AI chatbots are useful, but whatever you type into them can be stored, read by people and sometimes used to train the next version, and the answers they give can be confidently wrong. Tools such as ChatGPT, Google Gemini and Microsoft Copilot are now part of everyday life: Ofcom reports that ChatGPT had 1.8 billion UK visits in the first eight months of 2025, up from 368 million in the same period of 2024. This guide is for anyone who uses a chatbot at home, for school or in a very small business. It explains in plain English what to keep out of a chatbot, which privacy settings to change, how to check answers, how criminals use AI against you, and what to do if something goes wrong.
Key takeaways
- Treat a chatbot like a stranger with a notebook. The UK National Cyber Security Centre (NCSC) warns that what you type is visible to the company running the service, is stored, and may be read by its staff or contractors.
- Never type passwords, bank details or other people's private information. If it would cause harm if it became public, leave it out.
- Change two settings today. Turn off the option that lets your chats train the model, and set your chat history to delete itself after a short period.
- Check anything that matters. Chatbots can "hallucinate", which means they make up facts that sound convincing. Check health, money and legal answers with an official source or a professional.
- Expect AI-powered scams. Criminals use AI to write better scam messages and to copy the voices of people you love. Slow down and check through a number you already know.
What happens to what you type
When you type a question into a chatbot (this question is often called a "prompt"), it travels to the company's computers. It does not stay on your phone. The NCSC points out that your questions are stored, that they could be read by the provider, and that they could be leaked, hacked or made public by mistake. It also notes that many small questions from the same account can be combined to build a picture of you over time.
The details differ by service. These are the defaults the companies describe on their own help pages, checked in October 2026. Menus and wording change often, so look for the closest match in your app.
| Service | Setting to look for | What the company says |
|---|---|---|
| ChatGPT (OpenAI) | Settings > Data Controls > "Improve the model for everyone" | Turn it off to stop your chats being used for training. "Temporary Chat" is not used for training and is deleted within 30 days. Data Controls also lets you export your data or delete your account. |
| Google Gemini | "Keep Activity" (Gemini Apps Activity) | Chats are auto-deleted after 18 months by default; you can choose 3 or 36 months. A sample of chats is reviewed by people, and reviewed chats can be kept for up to three years even if you delete your activity. With Keep Activity off, chats are still kept for 72 hours. |
| Microsoft Copilot | Privacy controls: model training and conversation history | Signed-in users can opt out of model training for future chats. You can delete single conversations or your whole history. Conversations are stored for 18 months by default. |
Two points stand out. First, turning off training does not delete what you have already shared. Second, Google says plainly that you should not enter confidential information you would not want a human reviewer to see. Assume the same is true of any chatbot.
What never to share
A simple test: if a stranger found it on a park bench, would it hurt you or someone else? If yes, do not type it. Get Safe Online gives the same advice: assume anything you enter may be stored and used in some way.
- Passwords and codes: no chatbot needs your password, PIN or a one-time code. If one asks, stop.
- Money details: card numbers, bank account numbers, screenshots of statements, or anything with your National Insurance or Social Security number.
- Identity documents: photos of passports, driving licences or utility bills. These are exactly what fraudsters need.
- Health and family details: you can ask general health questions, but leave out names, dates of birth, addresses and test results that identify you or a relative.
- Other people's information: do not paste in a friend's message, a child's school report or a customer's details. They have not agreed to share it.
- Work secrets: if you run a small business, keep customer lists, contracts and supplier bank details out of free chatbots. For organisations, see the Generative AI Data Loss Prevention Guide.
If you need help with a document that contains personal details, replace them first: "Customer A", "my bank", "[address]". The chatbot can still help with the wording.
Set it up safely in 10 minutes
- Use the real app. Download chatbot apps only from the official Apple App Store, Google Play or the company's own website. Copycat apps and fake "AI" browser extensions exist and can steal data.
- Protect the account. Sign in with a strong, unique password (three random words works well) and turn on two-step verification. See the Two-Factor Authentication for Everyone Guide.
- Switch off training. Find the model training setting from the table above and turn it off.
- Shorten history. Set auto-delete to the shortest option, or use a temporary or private chat mode for anything personal.
- Review "memory" and personalisation. Some chatbots remember facts about you between chats. Look in settings for memory or personalisation, check what it has saved and delete anything you would not want kept.
- Check connected apps. Some chatbots can link to your email, calendar or files. Only connect what you really need, and remove connections you no longer use.
- Read the privacy page once. The NCSC recommends reading the terms and privacy policy before asking anything sensitive. Look for who can read chats and how long they are kept.
How to check an answer
Chatbots predict words that sound right. That is not the same as being right. The NCSC notes that they can be biased, can be led by the way you phrase a question, and can give answers that are partly or wholly wrong while sounding certain.
- Ask for sources, then open them. Chatbots sometimes invent links, book titles or quotes. If the page does not exist or does not say what the chatbot claims, do not rely on it.
- Use official sites for high-stakes topics: the NHS website for health, GOV.UK or HMRC for tax and benefits, your bank's own website for money, Citizens Advice for consumer rights.
- Be wary of instructions to change security settings. If a chatbot tells you to turn off antivirus, install a program or run a command you do not understand, stop and check with someone you trust.
- Do not let it make the final decision. Use it to draft, explain and brainstorm. Make payments, medical choices and legal decisions yourself, with proper advice.
- Watch out for hidden instructions. If a chatbot can read web pages or your emails for you, a booby-trapped page can include hidden text that tries to steer it (this is called "prompt injection"). Be cautious when a chatbot summary suddenly pushes you to click a link or call a number.
Signs an AI is being used against you
The same tools that help you write a letter help criminals write scams. The NCSC warns that chatbots can help attackers write more convincing phishing emails in many languages. Typos are no longer a reliable warning sign.
- A loved one's voice asking for urgent money: the US Federal Trade Commission (FTC) warns that criminals can copy a voice from a short clip posted online. Hang up and call the person on a number you already have, or check with another family member.
- Pressure and secrecy: "Do not tell anyone", "pay in the next hour" and "buy gift cards and read me the numbers" are scam signs whatever the voice or video looks like.
- Payment by gift card, cryptocurrency or bank transfer: these are hard to reverse, which is why criminals ask for them.
- Video calls and celebrity adverts: faces can be faked too. See the Spotting Deepfake Scams Guide.
- Fake "AI investment" schemes: adverts promising guaranteed returns from an AI trading bot are a common hook. See the Romance and Investment Scam Awareness Guide.
A family code word, agreed in person and never posted online, gives everyone a quick way to check a panicked call.
Children and chatbots
Many children now use chatbots for homework and for company. Get Safe Online points out that leaning on AI for essays can leave young people unprepared for exams and later study.
- Check the age rules: each service sets a minimum age in its terms. Check them before setting up an account for a child.
- Talk about sharing: explain that a chatbot is not a friend keeping a secret. Real names, school, address and photos stay out.
- Stay curious, not cross: ask what they use it for and what it has said. Chatbots can produce upsetting or wrong content, and children should feel able to tell you.
- Use family controls: phone and console parental controls can limit app downloads. See Protecting Children Online: A Parent's Guide.
What to do if something goes wrong
- You shared a password or code: change that password straight away, on every site where you reused it, and turn on two-step verification. Then delete the chat.
- You shared bank or card details: call your bank using the number on the back of your card. In the UK you can also call 159 to reach most major banks.
- You shared ID documents: watch for letters about accounts you did not open. See the Identity Theft Recovery Guide.
- You want your data removed: delete the chats, then use the company's privacy or data request form. In the UK and EU you have the right to ask what a company holds about you and, in many cases, to have it erased. The Information Commissioner's Office (ICO) explains these rights.
- You lost money to a scam: tell your bank first, then report it (see below).
Where to get help
- UK, scam emails: forward them to [email protected], the NCSC's Suspicious Email Reporting Service.
- UK, fraud victims: in England, Wales and Northern Ireland report to Report Fraud at reportfraud.police.uk or 0300 123 2040. In Scotland call Police Scotland on 101.
- UK, advice: Get Safe Online for everyday safety tips, Citizens Advice for consumer help, and the ICO for data protection rights.
- US: report scams to the FTC at ReportFraud.ftc.gov, and identity theft at IdentityTheft.gov. Online crime can also be reported to the FBI's Internet Crime Complaint Center (IC3) at ic3.gov.
Common failure patterns
- Pasting the whole document: people paste an entire letter, bank statement or medical report when only one paragraph was needed. Cut it down and remove names first.
- Thinking "delete" means gone at once: as Google's own page shows, reviewed chats can be kept for years and chats are held briefly even with history off. Do not share it in the first place.
- Trusting a confident answer: made-up legal cases, wrong medicine doses and invented tax rules all come out in the same calm tone as correct answers.
- Reusing a weak password: a chatbot account holds a long record of your questions. If its password is reused elsewhere and leaks, someone else can read your history.
- Believing a familiar voice: families have sent money after a call that sounded exactly like a grandchild. Always call back on a known number.
- Connecting everything: giving a chatbot access to all your email and files "just in case" means one mistake or one tricked summary can expose far more.
Practitioner checklist
Work through this list once, then repeat it every few months or when you start using a new chatbot.
- I installed the chatbot from an official app store or the company's own website.
- My chatbot account has its own strong password and two-step verification.
- I have turned off model training in settings.
- I have set chat history to auto-delete, or I use temporary chats for anything personal.
- I have checked what the chatbot "remembers" about me and deleted what I do not want kept.
- I never type passwords, codes, card numbers, ID documents or other people's private details.
- I check health, money and legal answers on an official website or with a professional.
- My family has a code word for emergency calls, and we always ring back on a known number.
- My children know what not to share with a chatbot and that they can tell me if something upsets them.
- I know where to report a scam: [email protected] and Report Fraud in the UK, ReportFraud.ftc.gov in the US.
Standards and references
- NCSC: ChatGPT and large language models: what's the risk?
- NCSC: Reporting a scam email
- OpenAI Help Center: Data Controls FAQ
- Google: Gemini Apps Privacy Hub
- Microsoft Support: Privacy FAQ for Microsoft Copilot
- FTC: Scammers use AI to enhance their family emergency schemes
- Get Safe Online: Artificial intelligence
- Ofcom: From apps to AI search, how the UK goes online in 2025
Related NHI Mgmt Group resources: AI and Privacy Guide · How to Spot a Phishing Email Guide · Password Managers Explained for Everyone Guide · Cyber Hygiene Checklist Guide