Skip to content

What not to paste into an AI tool when documenting your work

Reviewed on 29 October 2026

A padlock resting on a laptop keyboard
Photo: rupixen on Pixabay

Short answer: leave out anything you wouldn't want a stranger to read. Public chat tools aren't a private notebook.

Before anything goes into a chat toolRead your notes line by lineSwap real details for labelsCheck the tool's data settingsPaste the cleaned text
Remove first, then paste.

Why it matters

The UK's National Cyber Security Centre says a query will be visible to the organisation providing the tool. It adds that such queries are stored and will almost certainly be used for developing the service or model at some point.1 Its advice is not to include sensitive information in queries to public tools, and not to submit anything that would cause trouble if it became public.1

In Australia, the privacy regulator (the OAIC) recommends that organisations do not enter personal information, and particularly sensitive information, into publicly available generative AI tools.2 That's the regulator's recommended practice. It isn't a full statement of the law, so ask the regulator or a lawyer about your own duties. In the UK and the US the rules differ. Check your own country's privacy authority.

Leave these out

  • Customer details. Names, addresses, phone numbers, emails, order history, health or financial details. Use "the customer" or "Customer A".
  • Staff details. Pay rates, performance notes, medical information, disciplinary matters.
  • Passwords and access codes. Also alarm codes, safe combinations, wifi keys and the answers to security questions. A procedure can say "the code is in the password manager", never the code.
  • Payment card and bank details. Yours or anyone else's.
  • Prices and terms you keep private. Supplier rates, wholesale costs, contract terms, quotes. Use placeholders such as [supplier price].
  • Contracts and legal papers. Especially ones with a confidentiality clause.
  • Anything under an agreement. If a customer or partner told you to keep it private, it stays out.
  • Screenshots and photos. They carry names, faces and numbers you may not notice. The OAIC says many photographs or recordings of individuals contain sensitive information.2

Swap in labels

Procedures rarely need the real detail. They need the shape of the job.

  • "Enter the client's account number" works as a step with no number in it.
  • "Apply the standard discount from the price list" works without the figure.
  • "Log in with the shared account" works without the password.

Fill in real details afterwards, on your own computer, if the page needs them. Often it doesn't.

Check the tool's settings

Plans and settings differ. Take one popular chat tool, with the maker's page checked on 29 October 2026. The maker says data sharing for model training is enabled by default on its personal Free, Plus and Pro plans.3 It also says that turning that setting off does not delete or hide saved chats.3 The same page says content from its Business, Enterprise, Edu and healthcare workspaces isn't used for training by default.3

Other tools differ. Before using any of them for work:

  • Read the terms of use and privacy policy. The NCSC says they need to be thoroughly understood before asking sensitive questions.1
  • Find out whether chats are kept, and for how long.
  • Work out who in your business can see what staff paste in.
  • Write a one-page rule for staff: what's allowed, what isn't, which tool.

Ask whether you need the tool at all

The OAIC says AI products shouldn't be used simply because they are available.2 If a job involves a lot of private detail, write that procedure by hand. A plain template can do the job.

For a mixed case, write the structure with the tool using placeholders, then fill in the sensitive parts yourself.

If something has already been pasted

  • Don't paste more.
  • Check whether you can delete the chat in the tool's settings.
  • Change any password that went in.
  • If it was customer or staff information, find out whether you have to report it. In Australia the OAIC has the guidance. Elsewhere, ask your privacy authority.

Reference list

  1. ChatGPT and large language models: what's the risk?, National Cyber Security Centre (UK)
  2. Guidance on privacy and the use of commercially available AI products, Office of the Australian Information Commissioner (Australia)
  3. Data controls in ChatGPT, OpenAI Help Center