AI for Older Adults

AI for Older Adults

by Alvin Blackshear | Historian & Researcher

Artificial intelligence tools like ChatGPT, Claude, Gemini, Perplexity, and Copilot have moved from novelty to necessity faster than almost any technology in memory. For adults over 60, that speed can feel less like an invitation and more like being left behind at the station. But research on how older adults actually use, and struggle with, these tools tells a more encouraging story: the gap isn't about ability, it's about approach. A 2026 study of older adults' technology support requests found that when queries were rewritten to add missing context and clearer language, the accuracy of AI-generated solutions jumped from 46% to 69%, and related Google search results improved from 35% to 69% accuracy. In other words, the tools already work well enough. The skill worth learning is how to talk to them.

This article walks through four practical steps: choosing the right AI for your needs, writing prompts that get better answers, understanding what AI "agents" are and when to trust them, and avoiding the mistakes that trip up new users of every age, but especially those who didn't grow up parsing tech jargon.

Choosing the Right AI

There is no single "best" AI assistant. Each is tuned for different strengths, and picking the right one for the task at hand saves time and frustration. A simple way to think about it:

·   Best for everyday questions: ChatGPT is the most general-purpose assistant, well suited to answering quick questions, drafting emails, or explaining a topic in conversation.

·   Best for writing and long documents: Claude tends to produce clear, well-organized long-form writing and is a strong choice for letters, essays, or reviewing lengthy documents.

·   Best for web research: Perplexity is built around search, pairing its answers with cited sources so you can click through and verify what it tells you, which is useful when accuracy matters most.

·   Best for Google users: Gemini integrates directly with Gmail, Google Docs, and Google Search, which makes it a natural fit if that's already your everyday ecosystem.

·   Best for Microsoft 365 users: Copilot is built into Word, Outlook, and Excel, so if your family or workplace runs on Microsoft software, it can work inside the programs you already know.

None of these choices is permanent or exclusive. Many people end up using two: one for quick daily questions and another for anything they want double-checked with sources. The point isn't to find the "correct" one, but to match the tool to the task, the same way you'd choose a phone call over a text message depending on what you're trying to accomplish.

Learning to Write Better Prompts

The single biggest lever an older adult has for getting better answers from AI isn't a special app or setting. It's the wording of the question itself. Research on communication between older adults and AI systems identified four recurring patterns that reduce answer quality: verbosity (explaining too much surrounding detail), incompleteness (leaving out a key fact the AI needs), over-specification (adding assumptions that aren't actually true), and under-specification (being too vague about what's wanted). The good news from that same research is that these patterns are fixable, not through retraining how you think, but through a few reliable habits when you type or speak your question.

Ask for plain English. AI models default to whatever register they think fits the question, and that can drift into jargon. Simply adding "explain this in plain English, no technical terms" or "explain it like you would to a smart friend who isn't a computer person" changes the entire tone of the reply.

Request step-by-step explanations. Instead of asking "how do I attach a photo to an email," try "walk me through this one step at a time, and wait for me to say 'next' before moving to the next step." This breaks a wall of instructions into a manageable back-and-forth, and it also gives you a natural point to ask a follow-up question without losing the thread.

Ask for examples. Abstract instructions are harder to follow than concrete ones. Adding "give me a real example" or "show me what that would actually look like on my screen" turns a vague description into something you can match against what you're actually seeing.

Ask the AI to critique its own answer. This is a lesser-known but powerful technique: after receiving an answer, ask "is there anything wrong with this answer, or a simpler way to do it?" AI models are often able to catch their own errors or oversimplifications when prompted to double-check themselves, and this habit builds in a layer of quality control you don't have to supply yourself.

The underlying finding from the research is worth repeating: rephrased, well-specified questions were understood far better by others (93.7% versus 65.8% comprehension in one comparison) and led to solutions people felt confident following. A better-written question isn't a nicety. It's the difference between a correct answer and a confusing one.

Using AI Agents

You may have started to hear the term "AI agent," distinct from a regular chatbot. It's worth understanding the difference, because agents come with real convenience and real risk.

What an agent is. A standard AI chat answers a question and stops. An agent is designed to take a goal and carry out a sequence of actions on its own (searching the web, filling out a form, booking something, or working through a multi-step task) before reporting back. Industry analysis on the state of AI agents in 2026 notes that the defining shift in newer systems is their ability to handle progressively longer and more complex tasks, maintaining context across many steps rather than answering a single quick question.

When to use one. Agents are useful for tasks that are repetitive, well-defined, and low-stakes if something goes slightly wrong, for example, summarizing a folder of documents, comparing prices across several websites, or drafting a set of similar emails. They save the most time exactly where a task has many small, mechanical steps that would otherwise take you an hour of clicking around.

Risks of autonomous agents. The tradeoff for that convenience is reduced oversight. Because an agent acts across many steps without checking in at each one, a small misunderstanding early on can compound into a larger error by the end, the AI equivalent of a wrong turn early in a road trip that takes you far off course before anyone notices. This matters most for anything involving money, personal information, medical decisions, or sending communications on your behalf. Never give an agent standing access to your bank accounts, passwords, or ability to make purchases without a manual confirmation step at the end.

Practical everyday examples. Reasonable, lower-risk uses include: asking an agent to research and summarize options for a specific purchase (with you making the final decision), having it organize digital photos or files, or using it to draft, but not send, a batch of similar messages, such as holiday cards or appointment reminders, for you to review individually before anything goes out.

Avoiding Common Mistakes

Even with the right tool and better prompts, a handful of habits cause most of the frustration people report with AI, regardless of age.

Trusting AI too quickly. AI models can sound confident even when they're wrong. They occasionally generate plausible-sounding but incorrect information, a phenomenon often called "hallucination." Treat a confident tone as no guarantee of accuracy, especially for anything involving health, legal, or financial specifics.

Asking vague questions. As the communication research above showed, vague or incomplete questions are the most common source of poor answers. If a response feels off-target, the fastest fix usually isn't switching tools. It's adding the missing context and asking again.

Accepting the first answer. The first response an AI gives is a starting point, not a verdict. Asking a follow-up like "are you sure?" or "what would make this answer wrong?" often surfaces caveats or corrections the first answer left out.

Failing to verify important facts. For anything that matters, such as a medication interaction, a legal deadline, or a financial figure, cross-check the AI's answer against a trusted human source, such as your doctor, pharmacist, financial advisor, or a reputable website. Tools like Perplexity, which show their sources directly, make this step easier by letting you click through and confirm the claim yourself rather than taking it on faith.

The Bigger Picture

None of this requires becoming a technical expert. The research is fairly consistent on this point: older adults who use AI report high confidence and ease once they receive well-matched, clearly explained answers. The 2026 diary study found 89.8% of older adult participants felt able to answer AI's follow-up questions, and 94.7% felt able to follow the solutions they were given. The barrier isn't capability; it's a mismatch in how questions get asked and answered. Broader research on aging and generative AI echoes the same conclusion, pointing to real promise in health and cognitive support as long as tools are built and used thoughtfully, with attention to privacy and genuine ease of use. Choose a tool that matches the task, ask questions the way outlined above, use agents cautiously and only for low-stakes chores, and verify anything that truly matters. That combination, not fluency in tech jargon, is what determines whether AI becomes a genuinely useful assistant in daily life.

___________________________ 

Sources:

Dhanuka, N. (2026, February 26). State of AI agents 2026: Autonomy is here. Prosus. https://www.prosus.com/news-insights/2026/state-of-ai-agents-2026-autonomy-is-here

Ernst & Young Global Limited. (2026, March). Understanding older generations' adoption of AI [Ripples Research Report]. https://www.ey.com/content/dam/ey-unified-site/ey-com/en-gl/about-us/corporate-responsibility/documents/ey-gl-how-older-generations-are-engaging-with-ai-03-2026.pdf

Liu, T. (2026, May 29). The role of multimodal generative AI in older adults' health management: Systematic scoping review. JMIR AI. https://ai.jmir.org/2026/1/e84695

Shomee, H. H. (2026, January 15). Empowering older adults in digital technology use with foundation models. arXiv. https://doi.org/10.48550/arXiv.2601.10018