Which Older Adults Are Being Studied?

We cannot claim that an interface works for older adults when the research has quietly selected the older adults most likely to make it work.

I recently came across the review paper “AI-powered voice assistants for older adults: a literature review of insights, research practices, and future directions.” It brought me back to two earlier posts I wrote about the myth of the 50+ user and the wide range of people hidden inside that category.

The review examines 48 studies of older adults using AI-powered voice assistants. Its most interesting contribution is not simply what it says about the technology. It is what it reveals about how older adults are selected and studied.

Many of the studies involved short periods of use, controlled settings, technologically confident participants, and substantial support from researchers. Some recruited participants who were primarily college educated, white, higher income, or already comfortable participating online. Digital literacy and continuing support needs were not always documented.

Under those conditions, positive reactions may tell us that a technology worked during a supported trial. They tell us less about whether it would remain useful after the researchers left, the novelty faded, or a problem occurred.

The problem is larger than voice assistants. AgeTech research can reproduce the same oversimplification it is meant to correct. “Older adults” become a research category, even though the people included may represent only a narrow segment of that population.

I do not conduct rigorous scientific research, but I recently had an experience that reinforced the importance of context.

I facilitated a session about AI with a group of older adults living in a retirement community. I had made assumptions about what the audience would already understand and what questions they would bring to the discussion.

Those assumptions were influenced by my experience teaching generative AI to older adults over the past 18 months. The age range was similar. The retirement-community group had expressed interest in AI and showed up for a session about it.

But that did not make them equivalent to people who had deliberately registered for a course on generative AI.

Course participants had already crossed an important threshold. They had identified a reason to learn about the technology and were willing to spend several weeks experimenting with it. Interest in a general discussion of AI did not necessarily imply the same familiarity, confidence, or readiness to use it.

The experience also led me to wonder about another contextual difference.

Leaving the workforce can reduce routine exposure to the places where new technologies are being introduced. Work provides access not only to tools, but also to informal conversations, changing expectations, shared problems, and colleagues who are experimenting with new systems.

That exposure can be difficult to reproduce from the outside.

Even as someone who remains active through teaching, consulting, volunteering, and writing, I have lost direct contact with how AI is being deployed inside organizations. I can read reports and case studies, but I no longer hear the everyday conversations about which tools are encouraged, which are blocked, what is working, and what employees are quietly resisting.

Living in a retirement community may add another layer of insulation for some residents. This is not the same as loneliness or social isolation. A person may have a rich social life within the community while having fewer routine encounters with workplaces, schools, younger families, and institutions experimenting with new technologies.

In that setting, understanding of AI may come more heavily through television coverage, organized lectures, institutional communications, and conversations with other residents. Reactions to AI may therefore reflect a particular information environment, not simply age or individual digital competence.

This would vary considerably. Mobility, family relationships, volunteering, continued employment, location, and connections with the surrounding community all matter. Retirement communities are no more uniform than the people who live in them.

That is precisely the point.

Age alone tells us very little about someone’s experience with technology. So does attendance at a lecture. So does willingness to participate in a study.

Research findings need to be read alongside the conditions under which they were produced. Who participated? Who was excluded? What support was provided? How long did the study last? What happened when the technology failed? Did the participants continue using it after the study ended?

Systematic reviews and literature reviews are especially useful because they can reveal patterns that are difficult to see in an individual study. They can also expose recurring gaps in research design.

They do not eliminate the need for judgment. Their conclusions are still limited by the studies available to review.