November 19, 2026 · AI writer · Oak City Intelligence
What Never Goes Into the Chat Box

The List
Five kinds of information stay out of any chat assistant, whatever plan you pay for:
- Full card numbers, including the ones on a photo of a receipt.
- Social Security numbers and anything else a thief could open an account with.
- Health details, if you run a clinic, a dental office or anything else that falls under HIPAA.
- Privileged client matter, if you practice law. Your state bar's guidance on AI governs here, and an assistant's terms of service don't override it.
- Passwords and login codes, including the ones a customer reads out over the phone.
Everything else calls for judgment, and the test fits in one question: would you be comfortable if the paragraph you just pasted turned up in front of a stranger? A customer's first name and the fact that their water heater leaks pass easily. Their name, home address, gate code and the days they travel don't.
The Setting
On the free and personal paid plans, the three big assistants may use what you type to train future models unless you tell them otherwise. Each has one switch for it, and each takes under a minute to flip:
- ChatGPT: Settings, then Data Controls, then turn off "Improve the model for everyone." New chats stop going into training, and your history stays where it was.
- Claude: Settings, then Privacy, then turn off "Help improve Claude." Anthropic's consumer plans (Free, Pro and Max) train on chats unless you switch it off.
- Gemini: turn off "Keep Activity" in your Gemini Apps activity settings. Google still holds chats for up to 72 hours with the setting off, and says human reviewers may still see some of them.
If your shop uses Microsoft Copilot, look for its privacy settings the same way, and remember that signing in with a work account on a Microsoft 365 business plan puts it under your business terms rather than personal ones.
Flip it today on every account anyone in your business uses, including the one your office manager set up on a personal email.
Why a Business Plan Changes Things
The team and business plans work differently. ChatGPT Business, Claude's Team and Enterprise plans, and Gemini on a Google Workspace business account all leave your chats out of training by default, and Google says Workspace chats aren't seen by human reviewers either. You also get an admin who can see which accounts exist and shut one off when someone leaves, a power the personal plans never give you.
If more than one person in your shop uses an assistant for work, that admin control alone usually justifies the switch. Ask what the plan costs per seat, compare it against what you already pay for Google Workspace or Microsoft 365, and check whether the tool you already pay for includes an assistant before you buy a second one.
What the Setting Won't Do
None of these switches makes an assistant safe for the five items on the list. Training settings decide what the vendor learns from. They do nothing about whether your chat history sits on someone else's servers, gets read by a person reviewing for abuse, or turns up when a staff member's account gets broken into. Keep the list short, tape it near the computer, and treat the setting as a backup to the list, never a substitute for it.
Sources
- OpenAI, Data Controls FAQ: https://help.openai.com/en/articles/7730893-data-controls-faq
- Anthropic, model improvement privacy settings: https://privacy.claude.com/en/articles/12109829-how-do-i-change-my-model-improvement-privacy-settings
- Google, Gemini Apps Privacy Hub: https://support.google.com/gemini/answer/13594961
Written by Oak City Intelligence's AI writer. Reviewed before publication.