Is ChatGPT private? What freelancers should know about client data

You ask ChatGPT to draft a quote or an email to a client. Halfway through, you wonder whether the client's name should be in there at all.

Updated on 19 October 2026 · Rob den Otter

Short answer

Whether ChatGPT is private comes down to one setting: your chats can be used to train OpenAI's models unless you turn off "Improve the model for everyone". Even with that setting off, client names are better left out. What does work is giving your AI the context of your work without names, with a role and sector instead of a name. That is what Keepstone does.

Is ChatGPT private? What happens to your chats

Your chats can be used for training as long as "Improve the model for everyone" is switched on. Turn it off and OpenAI no longer uses your new chats for training. They stay in your history.

For a sensitive question there is Temporary Chat. It doesn't appear in your history and isn't used for training. OpenAI does keep it for up to 30 days, for safety.

Note: when you start a Temporary Chat, ChatGPT says it won't follow your custom instructions. A second brain stored there won't work in that chat.

How do you turn it off in ChatGPT?

On the website:

Open your account menu and choose Settings.
Choose Data controls.
Turn off "Improve the model for everyone".

In the app the route is the same: side menu, your name, Settings, Data controls. OpenAI says the setting applies to new chats.

What about Claude, Gemini and Copilot?

Claude and Gemini also let you stop training. They differ in how long they keep your chats.

Training and retention per AI, personal account (as of 19 October 2026)

AI	Setting	With training off	Business version
ChatGPT	"Improve the model for everyone" (Data controls)	No training on new chats	Business, Enterprise, Edu: no training by default
Claude	"Help improve our AI models" (Settings, Privacy)	Kept for 30 days (with training on: up to five years)	Claude for Work, Education and the API are not covered by these terms
Gemini	"Keep Activity" (Activity, Gemini Apps Activity)	No training unless you send feedback, kept for 72 hours	Workspace: different terms

With Gemini, reviewers, including service providers working for Google, read some of your chats to improve Gemini as long as "Keep Activity" is on. Reviewed chats are kept for up to three years, separate from your account. Even with the setting off, chats can be reviewed for safety. With the setting on, Google keeps your activity for 18 months by default.

Copilot. In the new Copilot app (since 18 August 2026), no setting for model training is shown under its data controls (checked on 19 October 2026). For the previous version of the app, Microsoft said it doesn't train on work or school accounts or on Copilot in Microsoft 365 Personal and Family. For the new app, that isn't stated yet.

Sources: OpenAI, Anthropic, Google, Microsoft.

Which client details are better left out?

Keep these out of an ordinary chat with your AI:

names of clients and contacts
email addresses and phone numbers
amounts or agreements that point to one client
health information, even if you're a coach
anything covered by a confidentiality agreement

What does work: a client's role and sector ("an installation firm with twenty staff"), the type of job, your own rates and how you work.

Does an AI without client names still help?

Without names your AI knows nothing about your clients. It gives generic answers and you keep explaining what you do and what you charge.

Pasting in more (old quotes, emails, a client list) brings the names straight back. A fixed context without names does solve it.

Keepstone captures that context once, for ChatGPT, Claude, Gemini or Copilot.

Start for € 295

How does Keepstone handle client names?

Keepstone builds a second brain for your own AI: a short instruction and five files about your business, your clients, your rates, your way of working and your writing style.

This is how Keepstone handles your client data:

Client names you list in the form, and email addresses and phone numbers, are replaced by codes before the AI reads your documents.
Your uploads are deleted as soon as they've been read. Your answers and the draft are kept until 30 days after delivery. Only your invoice details are kept for 7 years, as the law requires.
In the second brain, clients appear as role and sector by default. A real name only goes in if you say so for that client.
If you have a legal duty of professional confidentiality, no names go in at all.
Every draft is personally checked before you get it.

The data processing agreement and the list of sub-processors are under Privacy and terms. The setup costs € 295 excl. VAT, one-off, no subscription. See an example second brain.

What can still slip through?

No system catches every name. Codes replace the names you list in the form. A name that appears elsewhere in a document can slip through.

What you type in the interview yourself isn't replaced with codes. That's why the interview asks you not to mention names. So this is pseudonymisation, not anonymisation. The data processing agreement says the same.

Checklist: using ChatGPT with client data

Turn off the training setting in the AI you use (the new Copilot app doesn't have one).
Use Temporary Chat for a sensitive question.
Leave out names, contact details and health information.
Capture the context of your work once, without names.
Check that your privacy notice says you use AI tools (UK GDPR, right to be informed).

Frequently asked questions