrieval hide inside voice agent silence?. That matters if the pauses themselves are what slower listeners depend on.

The less obvious takeaway is that the same vocal signal can mean different things depending on the listener and the situation. In one study, voice features that signalled an outgoing personality in a calm interview signalled anxiety under stress Does personality sound the same in stressful and neutral conversations?. The older-adult studies suggest perception works the same way: a listening cue reads as attentive to one listener and as disruptive to another. A voice agent built for older adults may need to treat timing and the density of its cues as settings it adjusts for each listener, not fixed features of its personality.