AI tools' data settings explained: chat history, training opt-out and business plans

Every AI tool has settings that sound like they protect your data. They protect different things. This is what the vendors' own help pages say about ChatGPT, Claude and Gemini as of October 2026, with links, so you can check the current wording yourself.

Four different things

Settings get confused because one word covers several questions:

• Training. Whether your conversations may be used to improve the vendor's models.
• History and retention. Whether the chat is saved in your account, and for how long the vendor keeps it.
• Human review. Whether staff or contractors can read some conversations.
• Feedback. What happens to a conversation when you press thumbs up or thumbs down.

Turning off one does not turn off the others. And none of them changes the basic fact that your text is sent to, and handled by, a third party.

ChatGPT

OpenAI's help page "Data controls in ChatGPT" says:

• When Improve the model for everyone is off, new conversations are not used to train OpenAI's models. They can still appear in your chat history. Turning the setting off does not delete saved chats.
• Temporary chats do not appear in your history, do not create or update memories and are not used to improve models. They may be retained for up to 30 days for safety purposes.
• If you give feedback such as a thumbs up or down, the entire conversation associated with it may be used to train OpenAI's models, even if you opted out of training.
• By default, content from ChatGPT Business, Enterprise, Edu and Healthcare workspaces is not used to train its models. Managed workspaces can have their own retention and access settings set by the organization.

Available controls depend on your plan and your workspace settings.

Claude

Anthropic publishes separate pages for consumer and commercial products.

Consumer (Free, Pro, Max). The Privacy Center page says Anthropic will use your chats and coding sessions, including to improve its models, if:

• you choose to allow it in your privacy settings;
• a conversation is flagged for safety review, in which case it may be used or analyzed to improve safety systems; or
• you explicitly opt in to training in another way, for example by joining a testing program.

Incognito chats are not used to improve Claude, even when model improvement is switched on. If you use the thumbs up or down buttons, the entire conversation is stored for up to 5 years and may be used to train models.

Commercial (Claude for Work, the API, Claude Gov). By default, inputs and outputs are not used to train Anthropic's models. Feedback or bug reports you submit, or data you choose to share, can change that. On Team and Enterprise plans an owner can turn off the ability to send feedback with the thumbs buttons.

Gemini

Gemini with a personal Google account. Google's Gemini Apps Privacy Hub covers the questions that matter here: who has access to chats and how human review works, how long temporary chats and chats with Keep Activity off are retained, and what the Keep Activity setting controls. The page is long and changes, so read those sections yourself before you decide what to enter. In particular, do not assume that switching Keep Activity off means a chat is deleted at once or never read.

Gemini with a Google Workspace account. Google's Workspace privacy page says:

• Prompts and generated responses are not used to train generative AI models without the customer's permission, and are not human reviewed outside your domain without permission.
• When conversation history is off in the Gemini app, new chats are saved in user accounts for up to 72 hours so Google can provide the service and process feedback.
• When history is on, administrators choose how long chats are kept: 3, 18 or 36 months, with 18 months as the default.

Side by side

This is a summary of the pages linked below, as read on 18 October 2026. Vendors change their terms. Check the page before you rely on it.

What no setting changes

• The text still leaves your computer and is processed on someone else's servers.
• A chat can be kept for safety or legal reasons even when training is off.
• Feedback buttons can send a whole conversation back to the vendor.
• For professionals with confidentiality duties, a privacy setting is not the same as client consent or a contract. The article on lawyers and confidentiality discusses the point for law firms.

The safest content is content that never held the client's identity. See how to use AI tools with client documents and the checklist for small firms.

Frequently asked questions

If I turn off training, is the chat deleted?

No. On ChatGPT, for example, turning the setting off does not delete or hide saved chats.

Is a paid personal plan the same as a business plan?

No. The help pages treat personal and business or workspace plans differently, and the defaults are different.

Should I avoid pressing thumbs up or down?

Be aware that it can send the whole conversation to the vendor. If the chat touched client material, do not.

Do these settings make it safe to paste a client's document?

They reduce one risk. Replace the identifying details first.

ClientSafe works on your PC and sends nothing to an AI service. You choose what to paste into the AI tool you use, and the safe copy contains placeholders in place of names and numbers.

Information as of October 2026, taken from the vendors' help pages. Not legal advice.