Client data in ChatGPT. One switch to check before you paste a client's email

Someone in the office pastes a few complaint emails into ChatGPT and asks for polite replies. The replies are good. The chat now also holds the clients' names, phone numbers and order numbers. That is how client data in ChatGPT usually starts, with no bad intent, just a busy afternoon.

Nothing terrible has happened. But if you do this every day, it helps to know which setting decides what happens to those chats next.

The setting that decides what happens to client data in ChatGPT

ChatGPT has one switch that decides whether your conversations can be used to train OpenAI models. It is called Improve the model for everyone, and it sits in Settings under Data controls.

On the web, open your account menu, go to Settings, then Data controls, and turn the switch off. On a phone, open the sidebar and tap your profile icon to get to Settings. Once you are signed in, the setting applies to your whole account, on the web and on mobile, so you only need to change it once.

With the switch off, new conversations are not used for training. They still show up in your chat history. Nothing gets deleted.

There is one detail few people know about. If you rate an answer with a thumbs up or thumbs down, the whole conversation may be used for training, even with the switch off. In chats with client data, leave the thumbs alone.

Temporary chat for a one-off case

For a single case you don't want to keep, there is Temporary chat. It doesn't show up in your history, doesn't create memories and isn't used to improve OpenAI models. OpenAI may keep it for up to 30 days for safety reasons.

Claude has incognito chats, which you start with the ghost icon on a new chat. They aren't saved to your history or used for training, and they are kept for 30 days.

Claude and business plans work differently

Claude has its own switch with the same meaning. You'll find it in Settings, under Privacy, and it covers Free, Pro and Max accounts.

With it on, Anthropic may keep chats for up to 5 years. With it off, the usual 30-day period applies.

On business plans it works the other way round. By default, OpenAI does not use content from ChatGPT Business and Enterprise workspaces to train its models. Anthropic does not use chats from Claude Team and Enterprise plans by default either, unless someone sends feedback with the thumbs buttons.

If several people in your business use AI on personal accounts, each of them may have a different setting and nobody checks. One shared business plan keeps this simple.

Client A instead of a name

Settings are half of it. The other half is what goes into the chat in the first place.

Before you paste an email, replace the name with “Client A” and delete the phone number and address. The answer will be just as good. The model doesn't need to know who the client is to write a polite reply to a complaint. You put the name back in your email program.

The same goes for spreadsheets. If you ask for help tidying up an order list, delete the name and phone columns first. The model works with dates and products anyway, not with who placed the order.

Passwords, ID numbers like the Polish PESEL and card numbers don't go into a chat at all. Once pasted, they sit on someone else's server, and no switch takes that back.

Whether this use of AI needs a data processing agreement or a change to your records of processing depends on your business. Check GDPR questions like these with a lawyer, ideally before AI becomes an everyday tool in the office.

If the same kind of email goes through a chat every day, that step can be set up more carefully. With AI automation, where the data goes is decided at the start, not after the fact.

Today, open ChatGPT's settings and check that one switch under Data controls. Then agree on a simple office rule that names and numbers become “Client A” before anything goes into a chat.