Grandparents and grandkids may talk differently with AI — not over who leads, but whether small cues like 'mm-hm' land right.

This explores whether older adults and younger family members handle turn-taking, interruption and topic-steering differently, especially when talking with AI. Up front: the corpus has no study comparing the two groups side by side, so it can't tell you how a grandparent and a grandchild differ in the same conversation. What it does have is a cluster of findings about older adults and voice agents. Those findings suggest the most important difference may not be who takes control. It may be how well the small signals that manage control come through.

The clearest example is backchannel, the 'mm-hm', 'right' and 'yeah' that listeners make while someone else is talking. Younger listeners barely notice these cues. For older adults, a voice agent that backchanneled often lowered their ratings of the experience. The reason was not rudeness. Age-related slowing in how the brain processes sound made rapid cues hard to separate from the conversation, so they felt like noise rather than attention Does frequent backchannel from voice agents help or hurt older adults?. A signal designed to say 'I'm listening' can become an interruption depending on the listener's hearing.

The surprise is that older adults did not reject interruption itself. A 'barge-in' agent that could interrupt mid-sentence, and be interrupted, felt more natural, engaging and fluent to older adults than a polite strict-turn-taking agent. When the agent cut in to change the topic, steering toward things participants cared about, people didn't resent it Do voice agents that interrupt feel more natural to older adults?. The stereotype that older users want slow, deferential, one-at-a-time exchanges doesn't hold up. What matters is how often and when the cues arrive. Interruptions with a clear purpose work. Constant filler cues don't.

A wider lens helps explain why. In a month-long study of adults talking with an AI companion, control was neither held by one side nor fought over. It was negotiated turn by turn: people started topics and set boundaries, while the AI asked questions and persisted How does conversational control actually shift between humans and AI?. A review of 'alignment', meaning how speakers match each other, found that matching word choice helps get tasks done, while matching emotion and tone builds warmth and trust Do different types of alignment serve different conversational goals?. Put together, generational differences are likely to show up in which channel carries the negotiation. Tone, timing and prosody are exactly where age-related hearing changes have an effect. The structure of a conversation alone can predict whether it goes well almost as well as its content can Can conversation shape predict whether it will work?, which suggests timing and turn patterns are worth studying for their own sake.

One caution applies to anyone designing for mixed-age households. Today's chat models are trained in ways that reduce 'grounding' moves, such as checking understanding or asking a clarifying question, to well below human levels Does preference optimization harm conversational understanding?. Those are the repair moves a listener with slower auditory processing may need most. The open question the corpus leaves is whether one agent can adjust its timing and repair behavior to whoever in the family is speaking. No paper here tests that yet.