Where to start: can I trust what AI says?
The short answer: trust it the way you would trust a confident helper who sometimes gets things wrong. It is good for a first draft and for explaining things. For anything about health, money, or the law, check with a real person or an official source before you act. Keep private numbers out of the chat. And for photos, videos, and voices, check where they came from rather than trusting your eyes and ears.
Why it can sound sure and still be wrong
A chatbot is built to give a helpful-sounding answer. It does not always know when it is guessing. When it states something false in a confident voice, people call it a “hallucination.” It also tends to agree with you. Researchers at Duke say people like chatbots that agree with them, so a chatbot may not push back, even when it should.
This matters most for health. In one 2026 study of popular chatbots, about half of the answers to 50 medical questions were rated “problematic.” Chatbots can also invent sources: in one study, only about a third of the citations they gave were accurate, and nearly half were at least partly made up.
How to check an answer in a minute
- Ask how sure it is. Use the example below.
- Ask for a source, then open it yourself. Do not assume the link is real or says what the chatbot claims.
- Compare with a trusted site. For health, think of your doctor's office or Medicare.gov. For money or taxes, think of the bank, the IRS, or a licensed professional.
- For anything big, ask a person. AI can help you prepare good questions to ask them.
“How sure are you about that? What might be wrong or missing? Please tell me a trusted source I can check myself.”
Health: helpful for questions, not for decisions
Many older adults are already asking AI health questions. In AARP's 2026 survey, about two in five adults 50 and older who had used AI had asked it a health question, yet 69% were not comfortable sharing personal health information with an AI tool. They are right to be careful. For symptoms, medicines, and new or worsening problems, call the doctor's office. See our guide on writing a message to the doctor's office.
Also, six in ten adults 50 and older did not know whether their health care provider uses AI. It is fine to ask your doctor's office.
Privacy: what to keep out of the chat
- Never type Social Security, Medicare, bank, or card numbers, or passwords.
- Leave out names and birth dates when you do not need them.
- Check one setting. In ChatGPT, open Settings, then Data controls, and look at Improve the model for everyone. Other chatbots have a similar privacy or data setting, usually under Settings. Our ChatGPT guide shows where to find it.
Worry about personal data is common. In Pew's 2026 survey, 71% of U.S. adults said they expect AI to make their personal information less secure.
The box at the top of search results
Many search engines now show a short answer written by AI above the usual results. Pew found that six in ten U.S. adults read these. They can be wrong too. For anything that matters, scroll down and open a source you recognize.
Real or fake?
AI can now make fake photos, videos, and voices. A Quinnipiac University poll found that 42% of people were not confident they could spot AI-generated video or audio. You are not behind if you cannot tell. Nobody can, every time.
- Clues that sometimes show up: odd blinking, lips that do not match the sound, missing reflections, or objects that bend or vanish in the background. These clues are becoming less reliable as the technology improves.
- The stronger habit is to check the source. Who posted it? Is the same story on news sites you know? Search for it yourself.
- Treat famous people promoting investments as a red flag. This is a common deepfake scam.
- Around elections, fake videos, images, and robocalls can spread fast. Check with trusted news or your local election office before you believe or share.
Many older adults want to know when something was made by AI. In a University of Michigan poll, nine in ten did. It is a good instinct. When in doubt, ask who made it.
Quick questions
Is AI more wrong than a web search? It is different. A web search shows you where information comes from. A chatbot gives one smooth answer, which can hide mistakes. That is why checking matters.
Can I use it for a first draft? Yes. Notes, messages, and plans are good uses, as long as you read it over and make it your own.