Does ChatGPT or Copilot train on your business data?
Whether ChatGPT trains on business data depends on the account. Business plans such as ChatGPT Team and Enterprise, and Microsoft 365 Copilot used with a work sign-in, are generally set up so your chats are not used to train the models. Free and personal accounts are different. Training is often on by default there, unless the user switches it off. And training is only one of the risks.
Consumer accounts and business accounts
AI companies sell two kinds of account. They treat the data differently.
Consumer accounts. Free and personal paid plans. Built for individuals. The terms usually allow chats to be used to improve the models. The person using the account controls the settings, not the business.
Business accounts. Team, Enterprise and work versions of Copilot or Gemini. These usually come with a promise not to train on your content. They also give an administrator some control over settings and who can use them.
The exact terms change. Read the provider's current data page before you rely on any of this. Our guide to ChatGPT Team vs Enterprise vs free goes through the plans one by one.
Copilot is not one product
The name covers several tools. The free Copilot in a browser or Windows is a consumer product. Microsoft 365 Copilot, used with a work account, sits inside your company's Microsoft setup and follows its business terms.
A simple check: when someone opens Copilot, is there a sign it is a work account? Microsoft usually shows a shield or a work label. No label, treat it as personal. There is more in is Microsoft Copilot safe for client data.
Settings to check on ChatGPT and other tools
Settings move and get renamed. These are the ones worth finding on any AI tool your team uses.
- Model training. Often called something like "improve the model for everyone". On consumer accounts, turn it off.
- Chat history. Whether past conversations are kept, and for how long.
- Temporary or private chats. A mode that is not saved to history. Providers may still keep it for a short time for safety checks.
- Memory. Some tools remember details across chats. Check what is stored and clear it.
- Shared links. A shared chat can be opened by anyone with the link.
- Connected apps. Links to email, drives or calendars widen what the tool can see.
Rule of thumb: a setting on a personal account belongs to the person, not the business. You cannot see it, check it or switch it back on if it changes.
Why training is only one of the risks
Owners often ask about training first. It is a fair question. But a chat that is never used for training can still cause a problem.
Data sits in chat history
A pasted client letter stays in the history. Anyone who gets into that account can read it. That includes a former employee who still has the login, or someone using a stolen password.
Staff use personal accounts
This is the common gap. A firm pays for a business plan. Then someone uses their own free account on a busy afternoon, because it is already open. The business plan's promises do not cover that chat.
Not every tool has business terms you have agreed
Staff try new tools. Some are run from countries with different data rules. Our piece on DeepSeek at work covers one example.
Your own duties still apply
Under UK GDPR, you are responsible for client personal data. Sending it to an outside service needs a lawful basis and proper terms. A setting does not change that duty. Nor does it answer a client or insurer who asks what controls you have.
Settings decide what the AI company does with data. They do not decide what your team sends.
What actually covers the gap
Settings are the first layer. They are worth doing. Most firms add a few more.
- Pick approved tools. Business accounts only, with training off.
- Write a short AI policy. What can go in, what cannot. Get each person to agree to it, with a date.
- Show people how to take data out. Names become labels. Account numbers come out. Our guide on redacting before pasting into ChatGPT walks through it.
- Check what happens in practice. A policy nobody can check is a hope.
Worked example: Priya at Ashcombe wants help with a letter to Fenwick Joinery about invoice 4471. She pastes it with the client name replaced by "Client A" and the bank details taken out. The question still goes. The client data does not.
Where Helio Guard fits
Helio Guard works on step four. An agent on each Windows PC adds an extension to Chrome and Edge. It checks pastes, messages and file uploads on listed AI sites, on any account, business or personal. Card numbers, UK bank details, National Insurance numbers and client names are taken out or swapped for stand-ins such as "Client A". Checked on the PC, with fixed rules. We never see what your team types. It also records each person's agreement to your AI policy. It costs £1 a person a month, from 5 people, with a 14-day free trial and no card needed. Details are on the pricing page.
Common questions about AI tools and business data
Does ChatGPT Plus train on my data?
ChatGPT Plus is a personal plan, so it is treated like a consumer account. Chats can be used to improve the models unless the user turns that setting off. Paying for Plus does not give a business the admin controls or the no-training default that come with ChatGPT Team or Enterprise.
If training is off, is it safe to paste client data into ChatGPT?
Turning training off in ChatGPT reduces one risk but not all of them. The chat can still sit in history, be read by anyone with the login, or be shared by link. Client data still leaves your business, so UK GDPR duties apply. Taking names and numbers out first is the safer habit.
Can I stop staff using personal ChatGPT accounts at work?
A written AI policy can ban personal ChatGPT accounts for work, and most firms should say so plainly. A policy alone cannot check what happens on each PC. Many businesses add a tool that checks what is pasted into AI sites, whichever account is signed in, or block unapproved sites outright.
Related: Is pasting a client email into ChatGPT a data breach under UK GDPR?