What Not to Tell an AI Chatbot: 8 Kinds of Information to Keep Out

Treat a chatbot like a postcard

A chatbot feels private because it is just you and a text box. It is not. Your messages travel to a company’s servers, are stored for some time, and in some products are read by trained reviewers who check quality and safety. The UK National Cyber Security Centre puts it plainly: queries are visible to the provider, and it recommends not including sensitive information in queries to public chatbots.

That does not mean you should avoid these tools. It means deciding in advance what goes in.

What to keep out

1. Passwords, PINs and recovery codes. People paste these while asking for help with a login problem. A chatbot never needs them. If you already did, change them.

2. Card, bank and tax details. Account numbers, full card numbers and tax IDs are exactly what fraudsters look for. If you want help with a budget, use rounded figures or fake labels.

3. Government ID numbers and documents. Social Security numbers, passport or driving licence details and photos of ID cards stay out. They are the raw material for identity theft.

4. Medical records and test results. You can ask what a term means. Avoid uploading full records with your name, date of birth and insurer attached.

5. Other people’s private information. Your client’s contract, a patient’s notes, a friend’s confession, your child’s school record. You do not have their permission to hand it to a third party.

6. Work secrets. Source code, unreleased numbers, customer lists and legal material. Many employers ban this, because the data leaves the company’s control.

7. Anything you could be blackmailed or embarrassed by. Chats can be breached, subpoenaed or shared by mistake.

8. Full identity bundles. Name plus address plus date of birth in one message is a ready-made profile for a scammer. Our guide to phishing messages shows how such details are used against people.

How chats end up exposed

Several ordinary routes exist. Providers store history. Some chats are reviewed by people. Accounts get taken over. And sharing features can backfire: in 2025, OpenAI removed an option that let shared ChatGPT conversations be found through Google after people ticked a “make this chat discoverable” box without fully realising what it did. It was not a hack. It was a setting that led to accidental publication.

Redact, then ask

You usually get the same quality of help without real details. Instead of “My name is Priya Shah, account 4417…, why was I charged a fee?”, write “A customer was charged a $35 overdraft fee after a $20 purchase. What are the common reasons?” Describe the situation, not the person.

Use a temporary or incognito mode for sensitive topics, and review your privacy options. Our walkthrough of privacy settings and data retention covers ChatGPT, Claude and Gemini.

If you slipped

Delete the conversation, change any credential you shared, and turn on two-step verification for the affected account. If the leaked information could enable identity theft, follow the steps in what to do after a data breach. Keep an eye on recent incidents too, since chatbot accounts themselves can be targeted.

Used with care, chatbots are useful. The simple habit of pausing before you send keeps them that way.

This article reflects public information as of 2026-11-05. AI tools and rules change quickly; we update articles when new facts are published. Spot an error? See our editorial and corrections policy.