Do ChatGPT, Claude and Gemini use your company's data?

A question owners often ask is not "which AI is the best". It is "if my salesperson pastes a client's quote into it, does that quote end up inside the AI?".

The short answer: it depends on the plan, not the brand. The same ChatGPT, the same Claude, the same Gemini behave differently depending on whether you use them with a personal account or a business contract.

The general rule

On business plans and APIs, the main providers state that they do not train their models on your data, unless you consent ^{1, 4, 6, 7}. On personal accounts, free or paid, the rule varies, and in several cases training is possible until you switch it off ^{2, 5, 8}.

So a company's weak point is almost never the tool the owner chose. It is the personal account an employee opened because it is "quicker".

OpenAI: ChatGPT and API

For the APIs, OpenAI states that the data you send is not used to train or improve its models, unless you choose to share it, and that this rule has applied since 1 March 2023 ¹. Logs kept for abuse monitoring are retained for up to 30 days ¹.

For ChatGPT on personal plans, check the data controls section in the settings and the option on model improvement. The OpenAI help pages describing the default behaviour could not be consulted at the time of checking, so we do not report what they say here: look directly in your own account.

Anthropic: Claude

From 28 August 2025, Anthropic trains its new models on data from Free, Pro and Max accounts when the setting is on ². If you leave it on, your data is kept for up to 5 years; if you switch it off, it stays at 30 days ².

The toggle is called "Help improve our AI models" and is in the settings, under Privacy ³. Changing it applies to new chats, not past ones ⁹.

Commercial plans, that is Claude for Work and the APIs, are excluded from this change ². For these, Anthropic states that by default it does not use inputs and outputs to train its models ⁴. The exception is feedback you send yourself, such as a thumbs up or down ⁴.

Google: Gemini

With the personal Gemini app, some chats are read by human reviewers ⁵. If you turn off Gemini Apps activity, chats are no longer used for training, apart from feedback you send ⁵. With activity on, chats are deleted automatically after 18 months ⁵.

With a business Google Workspace account, Google states that customer data is not used to train models without the customer's permission and is not read by human reviewers ⁶. How long conversations are kept is decided by the administrator ⁶.

Microsoft: Copilot

In business Microsoft 365 Copilot, prompts, responses and data read from the company's documents are not used to train the models ⁷.

For personal Copilot, Microsoft states that it uses conversation data for training, except for some categories of users and for those who have opted out ⁸. The exclusions include business accounts and subscribers to Microsoft 365 Personal and Family ⁸. By default, conversations are kept for 18 months ⁸.

What to do now

• Make a list. Ask around the company who uses which tool, and with which account. The surprise is often there.
• Give a business tool to whoever uses AI for work. It costs less than a quote ending up in the wrong account.
• Switch off training on the personal accounts that remain, using the settings described above.
• Write a one-line rule: what must never be pasted into an AI (client data, confidential prices, staff data) unless it is done with the business tool.
• For the most sensitive data, consider a local model, which runs on your own computers and sends nothing out.

A business plan settles the training question. It does not settle the GDPR: the privacy notice, the legal bases and the processing agreement with the provider still need to be done.