Public AI tools at work: what providers say happens to the text staff paste in

What does OpenAI say?

OpenAI separates individual plans from business ones. For Free, Plus and Pro, it says it may use your content to train its models, and that you can opt out by turning off a setting called Improve the model for everyone, or through its Privacy Portal. Temporary chats are not used while they stay temporary. If someone saves one, it becomes a regular chat and follows the account's settings.

For ChatGPT Business, Enterprise and Edu, and for the API, OpenAI says it does not use inputs or outputs to improve its models by default. Its enterprise privacy page adds that business data may pass through automated classifiers and safety tools, and that human review depends on the service. The catch on individual plans is feedback. OpenAI says the entire conversation attached to a thumbs up or down may be used to improve its models even after you opted out. On business plans it says feedback does not permit training.

What does Anthropic say?

Anthropic's page for its consumer plans, Free, Pro and Max, names three cases in which chats and coding sessions may be used to improve Claude: you choose to allow it, a conversation is flagged for safety review, or you opt into a program such as its trusted tester program. It says Incognito chats are not used to improve Claude, even when model improvement is switched on.

For commercial products such as Claude for Work and the API, Anthropic says it does not use inputs or outputs to train its models by default. The exception is feedback: if you explicitly report feedback or bugs, for example with the thumbs button, it may use those chats and coding sessions to train. Owners of Team and Enterprise plans can switch the feedback button off for their organization.

What does Google say?

Google's help page for the consumer Gemini app says that, with Keep Activity on, your activity can be used to improve Google services, including the generative AI models behind them. It says trained reviewers, including service providers, read some chats. With the setting off, it says future chats are not used to improve Google's models unless you send feedback.

Workspace is a separate product with separate terms. Google's privacy hub for Gemini in Workspace says your content is not human reviewed or otherwise used for generative AI model training outside your domain without permission, and that Gemini only reaches content the user can already open. Administrators set how long prompts and responses are kept.

What does Microsoft say?

Microsoft's page on Copilot at work says prompts, responses and data accessed through Microsoft Graph are not used to train the foundation models behind Copilot. It says Copilot only surfaces organizational data the user already has permission to view, so the permissions already in your tenant decide what it can reach. It also says interactions are stored, and that administrators can search them and set how long they are kept.

What does this mean for a staff rule?

A rule that names brands goes out of date and misses the point. The same person can use ChatGPT on a personal account in the morning and in a company workspace in the afternoon, and the documents give those two accounts different defaults. The line to draw runs between accounts the company administers and accounts it does not.

On a personal plan, the protection is a setting each person has to find and keep switched off, with a feedback button that can undo it for one conversation. On a business plan, the default is set for everyone and an administrator holds the controls. A rule should write that difference down: which accounts are approved, who administers them, and what may never go into the others.

In the patterns we see, approving a business-plan tool that people actually want to use works better than banning the personal ones, because a ban moves the use out of sight. That is our reading, and the choice is yours. What you cannot do is learn the answer from a provider's page, because the pages describe the products and say nothing about your staff's habits.

What should you watch for in these documents?

These pages change without notice, and some carry no date. OpenAI's help article showed only a relative update time when we read it. Settings can also sit apart from the account setting: OpenAI says its coding agent Codex has a separate setting for training on full environments, and that changing the ChatGPT setting does not change it. Product names move as well. Microsoft's page notes that Microsoft 365 Copilot is now named Microsoft Copilot, so a rule that lists product names needs an owner who rereads them.