Can you stop ChatGPT from using your conversations for training?

Until yesterday, if you didn't want your ChatGPT conversations used to improve OpenAI's models, you had to go through a separate opt-out process. On 25 April 2023, OpenAI made it a simple switch. Anyone using ChatGPT can now switch off chat history, and any conversation started while it is off stays out of model training.

The change is rolling out to all users, and you can turn it on or off at any time. OpenAI also announced a new way to download everything ChatGPT stores about you, and a business plan with stricter data rules.

What changed

OpenAI's announcement covers three things:

Change	What it means for you	Available
Turn off chat history	New chats are not saved in your sidebar and are not used to train OpenAI's models	Rolling out to all users from 25 April
Export your data	A file with your conversations and other stored data is sent to your email	Now, in settings
ChatGPT Business	A subscription for professionals and companies; user data not used for training by default, as with OpenAI's developer API	Planned for the coming months

Two details are easy to miss. First, the switch applies to conversations started while history is off, so decide before you begin a sensitive chat. Second, "off" does not mean "never stored." OpenAI says it keeps these conversations for 30 days and looks at them only when needed to watch for abuse, then deletes them permanently.

Why training on chats matters

To understand the switch, it helps to know how chatbots improve. ChatGPT is built on a large language model, a system trained on vast amounts of text to predict what words should come next. After the first round of training, companies keep refining the model with examples of real conversations, often rated by people, so it learns which answers are helpful. Your chats can become part of that material.

That is useful for making ChatGPT better, but it worries people who type private things into it. Language models learn from their training data, and there is concern that fragments of personal or confidential text could resurface in later answers. The Register reports that several large companies, including JP Morgan, Goldman Sachs, Wells Fargo and Verizon, have limited their employees' use of ChatGPT for exactly this reason.

How to use the new setting

Here is the basic process, based on OpenAI's description:

Open ChatGPT and go to Settings.
Find the chat history option and switch it off.
Start a new conversation. A notice confirms that history is off: this chat stays out of your sidebar and out of training, and OpenAI deletes it within 30 days.
To go back, switch the option on again. New chats will once more be saved and may be used to improve the model.
To see what ChatGPT has stored so far, use the new Export option in settings and check your email for the file.

The trade-off is convenience. With history off, you cannot scroll back to that conversation later, so copy anything you want to keep before you close it.

Try it: Request your data export and open the file when it arrives. Skim your past conversations and count how many contain names, work details or personal information. That count is a good guide to whether you want history off by default.

Why OpenAI is doing this now

OpenAI presents the switch as a simpler alternative to its old opt-out process. The timing is also hard to ignore. Italy's data protection authority temporarily banned ChatGPT last month while it examined whether the service complies with European privacy law, and regulators in Canada, France and Spain have opened their own inquiries.

ChatGPT also had a privacy scare in March, when a bug let some users see parts of other people's chat histories and some subscribers' payment details, such as names and email addresses. OpenAI took the service offline briefly to fix it, and CEO Sam Altman said the problem came from an open-source library the service relies on. Against that background, giving users visible control over their data is a practical step, even if it is not a complete answer to every concern regulators have raised.

What should you still keep out of ChatGPT, even with history turned off?

Turning history off reduces how your words are used, but it does not make a conversation disappear instantly. For 30 days, the text still sits on OpenAI's systems and may be reviewed if abuse is suspected. So the safest rule hasn't changed: don't type anything you would be harmed by if someone else read it.

In practice, keep passwords, bank and card numbers, government ID numbers and medical records out of any chatbot. For work, follow your employer's policy, and don't paste confidential documents, client data or unreleased plans unless your company has approved the tool. If you need to work with real text, remove names and identifying details first, or replace them with placeholders like "Client A."

If your organization wants to use ChatGPT with company data, the planned ChatGPT Business tier, which will follow OpenAI's stricter developer data rules, may be worth waiting for. Until then, the history switch is a helpful tool for individuals, and the export option is a simple way to see what you have already shared.