Before You Paste Company Information into an AI Chat — What Never to Enter, and the Setting That Stops Training

(Lead) AI chat is useful, but few companies have decided what may be typed into it. This guide explains where entered information goes, what must never be entered, and the setting to change on a representative service.

Where does entered information go?

On personal plans (free and individual paid tiers), what you enter may be used to improve the service and train models. Once content is used for training, in principle it cannot be taken back. Overseas, cases have been reported of employees pasting work source code into an AI chat and causing an information leak. Before any setting: treat anything you enter as having left the building.

What never goes in, whatever the settings

• Customer and personal information (names, addresses, phone numbers, order details)
• Unreleased business information (prices, sales, deal terms, new-product details)
• IDs, passwords, secret keys, card details
• Full contracts or quotes with real names and amounts

If you need AI help with wording, replace names with placeholders and remove amounts and dates first.

The ChatGPT setting that stops training

OpenAI's official help gives these steps:1. Open the account menu and choose Settings2. Choose Data controls3. Turn off "Improve the model for everyone"

With it off, new conversations are not used for training, and your chat history remains. One caution from the official help: rating an answer (thumbs up/down) can still make that conversation be used for training. The same setting exists in the iOS/Android apps under profile > Settings > Data controls.

For work, use a business plan

Business plans (such as ChatGPT Business / Enterprise) state that entered content is not used for model training. If the company uses AI chat, the starting point of information management is to provide a company-managed business plan — not personal free accounts — and to decide who may use it for what. Our guide "Generative AI Rules at Work: The Minimum Set for a Small Company" covers the rules.

Summary: today's checklist

• Decided which kinds of information must never be entered
• Stopped handling company information in personal accounts
• Checked the ChatGPT data-controls setting
• Considered a business plan for work use

Check your understanding

Tap an answer to check yourself. Nothing is saved or sent.

Q1. On a personal AI-chat plan, what can happen to what you type?

It is always deleted immediatelyIt may be used to improve the service and train modelsIt is encrypted so no one can see it

See the answer

Some personal plans share data by default. Treat anything entered as having left the building.

Q2. Which ChatGPT setting stops new conversations being used for training?

Settings > Data controls > turn off "Improve the model for everyone"Delete all historyLog out

See the answer

History remains, and new conversations are not used for training (OpenAI's official help). A rated conversation can be an exception.

Q3. You want AI to summarize a contract with real names. The right approach is:

Paste it as isReplace names with placeholders and remove amounts and dates firstPaste it as is because the setting is off

See the answer

Confidential information stays out regardless of settings. Mask first, then use.

Read next

See the latest threat intelligence

Start Here — security in 3 steps

← Back to the guide list