Someone in your team has a long customer email chain and a reply to write, so they paste the whole thing into ChatGPT and ask for a draft. Can you put client data into ChatGPT like that? Sometimes, but not on a personal account, free or paid, and not without knowing where the information goes and what you've promised the people it's about. This guide is for anyone responsible for a business or a team who suspects this is already happening and wants a straight answer and some rules to give the team.
This is general information to help you work through the question. It isn't legal advice, and if you're unsure about your own position you should take advice on it.
The short answer
Whether it's acceptable depends on three separate things: the kind of account the person is using, the kind of information they're putting in, and what you've agreed with the people the information is about.
On the information side, it helps to sort it like this:
| What the information is | What applies |
|---|---|
| Personal data: anything about an identifiable person, such as a customer's name and email, a staff record or a supplier contact | UK GDPR applies, and your business is responsible for what happens to it |
| Confidential business information: a client's draft accounts, deal terms, pricing, a fund's holdings | Your contract, engagement letter or professional duty applies, even if no individual is named |
| Information that's public or belongs only to you: your own marketing copy, a published price list, a general question | Generally fine to use, subject to your own judgement |
Much of what a business handles day to day falls into both of the first two rows at once. A customer's email about a disputed invoice contains personal data and is usually confidential too.
Why a personal account and a business plan are different things
The same chatbot can sit under quite different terms depending on how you sign in. A personal account, free or paid, is generally covered by the provider's consumer terms and privacy policy. Business plans and the API (the route developers use to build a provider's models into other software) are generally covered by a business contract, often with a data processing addendum: a contract setting out how the provider handles personal data on your behalf.
Here is what each provider's own pages said when we checked them on 11 November 2026. Terms change, so read the current version before deciding anything.
OpenAI (ChatGPT). OpenAI's business data page says that by default it doesn't use data from ChatGPT Enterprise, ChatGPT Business, ChatGPT Edu, ChatGPT for Healthcare, ChatGPT for Teachers or its API platform, including inputs and outputs, to train or improve its models. Its Data Controls help article says that on personal ChatGPT Free, Plus and Pro accounts, data sharing for model training is on by default, and you can turn it off under Settings, Data controls, "Improve the model for everyone". It also says that if you give feedback on a response, such as a thumbs up or down, the whole conversation may be used for training even after you have opted out. OpenAI's API documentation on data says that since 1 March 2023, data sent to the API isn't used to train or improve its models unless you opt in.
Microsoft (Copilot). Microsoft's page on enterprise data protection says that Copilot and Copilot Chat, as used by organisations, are covered by its Data Protection Addendum and Product Terms, with Microsoft acting as a data processor, and that prompts and responses aren't used to train its foundation models. Its Copilot Chat privacy page says a green shield shows at the top of the chat when that protection applies.
Google (Gemini). Google's Workspace generative AI privacy hub says that on qualifying Google Workspace editions, prompts are customer data under its Cloud Data Processing Addendum, and your content isn't reviewed by people or used to train generative AI models outside your organisation's domain without permission. The Gemini Apps Privacy Hub for personal accounts says that with the Keep Activity setting on, Google uses your activity to improve its services, including training generative AI models.
Anthropic (Claude). Anthropic's privacy policy for Claude.ai says it may use your inputs and outputs to train its models unless you opt out in your account settings, with some exceptions, such as conversations flagged for safety review, and that the policy doesn't apply to its business offerings, such as Enterprise accounts. Its commercial terms say it may not train models on customer content from its services, and that data is processed under its Data Processing Addendum.
Whichever provider you use, your business is the controller of the personal data you put in: the one responsible for it. Where a provider processes personal data for you, Article 28 of the UK GDPR says this must be governed by a written contract and lists what it must cover. As we read it, a personal account an employee set up for themselves gives your business no such contract with the provider.
What the UK GDPR actually asks of you
The principles are in Article 5. In our words rather than the law's, these are the questions to ask before personal data goes into any AI tool:
- Do you have a lawful reason to use the data this way? The ICO explains the options under lawful basis. Health information and some other types are "special category data", with extra conditions.
- Would the people concerned expect it? Article 5 says data must be processed "lawfully, fairly and in a transparent manner", so, as we read it, your privacy notice should cover it.
- Are you putting in only what's needed? If the tool only needs the gist of a complaint, it doesn't need the customer's address and account number.
- Is it secure, and is there a contract? Article 5 requires appropriate security, and Article 28 covers the processor contract.
- Does the data leave the UK? If so, the ICO's rules on international transfers may apply; its page has a tool to help you check.
Article 5 also says the controller must be able to demonstrate compliance, so keep a note of what you decided and why.
For a new use of AI with personal data, read the ICO's guidance on when a data protection impact assessment (DPIA) is needed. A DPIA is a written assessment of the risks to people and how you'll reduce them.
The ICO's guidance on AI and data protection (last updated 15 March 2023) and several other ICO pages linked here were marked as under review after the Data (Use and Access) Act 2025 when we checked on 11 November 2026, so check the current position.
Confidentiality is a separate question
Data protection law covers information about people. Confidentiality comes from what you've promised, and it can apply where no individual is named at all. A client's draft accounts, the terms of a deal, a supplier's pricing or a fund's holdings may contain no personal data and still be covered by your contract, an NDA, your engagement letter or a professional duty.
So, separately from GDPR, check your client contracts and NDAs for clauses about subcontractors, third parties, where data is held and, in newer contracts, AI. If a contract says you won't share a client's information with third parties without consent, pasting it into a personal account is hard to square with that. A business plan doesn't get round that on its own, because the provider is still a third party. It may fit what the contract allows, or you may need the client's agreement, so check the wording.
The real risk is usually personal accounts, not the tool
Illustrative example, not a client: an office manager has a tricky reply to write to a customer who is unhappy about a late delivery. She copies the whole email chain, with the customer's name, address and order history, into a free ChatGPT account she set up on her own email last year, and asks for a polite reply.
As we see it, much of the risk comes from this: people trying to get through their work, with nobody having decided anything. Banning AI outright is more likely to push it out of sight than stop it. Dealing with it is mostly a management job:
- Find out what's in use. Ask the team which tools they use and for what, and make clear you're asking so you can set things up properly, not to catch anyone out.
- Decide which tools and accounts are approved. Approved should mean a business account signed in with a work email, with the settings checked.
- Give people an approved way to do the same job. If an approved tool drafts replies just as well, there's no reason to go round it.
Some AI is already switched on inside software you pay for, such as your email or accounting package. We've written separately about the questions to ask your software suppliers about their new AI features.
A simple way to sort your data: red, amber, green
This is a starting point. Where the lines fall depends on your business and what your contracts say.
| Colour | Examples | Rule |
|---|---|---|
| Red | Health information and other special category data, bank details, ID documents and passport scans, passwords and other log-in details, anything a client has given you in confidence | Never into a personal or public account. Only into an approved tool if you've specifically checked and agreed it |
| Amber | Customer names and emails, invoices, staff records, draft accounts, internal reports | Only into an approved business account, after checking its terms |
| Green | Public information, your own marketing copy, general questions with no client or staff details | Fine in any approved tool |
Removing a name doesn't always make information anonymous. An email about "our client in Leeds with the warehouse fire" can identify someone as well as a name can. The ICO's guidance on pseudonymisation says that pseudonymised data, where names are swapped for codes but you hold the key, is still personal data in the hands of someone who holds the additional information. If you rely on stripping details out before using a tool, be strict about it.
A one-page AI policy starter
A policy for a small team doesn't need to be long. Something like the rules below, adapted to your business, covers most of it. It's a starting point to work from, not a finished policy, and not legal advice.
- Only use the AI tools on our approved list, signed in with your work account. Don't use personal accounts for work.
- Before putting information into an approved tool, check it against the red, amber, green list.
- Never put red information into any AI tool unless [named person] has agreed that specific use.
- A person must check anything AI has helped produce before it goes to a client, a customer or a regulator.
- AI must not make decisions about people, such as hiring or credit, without a person reviewing them.
- Don't switch on AI features in our existing software without asking [named person] first.
- If you're unsure whether something is allowed, ask [named person] before you go ahead.
- If something goes into a tool by mistake, tell [named person] straight away. It may need reporting.
- We will review this list and these rules every six months.
On the mistake rule: putting personal data somewhere it shouldn't go may count as a personal data breach. The ICO's page on personal data breaches explains how to assess the risk, and says that if a risk to people is likely, you must notify the ICO as soon as possible and, where feasible, within 72 hours.
A policy like this only works if the team knows about it and has an approved tool to use. If you'd like help, we can write a short AI policy for your team with you in plain English, though plenty of businesses can do it themselves from a list like this one.
Common questions
Is ChatGPT GDPR compliant in the UK?
That depends on how you use it as much as on the tool. Your business is the controller of the personal data you put in, so you need a lawful basis, appropriate security and, where the provider processes personal data for you, a written contract with it.
Does ChatGPT use my data to train its models?
It depends on the account. OpenAI says that by default it doesn't use data from its business plans (ChatGPT Business, Enterprise and Edu, among others) or its API for training. On personal Free, Plus and Pro accounts, OpenAI says conversations can be used for training unless you turn off "Improve the model for everyone" in Settings, Data controls. That setting doesn't stop it being able to use a whole conversation you have given feedback on, and it isn't a substitute for a business plan if you are handling client data.
Do I need to tell clients I use AI?
Your privacy notice should explain how you use personal data, which may include AI tools. For confidential work, check your contracts and engagement letters; the professional bodies' AI guidance for tax work suggests mentioning the potential use of AI tools in the engagement letter.
What should I do if someone has already pasted client data into ChatGPT?
Find out what went in, which account was used and when. If it included personal data, assess the risk using the ICO's breach guidance; if a risk to people is likely, the ICO must be told as soon as possible and, where feasible, within 72 hours. Check too whether any client contract requires you to tell the client.
Talk to us
Not sure what your team is already putting into AI tools, or want a short set of rules in plain English? Book a free 30-minute call. We'll tell you what we'd check first, and whether you need any help at all.
This article is general information, not legal, financial or regulatory advice.