Quick answer: treat an AI chatbot like a public place. Do not type in anything you would not want to become public, and check any important answer against a reliable source before you act on it. Chatbots are useful, but they are not always right and they are not private by default.
Last checked against official sources: 7 November 2026.
What to keep out of your chats
The UK’s National Cyber Security Centre (NCSC) published advice on AI chatbots in March 2023. It recommends two simple rules for public chatbots: do not include sensitive information in your queries, and do not submit queries that would cause problems if they were made public.
The NCSC explains the reasoning. Queries are visible to the organisation that runs the chatbot, and stored queries could be hacked, leaked or accidentally made public. It also points out that your information can build up across many queries made with the same login. The NCSC gives the example of questions about health or relationships, which can be revealing.
In practice, avoid typing or pasting:
- passwords, bank details, card numbers and ID numbers;
- medical or other highly personal details about yourself or other people;
- confidential documents from your employer, unless your organisation has told you the tool is approved;
- anything you promised to keep private.
Read the privacy settings
The NCSC also advises you to understand a service’s terms of use and privacy policy before asking it anything sensitive. Chatbot services differ in how long they keep conversations and what they may use them for, and these settings change over time. Look in the account or privacy settings of the service you use, check what it says about saved chats and how your conversations may be used, and adjust the options to suit you.
Why you must check the answers
The NCSC notes that chatbots are built from huge amounts of text gathered from many places, and that it is not possible to filter out all inaccurate or offensive content from it. Chatbots can therefore produce answers that sound confident and are wrong. They can also invent details such as names, quotations or sources.
Use this simple routine for anything that matters:
- Check the facts. Look up figures, dates, names and quotations in an official or well-established source.
- Ask for where it came from, then verify it. If the chatbot names a source, make sure the source exists and says what the chatbot claims.
- Be extra careful with health, legal and money matters. Use a chatbot to help you prepare questions, then confirm the answer with a qualified professional or the official body concerned.
- Compare more than one source when the topic is new to you.
Use AI as a helper, not a final authority
Chatbots are good for drafting, brainstorming, explaining a concept in simpler words and organising your own notes. You stay responsible for what you send or publish, so read everything the chatbot writes before you use it. If you want clearer results, our guide to writing better AI prompts shows how to ask.
Common questions
Is anything I type in a chatbot private?
Do not assume so. The NCSC’s advice is to behave as though your queries could be seen or stored, and to read the service’s privacy policy.
Can I paste my work documents into a chatbot?
Only if your employer has approved the tool and the type of information. Many organisations have rules on this, so ask first.
Does a confident answer mean it is correct?
No. The tone of an answer tells you nothing about whether it is accurate.
Related guides
- Why AI Chatbots Make Things Up (and How to Check Their Answers)
- 8 Everyday Ways to Use an AI Assistant (and When to Be Careful)
- AI Voice Cloning Scams: How to Protect Yourself and Your Family
- AI Chatbot Privacy Settings: What to Review (ChatGPT and Gemini Examples)
More in Everyday AI Tools, or browse all guides.
Sources
This guide draws on the NCSC blog post “ChatGPT and large language models: what’s the risk?” (March 2023) and the NCSC’s later blog on the security of AI systems. The NCSC post is from 2023, so we have kept to its general principles rather than describing how any one service works today. See also our Editorial Policy.


