Asking an AI chatbot about money, health or government.

What chatbots are good and bad at

A chatbot writes fluent, confident answers in everyday language. Fluent is not the same as correct.

An AI chatbot is a computer program you type or talk to, which answers in ordinary sentences. People use them to put a hard word or a long letter into plainer language, to list the questions they want to ask before calling an office or seeing a doctor, or to start a draft they then finish themselves. Those uses have something in common: you can see a mistake and fix it before it costs you anything.

Where they go wrong

• They can be wrong and sound sure. The UK National Cyber Security Centre (NCSC) says these tools can get things wrong and present incorrect statements as facts, a flaw known as hallucination. New Zealand’s Privacy Commissioner says they often produce very confident errors of fact or logic, and should not be relied on without checking.
• They learn from the internet, mistakes included. The NCSC explains that the models behind them are trained on very large amounts of text, typically taken from the internet. There is too much to filter, so incorrect material is likely to be included.
• Leading questions get leading answers. The NCSC also says they can be biased and are often gullible when responding to leading questions. “What does the official guidance say about…?” is a better question than “Isn’t it true that…?”
• Money answers are no exception. The US Consumer Financial Protection Bureau (CFPB) found that chatbots sometimes get the answer wrong, and that the statistical methods behind generative chatbots are not well placed to tell correct information from incorrect information.

What not to type in

What you type may be stored, used to train the system, or shown to someone else.

• Passwords, PINs and one-time codes. Australia’s Scamwatch says never to share passwords or one-time codes with anyone who contacts you unexpectedly, even if they claim to be from your bank. A chat window that asks for one is a warning sign: the FBI says criminals put AI chatbots on fake websites to get people to click harmful links.
• Account and card numbers. Bank, card, loan and pension numbers, and the security questions that go with them.
• ID numbers. Social Security, tax, passport, driver’s license, health card or national insurance numbers.
• Full medical records. Test results, letters or records with your name, date of birth or record number on them.
• Other people’s details. Canada’s Privacy Commissioner says not to share personal information about other people, and never to share pictures or personal information of children.

The Office of the Privacy Commissioner of Canada warns that what you type may be collected and stored, used to further train the system, used to create profiles about you or make decisions about you, and could be disclosed to other users. It suggests limiting what you share and changing identifying details, for example by using a made-up name.

It is the same rule our scams and fraud session uses: nobody is ever asked to read out a code, a password or account details, even as a demonstration.

How to check an answer

Treat the answer as a starting point. The official source is the answer.

1. Find the official site yourself. Type the address of your government, regulator, bank or clinic, or find it through your country’s main government website, rather than following a link from a chatbot or a message. Scamwatch advises typing web addresses directly into your browser, and warns that scammers use AI tools to make copies of real websites. Our free help by country pages list official, free services we have checked.

2. Look for the date and the country. Rules, rates and deadlines differ by country, and often by state, province or council, and they change. MedlinePlus, from the US National Library of Medicine, advises checking how old information is, because even trusted sites can have out-of-date pages.

3. Ask for sources, then open them. Ask the chatbot where its answer comes from. Open each link and check that the page exists, that it is official, and that it says what the chatbot said. If you can’t find it, don’t rely on it.

4. Read laterally. Leave the page to find out who is behind it and what other reliable sources say. Elections Canada suggests finding the original source and comparing it with other sources; MedlinePlus says not to rely on just one website. It is the habit practiced in our Information you can trust session.

5. When it matters, ask a person. A pharmacist, your doctor’s office, your bank on the number on your card, or the agency’s own helpline can answer for your situation.

Money questions

Rates, fees and offers change, and a wrong answer about money can cost you.

• Check figures at the source. Interest rates, fees, benefit amounts and tax rules change and differ between providers and countries. An answer may not say how current it is. Check on the provider’s or regulator’s own website, or in the official documents for the product, before you act. The CFPB notes that people ask chatbots for recommendations on credit cards, bank accounts, mortgages and loans, even though chatbots can be wrong.
• A general chatbot is not your bank. To stop a payment or report fraud, call your bank on the number on your card or its official website.
• Scams now use AI voices. The US Federal Trade Commission (FTC) warns that a scammer can clone a family member’s voice from a short audio clip, perhaps taken from something posted online. Its advice: don’t trust the voice; call the person on a number you know is theirs. Scamwatch says a few seconds of a recording can be enough.
• And AI messages, photos and videos. The FBI says criminals use AI to write believable messages without the spelling and grammar mistakes that used to give them away, to make fake profile photos and identity documents, and to make videos of public figures promoting investments.
• Agree a family code word. The FBI suggests creating a secret word or phrase with your family to check it is really them.
• Watch how they want to be paid. The FTC says a request to wire money, send cryptocurrency, or buy gift cards and give the numbers could be a sign of a scam.

More in Money basics: spot a scam, including the first hour if it has already happened.

Health questions

A chatbot can help you prepare questions for a health professional. It can’t examine you.

• Emergencies come first. If someone may be seriously ill or injured, call your local emergency number, not a chatbot. In England, for example, the NHS says 999 is for life-threatening emergencies such as strokes and heart attacks, and NHS 111 can check your symptoms and tell you what to do. Many countries have a free health advice line.
• Don’t change medicines on a chatbot’s say-so. The US Food and Drug Administration advises not to skip doses or stop a prescribed medicine without first talking to your health care professional, even if you feel better, and says you shouldn’t choose or change a dose yourself.
• Take it to your clinic or pharmacist. MedlinePlus says online health information can never replace a talk with your provider, and suggests printing what you have read and bringing it to your appointment.
• Keep it general. You can ask about a condition or a medicine without giving your name, date of birth, record number or test results.

More in Finding and paying for care: health scams.

Government and elections

Rules and deadlines are local. An answer about another country, state or year can look exactly right.

• Check deadlines where they are set. Registration deadlines, polling places, benefit rules and application dates differ by country, region and council, and can change. Check your election office or local government website. In the United States, USAGov lists state and local election offices.
• For voting, only use your official election office. Elections Canada describes itself as the official source of information on how federal elections work, and warns that fake images, videos, text and sound are easy to create and can be impossible to tell apart from real content.
• Stop and consider. The Australian Electoral Commission warns that AI-generated text, images, video and audio can be used to deceive people about where to vote or how to complete a ballot paper. Its advice: check the content and the source, and if you’re unsure, ignore it and don’t share or repost it.
• Messages that pretend to be official. In October 2026 the FBI warned that scammers use AI tools to make realistic voices, images and messages impersonating candidates, campaigns and government bodies, and described texts falsely telling people they are not registered to vote, with a link to a fake registration page.

This guide takes no side on any party, candidate or ballot question. More in Taking part in local government: information you can trust.

Privacy settings and deleting chats

Every product is different, and settings change. These are the questions to ask of any of them.

• Find the privacy settings. Canada’s Privacy Commissioner suggests reviewing and adjusting them, and using a strong password. Where the account offers two-step sign-in, it protects your chat history too.
• Delete chat history if you can. The same office suggests deleting your chat history if the chatbot allows it.
• Read what happens to your chats. It also warns that platforms may not clearly explain how personal information is used, stored or shared. Look in the privacy policy for how long chats are kept, whether they are used for training, and what deleting does. New Zealand’s Privacy Commissioner notes the risk that what is typed in is kept or disclosed by the provider and used to keep training the model.
• On a shared computer, such as at a library, sign out when you finish.
• Check again after updates. Settings and policies can change when a product changes.

A one-screen checklist

Print it, or keep it on your phone.