Why We Need to Build AI for Seniors

Abstract: The AI industry designs for the young and treats adults over 65 as an edge case. We argue this is backwards. The failures older adults hit are failures of interaction rather than intelligence—timeouts, dense menus, recovery paths built by engineers for engineers—and they are fixable. We take seriously the two strongest objections: that senior-focused products have been failing for two decades, and that AI may harm this population faster than it helps. We propose design principles that follow from treating patience and trust as requirements rather than accommodations, and argue the result is better AI for everyone.

Today, the AI industry designs, tests, and markets its products for people who look like the people building them: young, fluent in software, tolerant of broken interfaces. Meanwhile, the fastest-growing demographic in nearly every developed country—adults over 65—is treated as an edge case. We think this is exactly backwards. Older adults are not a niche market for AI; they may be its most important one.UN World Population Prospects projects the 65+ population to roughly double worldwide between 2021 and 2050.

Building for this population is not a matter of shrinking features or enlarging fonts. It requires rethinking how AI listens, how it paces itself, and how it earns trust—and the result, as we'll argue, is better AI for everyone.

The most overlooked users in AI

By 2050, one in six people on Earth will be over 65. Yet if you audit the major AI assistants for onboarding flow, default type size, speech-rate assumptions, or error recovery, you find products implicitly tuned for a 28-year-old with perfect vision and a decade of touchscreen habits.

The irony is that older adults stand to gain more from capable AI than almost anyone: help managing complex medication regimens, a patient companion against loneliness, a defense against increasingly sophisticated scams, and a way to stay independent at home for years longer.Roughly nine in ten older adults say they want to age in their own homes—the single strongest preference in surveys of this group.

Why today's AI fails older adults

The failures are rarely about intelligence. They are about interaction. Chat interfaces assume fast typing and comfort with open-ended prompts. Voice assistants time out after two seconds of silence—an eternity of pressure for someone who speaks deliberately. Confirmation flows bury critical choices in small gray text. And when something goes wrong, the recovery path is a settings menu designed by and for engineers.Average speaking rate declines with age, and word-finding pauses lengthen—behavior that fixed voice-activity timeouts routinely misread as end-of-turn.

Worse, the industry's default posture toward older users is protective rather than empowering: locked-down “senior modes” that strip capability instead of adapting it. Nobody wants a dumber assistant. They want one that meets them where they are.

Two honest objections

Before going further, the two strongest arguments against everything that follows.

“This has been tried, and it failed.” Mostly true. The past two decades produced a graveyard of senior tablets, simplified phones, and companion robots, and the pattern of failure is instructive. The products were bought by adult children and abandoned by the parents they were bought for, because “senior-friendly” meant capability stripped out rather than interaction rethought. And the underlying technology could not meet the one requirement that matters most: none of it could hold a patient, natural conversation. Menus and touchscreens were the only interface available, and menus and touchscreens are precisely what this population struggles with. That constraint has now lifted. Conversational AI is the first technology that can deliver the interface seniors have been asking for all along—which is why the old failures argue for the opportunity, not against it.

“AI is at least as likely to harm this population as to help it.” This objection deserves to be taken seriously rather than waved away. The same voice models that could power a patient assistant are already powering grandparent scams—reported fraud losses among Americans over 60 reached $4.9 billion in 2024, and voice cloning is making the oldest tricks cheaper.FBI Internet Crime Complaint Center, Elder Fraud Report: reported losses among victims 60 and older were $4.9 billion in 2024, up from $3.4 billion in 2023—and reported losses understate the total, since elder fraud is heavily under-reported. A model that hallucinates a medication instruction is more dangerous to a user who was never taught to distrust confident answers. And a companion that is infinitely patient and always available could substitute for human contact instead of bridging to it. We believe these risks are tractable design problems—an assistant should verify health answers against the user's actual records, confirm irreversible actions out loud, and share what it does with the senior's care circle (the family and caregivers they have consented to) rather than operate in private—but nobody has demonstrated this at scale yet. Anyone building in this space should treat these as open problems, not solved ones.

The struggle is measurable

This is not anecdote. Pew Research has tracked older adults' relationship with technology for two decades, and the pattern is consistent: nearly three-quarters of Americans 65 and older say a new device usually requires someone else to set it up or show them how to use it, only about a quarter of older internet users feel very confident doing what they need to do online, and fewer than one in five would feel comfortable learning a new device entirely on their own.Pew Research Center, “Tech Adoption Climbs Among Older Adults” and “Older Adults and Technology Use.” Confidence figures are among internet-using seniors.

The gap shows up in adoption too. Seniors now own smartphones in large numbers—but they remain the least-connected age group by every measure, and the gap widens sharply with age.Pew Research Center, 2024–2025 surveys of U.S. adults. Within the 65+ group itself, adoption falls off considerably after the mid-70s.

The support gap

When something breaks, who do seniors turn to? Overwhelmingly, their adult children—who are busy, remote, and often impatient. Research on older adults' help-seeking finds that families frequently can't or won't walk through problems step by step, grandchildren fix the symptom without teaching the skill, and retail “experts” are distrusted as salespeople.A survey of 750 older Australians found adult children were the first source of tech help for 44%—and also the source most often described as unwilling to explain.

This is precisely the role a well-built AI can fill: infinitely patient, available at 2am, never annoyed at being asked the same question twice—a way to stop small frustrations from hardening into family obligations.

What seniors actually need

We think the help older adults actually need clusters in four places.

The first is health navigation, because that is where the stakes concentrate. A typical 75-year-old leaves the hospital with a discharge summary written for other clinicians, instructions from three specialists who have not spoken to each other, and a medication list that changed while nobody explained why. An assistant that can interpret the summary, reconcile the instructions, and help prepare questions for the next appointment is doing work that today falls to a stressed adult child—or to no one.

The second is scam defense. Older adults are the primary targets of fraud that gets cheaper to run every year, and the moment to intervene is before the money moves: a trusted second opinion on every suspicious call, text, and email—proactive, not after the fact.

The third is connection. Not a replacement for family, but a bridge to it: help writing messages, sharing photos, remembering birthdays.

The fourth is independence itself: managing bills, appointments, and transport so that living alone stays viable for longer.

Design principles

Voice and patience first

Speech is the interface humans have practiced the longest. An AI built for seniors should listen without timeouts, tolerate self-correction and long pauses, ask one question at a time, and never punish the user for going slowly. Latency budgets should optimize for comprehension, not throughput.

Trust as a feature

Every action should be explainable in one plain sentence. Every irreversible step should be confirmable out loud. And the system should be transparent with the care circle: everything shared with family or caregivers is visible to the senior, in plain language, revocable at any time—consent is not a checkbox, it's an ongoing conversation.Shared access is where most eldercare products quietly fail: either the senior loses agency, or the family loses visibility. Getting this boundary right is a product problem, not a legal one.

What good looks like

The principles are abstract until you hear them. Ruth is a composite, but every beat of this exchange is a documented failure mode of today's assistants: the pause that triggers a timeout, the self-correction that derails the flow, the action taken without plain-sentence confirmation.

The economics are on our side

Adults over 50 control the majority of household wealth in most developed economies, and the cost of eldercare is rising faster than nearly any other category of spending. This is not a charity market. It is one of the largest under-served markets in technology.Federal Reserve, Distributional Financial Accounts: households headed by someone 55 or older hold roughly 70% of U.S. household wealth.

The clearest way to see the opportunity is the cost of care itself. Once independent living becomes untenable, every alternative is dramatically more expensive—for families and for public payers.Care-setting costs per the Genworth Cost of Care Survey (2024 medians: assisted living ≈ $71k/year; a semi-private nursing-facility room ≈ $111k/year; part-time home care from hourly home-health-aide rates). The independent-at-home figure assumes no paid care.

Whether an assistant can actually delay a move to assisted living is unproven—to our knowledge, no one has run that study. But the asymmetry is stark: the intervention costs hundreds of dollars a year, and the outcome it competes with costs $70,000. Even a modest probability of working clears the bar, and it is a testable proposition—which is exactly why someone should run the study.

Three structural forces make the opportunity durable. Demographics: the 65+ population is the only large consumer segment guaranteed to grow for the next three decades. Labor: professional caregivers are already scarce, and the caregiver-to-retiree ratio worsens every year—capable software is not displacing anyone; it is filling a gap no workforce can. Payers: insurers and public health systems have a direct financial incentive to fund anything that keeps people healthy at home, which gives products in this space a rare second customer beyond the family itself.

A call to builders

Building for seniors is not about simplifying AI—it's about finishing it. An assistant that survives a six-second pause and asks one question at a time is not a niche product; it is what every voice interface should have been. Curb cuts were poured for wheelchairs, and everyone with a stroller, a suitcase, or a sprained ankle uses them. If the last decade of software was built for the young, the next one should be built for the people who raised them.

Citation

Please cite this work as:

Moses, Steve, "Why We Need to Build AI for Seniors", Co-Intelligence Labs: Thinking, Jun 2026.

Or use the BibTeX citation:

@article{moses2026seniors,
  author  = {Moses, Steve},
  title   = {Why We Need to Build AI for Seniors},
  journal = {Co-Intelligence Labs: Thinking},
  year    = {2026},
  month   = {jun},
  url     = {https://cointelligencelabs.com/thinking/why-we-need-to-build-ai-for-seniors}
}