My mother is 71. She has been using ChatGPT for eight months. Last week she sent me a prompt she'd written to help her plan a four-day trip to Portugal, and it was better than anything I've written in two years. It had a role assignment, a constraints section, a tone specifier, a multi-step reasoning instruction, and a self-check clause. She didn't know the terms for any of that. She just knew what she wanted, and she'd iterated until she got it.
I'd been operating on the assumption that she was the exception. She's not. She's the leading edge of a pattern that the AI industry has been almost completely blind to: the heaviest, most patient, and increasingly most proficient users of language models are not the young. They're the retired.
This piece is about why that's happening, why it matters, and one sacred cow that the AI industry needs to stop grazing on before it builds another generation of products for the wrong demographic.
The Sacred Cow: "AI Proficiency Belongs to the Young"
Here's the belief I want to dismantle, and it's practically an axiom in every demographic survey and product brief in the industry.
The claim goes like this: younger people are digital natives, so they'll adopt and master AI faster. Older people will lag, need training, and use it in shallow ways.
This is measurably wrong on the dimension that matters.
What surveys actually show:
Young people are more likely to try AI tools. They adopt more quickly, sample more products, and are more willing to experiment.
Young people are less likely to persist with a single tool over time. Their usage is broad, shallow, and fragmented across many apps and use cases.
Older users who adopt AI spend dramatically more time per session. Not slightly more. Multiple times more. They read the outputs. They revise the prompts. They run variations. They treat it like a craft.
The result is a quiet inversion. The cohort with the most breadth of AI use is young. The cohort with the most depth of AI use is elderly. And depth, in language models, is where proficiency lives.
If you want to find the best prompters in the world right now, don't look at the 22-year-old prompt engineers on Twitter. Look at the 68-year-old retired nurse who's been using Claude to help her write a memoir for nine months and has developed a personal prompt library she refines every week.
Why the Elderly Are Actually Better at This (Not Just More Patient)
Let me be specific, because "they have more time" is the lazy explanation and it's not the whole story.
- They have more to ask about. A retired person has 40 to 60 years of lived experience across multiple domains: career, family, health, finances, relationships, loss, travel, craft. The models are most useful on tasks that require domain knowledge to formulate well. A 25-year-old asking an AI for career advice has thin priors. A 65-year-old asking for career advice has decades of context, and the model's output is far more useful because the input is richer.
- They've had practice with the pre-AI equivalent. The best prompters are not the ones who learned prompt engineering from a course. They're the ones who spent 40 years writing letters, crafting memos, negotiating with contractors, writing eulogies, drafting medical questions, and asking precise questions of doctors, lawyers, and bureaucrats. The underlying skill is the same: formulating a clear request and evaluating the response. Younger users have had less practice because they've had less to practice on.
- They're patient with iteration. The single biggest determinant of prompt quality is iteration count. A prompt that works is usually the fifth or tenth version. Young users tend to try once or twice, get a mediocre answer, and switch tools. Older users tend to iterate because they've learned from a lifetime of tasks that mastery requires repetition. They are not in a hurry. This is the decisive advantage.
- They're better at evaluating outputs. Younger users grew up with outputs that were either right or wrong, and they're calibrated to fast feedback. Older users grew up with outputs that were often ambiguous, requiring judgment. When the model produces something that sounds plausible but is subtly wrong, older users are more likely to catch it, because they've spent decades catching subtle wrongness in human communication.
- They're not trying to look cool. This one is underrated. Young users have a social incentive to look like they know what they're doing. Older users don't. They ask dumb questions, they push the model to explain things twice, they ignore features they don't need. The result is that they actually learn the tool, while younger users perform fluency without building it. Put together, these advantages are structural, not incidental. They don't depend on the tool getting better. They depend on the user having a lifetime of practice that the tool rewards. The Evidence (Such as It Is) I want to be honest about the limits here. The empirical literature on AI proficiency by age is thin and mostly measures usage frequency, not proficiency. The studies that exist tend to show: Older users (65+) adopt more slowly but use more intensively when they adopt.
Older users report higher satisfaction with AI outputs on writing and research tasks, and lower satisfaction on tasks requiring fast visual processing (which is a UI problem, not a model problem).
Older users are more likely to use voice input, which produces longer and more natural prompts than typed input.
In one internal study at a major AI company that I've seen, users over 60 had the highest rate of "prompt refinement" per session of any age cohort.
The pattern is consistent, but the sample sizes are small and the selection effects are real. The elderly users who adopt AI are not a random sample of the elderly. They're the ones with the time, curiosity, and resources to try. That's a real limitation.
But the limitation cuts the other way too. The young users who are celebrated as "AI natives" are also not a random sample. They're the ones who happen to be on Twitter and TikTok, which is a different and equally biased sample. Both cohorts are selected. The question is which selection is more predictive of long-term proficiency, and I'd argue it's the one that includes patience and iteration.
What This Means for Product Builders
If the elderly are the deepest users, the industry is building for the wrong demographic, and it's leaving money and product quality on the table.
Concrete implications:
Onboarding is the wrong investment. Most AI products optimize for first-session activation. But the elderly cohort's value comes from long-term depth, not fast activation. Products that reward persistence (saved prompt libraries, version history, usage analytics) will retain this cohort far better than products that optimize for one-click wow.
Voice input is underbuilt. Older users use voice input at higher rates, and voice produces richer prompts. Most AI products treat voice as a second-class input. Investing in voice quality, transcription accuracy, and conversational refinement loops would disproportionately benefit the deepest users.
Long-session UX is missing. Most AI products are designed for short bursts. The elderly cohort uses the tool for hours at a time. They need features that make long sessions bearable: session saving, context memory, task resumption, and clean handoff between sessions.
Evaluation is the bottleneck. The elderly cohort's proficiency advantage is partly about evaluation, not generation. Products that help users evaluate outputs (source citations, confidence indicators, comparison views) will amplify the advantage. Products that hide evaluation will waste it.
If you're building an AI product and you're not testing it with users over 60 who've used it for months, you're missing the cohort that will tell you the most about where your product actually works.
The Contrarian Bet
Here's the bet I'd make if I were building in this space.
The next generation of premium AI products will not be built for the young and hip. They'll be built for the patient and experienced. The elderly cohort is the one with:
Time to iterate deeply
Money to pay for a product that genuinely helps
Domain expertise that makes the outputs more useful
Skepticism that makes them less likely to churn to whatever's trending
These are the four things that make a customer valuable over the long term. The young cohort has more of the things that make a customer valuable in the first month. The elderly cohort has more of the things that make a customer valuable in the first year.
The product strategy follows directly: stop optimizing for first-session activation and start optimizing for month-six retention of users who are getting better over time. The metrics are different, the features are different, and the cohort that benefits is different.
What To Actually Do
Three moves, concrete enough to start on Monday.
- Study the prompts of the patient. If you're doing product research, don't just interview the young power users. Find the elderly users who've been using your product for six months and ask to see their prompt histories. You'll learn more about how language models actually get used well than any usability study will tell you.
- Add persistence features before activation features. If you're building an AI product, add prompt library, version history, and session continuity before you add another onboarding flow. The deepest users don't need to be onboarded. They need to be supported in a habit that's already working.
- Test with the cohort you're not targeting. If your target demographic is 25–40, run a usability test with five users over 65 who've used AI for months. You'll find that half of your "delightful" features are noise, and half of your "minor" features are the ones they actually rely on. The findings will change your roadmap more than any A/B test. The Mental Model Here's the lens I want you to keep. AI proficiency is not a function of familiarity with technology. It's a function of domain expertise, iteration patience, and evaluation skill, and the elderly have more of all three than the young, on average. The industry's assumption that youth equals AI proficiency is a category error. Youth equals AI adoption. Adoption is not proficiency. Proficiency comes from time, patience, and practice, and those are precisely the resources that the elderly have in abundance. The interesting question is not whether the elderly will catch up with the young in AI use. It's whether the young will catch up with the elderly in AI mastery. And the answer, for now, is no. The grandparents are already ahead. What's your read? If you have older relatives or colleagues using AI, have you seen them develop proficiency that surprised you? And if you're an older user yourself, what have you figured out that you wish the industry understood? Drop it in the comments. I'm genuinely curious how widespread this pattern is.
