Is it safe to put client data into ChatGPT, Claude or Copilot?

The plan matters more than the brand

The same tool can behave differently depending on which plan you use. OpenAI's help centre says inputs to ChatGPT Business, Enterprise, Edu and the API are not used for training by default, while Free, Plus and Pro conversations can be, unless the user opts out.

Anthropic made a similar split. From its post of 28 August 2025, Free, Pro and Max users can choose whether their chats help train Claude. Opted-in data is kept for five years. The change did not apply to Claude for Work (Team and Enterprise), Claude Gov, Claude for Education or the API. Claude's pricing page also states that content on the Team plan is not used for training by default.

Microsoft and Google work differently

Copilot and Gemini sit inside the suites many firms already pay for. Their data terms sit in your Microsoft or Google agreement, and they depend on your licence and admin settings. We have not set out their terms here because they change by licence type. Ask your account admin, or your reseller, to confirm the position in writing for the exact licence you hold.

Training is only one question

Whether a vendor trains on your data is a separate question from whether you may send it there. The ICO has said that anyone using generative AI with personal data should settle a lawful basis, assess the risks with a data protection impact assessment before processing starts, and consider security. That applies to client names, contact details, case notes and anything else that identifies a person.

There is no exemption because the tool is new. If a client's personal data goes into an AI tool, your firm is still responsible for it. Check your client contract and your engagement letters too. Some clients restrict where their information can go.

Rules a small firm can set this week

Name the approved tools and plans. Say which plans are not approved, in plain words, for example personal free accounts.

Say what may not be pasted in. A short list works: client identifiers, health or financial details, anything marked confidential, and passwords or keys.

Say what can be used safely: public information, your own templates, and client material with the identifying details removed.

Say who to tell if something goes wrong. A personal data breach can have a reporting deadline. Decide this before you need it.

A check before you start

Open the admin settings of each tool you pay for and look for the data controls. Confirm the training setting, how long chats are kept, and who can see them. Write down what you found and the date, because plan terms change.

If you cannot confirm a setting, treat the tool as not approved for client data until you can.

Where this leaves you

A business plan with training off is a sensible starting point, not a finish line. The rest is a written rule, a short briefing for your team, and a habit of checking the settings when plans change.

This is a general guide, not legal advice. For your own position, speak to your data protection adviser. For a side-by-side look at the tools, see our comparison of AI assistants for UK small businesses. If you want help setting up the rules and the settings, get in touch through our Contact page.

Sources

OpenAI help centre, 'What is ChatGPT Business' and data controls articles, help.openai.com. Read 3 November 2026 through the help-lb.openai.com mirror. No fixed update date is shown.

Anthropic, 'Updates to our consumer terms', anthropic.com/news, 28 August 2025. Read 3 November 2026.

Anthropic, Claude pricing page, claude.com/pricing. Read 3 November 2026.

ICO, 'Generative AI: eight questions that developers and users need to ask', Stephen Almond, 3 April 2023, ico.org.uk. Read 3 November 2026.