What should adults over 60 know when using AI for the first time?

What an AI assistant is, and why it can sound more certain than it is

The text box in Chrome may open Google Gemini, an AI overview in Search, or another service. ChatGPT, Microsoft Copilot, Claude, Gemini, and similar tools differ in features, privacy settings, prices, and whether a particular response used the web. The basic caution is the same.

A generative AI assistant predicts and produces useful language, images, code, or summaries from patterns in its training and, in some modes, from tools such as web search. It is good at reshaping information and carrying on a conversation. That makes it particularly useful for getting unstuck.

It does not independently know that a response is true, understand your full real-world situation, or owe you professional care. A polished answer, a numbered explanation, or a list of citations is not proof. NIST notes that generative systems can present false content and even false citations confidently. NIST AI 600-1, section 2.2

Some assistants can search the web and display citations. That can improve a current answer, but it does not remove the need to open the sources. OpenAI specifically advises users to check whether cited sources support a claim and to prefer authoritative sources when accuracy matters. OpenAI, “Searching the web with ChatGPT”

Think of it as a very fast research and writing helper that needs supervision. It can help you formulate a better question for an accountant, doctor, librarian, travel agent, or family member. It should not replace them where the consequences are meaningful.

Five low-risk ways to start, with prompts you can actually use

Start with a task where a wrong answer would be annoying rather than costly or dangerous. You can correct, revise, or ignore the result. These first tasks also make it easy to see what the tool does well.

These are examples, not commands. You remain the editor and decision-maker. For the letter and suspicious-message examples, redact names, addresses, policy or account numbers, barcodes, dates of birth, and other identifying details before pasting anything.

A small experiment that builds confidence

Ask something you already know well: a hobby, local history, a recipe you make often, or the rules of a game. Notice what the assistant gets right, what it oversimplifies, and how it reacts when you correct it. That is a safer way to develop an instinct for its strengths and weaknesses than starting with tax, health, or investments.

A useful prompt has four parts

You can write naturally. A prompt is simply the instruction you give the assistant. More context usually produces a more relevant answer, but only give context you are comfortable sharing.

For example, instead of asking, “What is the best cruise?”, try:

“I am choosing between a small-ship Alaska cruise and a train-and-hotel trip in June 2027. I value accessible cabins, quiet evenings, and not changing hotels often. Compare the trade-offs in a table. List only information I need to verify with official operators, not claims of current availability.”

The prompt does not need to be perfect. Follow-up is one of the main benefits: “Make that shorter,” “explain that term,” “show the assumptions,” “give the other side,” or “what information would change your recommendation?”

Do not mistake a detailed explanation for a verified calculation. If you ask a tax question, say: “Explain the general issue and point me to the relevant tax authority publication. Do not calculate my return or give personal tax advice.” Then verify against the current authority in your country and, where needed, a qualified tax professional.

How to check an AI answer before relying on it

Use a lighter check for a dinner recipe and a stronger one when money, health, rights, deadlines, identity, or safety are involved.

flowchart LR
  A[Ask with non-private context] --> B[Read AI response]
  B --> C{Could a mistake affect money, health, rights, identity, or safety?}
  C -->|No| D[Review, edit, and use with judgment]
  C -->|Yes| E[Open primary or official sources]
  E --> F{Sources support the exact claim and are current?}
  F -->|Yes| G[Ask the appropriate professional or service if a personal decision remains]
  F -->|No or unsure| H[Do not act yet. Seek a reliable human or official source]

Visual brief: This flowchart is the practical “confidence checkpoint.” It shows that the same AI response can be useful for a low-stakes draft but insufficient for a consequential decision.

The five checks

1. Identify the exact claim. “It says my pension payment is taxable” is more checkable than “it gave me a lot of useful information.” Separate facts, assumptions, and recommendations.

2. Ask for sources and dates. Say, “Give me links to the primary sources and tell me the date of each.” A citation list is only a lead. Open it yourself.

3. Prefer the source that owns the fact. For a government deadline, use the government agency. For a product warranty, use the manufacturer. For a train time, use the operator. For health information, use a public-health agency or clinical organisation, then discuss personal questions with your clinician.

4. Check currency and fit. A good article from 2022 may not answer a 2026 rule. A correct general rule may still not apply to your income, medication, country, booking class, or contract.

5. Look for what would change the answer. Ask, “What assumptions did you make?” and “What facts could make this wrong?” This turns a polished response into a starting point for real research.

If the answer names a study, quotation, court case, regulation, statistic, or web page, verify that it exists and says what the assistant claims. OpenAI explicitly warns that language models can make up facts, quotes, studies, citations, and references. OpenAI, “Does ChatGPT tell the truth?”

Using two AI systems to compare wording can occasionally reveal a problem, but it is not fact-checking. Two systems can repeat the same outdated or invented claim. An independent primary source is better.

Privacy: decide what stays out of the chat box

Treat a public or consumer AI chat as a service with its own data rules, not as a locked personal notebook. Read the privacy and data-control page for the exact assistant you use before adding a file, turning on memory, connecting an email or storage account, or letting the tool take actions.

For example, ChatGPT offers controls to turn off use of new chats for model improvement and a Temporary Chat option, but Temporary Chats may still be retained for up to 30 days for safety purposes. OpenAI Data Controls FAQ OpenAI Temporary Chat FAQ Gemini’s privacy guidance likewise says connected apps can expose data to third parties and that its output is informational rather than professional advice. Google Gemini Apps Privacy Hub

Do not upload these to a general-purpose AI assistant

• Passwords, passphrases, one-time codes, recovery codes, security-question answers, or photos of identification.
• Bank, credit-card, brokerage, pension, tax, medical-record, insurance, or utility account numbers.
• A full passport, driver’s licence, Social Security number or national ID number, birth certificate, tax return, or unredacted medical record.
• A document containing another person’s private information unless you have a clear right and need to share it. This includes family group chats, client information, and school or work records.
• A live location, home security details, travel itinerary paired with an address, or a photo that reveals more than you intend.
• Anything you would not be comfortable being stored by a service provider or seen by a person reviewing a support request.

Redacting means removing the sensitive content before upload, not merely asking the AI to ignore it. Use placeholders such as [insurer], [policy number], and [date]. If you need help interpreting a document with confidential facts, a trusted human professional or the organisation that issued it is usually the safer route.

Spotting AI-assisted scams and impersonation

AI makes it easier for criminals to produce fluent messages, cloned voices, convincing fake photos, and plausible websites. It does not make every odd-looking message fake, and no single visual clue can reliably prove whether media is real. Urgency, secrecy, and an unusual payment request are much more useful warnings than trying to judge whether a voice “sounds real.”

The U.S. Federal Trade Commission warns that a criminal can clone a loved one’s voice from a short audio clip. For an emergency call, do not trust the caller’s voice alone. Hang up, then call the person on a number you already have, or check through another family member or friend. Requests for gift cards, cryptocurrency, wire transfers, or a code read aloud are especially strong warning signs. FTC, “Scammers Use Fake Emergencies To Steal Your Money”

Stop, verify, and only then respond

The SEC, FINRA, and NASAA warn that fraudsters use AI claims and deepfake audio or video to lure investors. They advise confirming registration and underlying sources rather than relying on AI-generated investment information. Investor.gov, “Artificial Intelligence (AI) and Investment Fraud”

An AI assistant can help you make a checklist of questions about a suspicious message, but it cannot authenticate the sender. It also cannot safely replace the independent call or login that resolves the question.

Health, financial, tax, and legal questions: use AI to prepare, not decide

These are not off-limits subjects. They just require a different role for AI.

Health

It can help turn medical jargon into plain language, prepare questions for an appointment, or make a symptom diary template. Do not use a general chatbot to diagnose a condition, choose or change treatment, decide whether symptoms are urgent, interpret test results in place of a clinician, or upload full medical records.

The World Health Organization urges caution with large language models in health because safety, autonomy, and privacy need protection. WHO, “Safe and ethical AI for health” For health information online, MedlinePlus recommends checking who runs a site, its purpose, funding, date, and sources, then discussing information with a provider before relying on it. MedlinePlus, “Online health information: what can you trust?”

If symptoms may be an emergency, contact local emergency services or urgent medical care. Do not wait for a chat response.

Money, tax, investments, and legal matters

It can explain terms, create a question list, or compare the wording of options you already have. It should not choose investments, calculate your tax return, create a legal plan you sign without review, or tell you that a decision is safe for your particular circumstances.

For investments, do not act on a chatbot’s stock idea, market forecast, or claim that an “AI trading system” guarantees returns. Use the relevant regulator’s registration lookup and a registered professional where advice is needed. The SEC’s warning is direct: AI-generated information may be inaccurate, incomplete, misleading, or wholly made up. Investor.gov AI investment fraud alert

For a tax, benefit, estate, or contract question, ask AI for a plain-language overview and a list of questions. Then consult the current official authority and, if the decision is personal or substantial, a qualified adviser licensed in your jurisdiction.

Make the experience easier to read, hear, and control

Accessibility is a normal part of setting up a useful tool, not an afterthought. Use the settings that help you read comfortably and reduce fatigue. For a long reply, ask the assistant: “Use 14-point-like short paragraphs, headings, and a five-item checklist,” or “Read this back to me in simple language, one step at a time.” The exact feature names vary by device and assistant.

• In Chrome on a computer, enlarge a page with Ctrl + + on Windows or Linux, or ⌘ + + on a Mac. Chrome can also remember a different zoom level for a site. Google Chrome Help: change text, image, and video size
• Chrome’s Reading mode can simplify supported webpages, adjust text, and read text aloud. It does not work on every type of page, such as many forms and interactive content. Google Chrome Help: Reading mode
• Chrome on Windows and Mac has Live Caption settings, and captions can be customised for readability. Google says the audio and captions for Chrome’s Live Caption are processed locally on the device. Google Chrome Help: captions and translations
• If you use a Chromebook, ChromeVox is its built-in screen reader. Google Chrome Help: ChromeVox
• Voice input can be useful, but pause before dictating personal information. Review the transcribed text before sending because voice recognition can mishear names, numbers, medication names, and addresses.

If a friend, family member, librarian, or trainer is helping, let the person decide what to ask and what data to share. The useful help is setting up a comfortable display, demonstrating a low-risk task, and practising verification, not taking over the conversation.

A calm first-session checklist

1. Pick one familiar, low-stakes task from the table above.
2. Check which assistant you are using and whether you are signed in. Find its privacy, history, memory, and connected-app settings.
3. Do not connect email, cloud storage, or financial apps merely to experiment.
4. Write a short prompt with a goal, relevant non-private context, output format, and a request to flag uncertainty.
5. Ask one follow-up question to improve the result.
6. For any claim that could affect money, health, deadlines, safety, or reputation, open an official or primary source and check it.
7. If a message, call, or video asks you to act urgently or secretly, stop and independently verify the sender before doing anything.

There is no need to use AI for everything. Ordinary search, official websites, a phone call, a local library, a trusted friend, or a professional may be faster and better for a particular task. The goal is not to become an “AI person.” It is to add one useful tool to the ways you already think, learn, and make decisions.

Limits and when another option is better

AI can misunderstand a question, reflect bias in its sources, miss a recent change, and fabricate a confident answer. It may not know your location, eligibility, accessibility needs, or personal priorities unless you provide them, and providing them can create privacy risk. It can be helpful company for conversation, but it is not a reciprocal relationship, crisis service, clinician, lawyer, accountant, or fiduciary.

Choose a different route when:

• You need an official answer, a binding decision, or an account-specific action. Go directly to the organisation or professional responsible.
• You need current local facts, such as a train cancellation or medical appointment. Check the provider’s current site or call.
• You are upset, rushed, or being pressured to send money or information. Pause and bring in a trusted person or independent contact channel.
• You need confidential help. Use a service designed and authorised to handle that information, rather than a general chatbot.