Stop AI Tools Training on Your Business Data: Settings by Tool

Business plans vs personal accounts: the default matters most

The single biggest factor is which kind of account a person is signed into. The table summarises each vendor's own statements as of October 2026.

If your team uses personal accounts for work, every person has to change the setting themselves, and you have no way to check. A business plan fixes the default for everyone and gives you admin control. That is the core point of our ChatGPT at work policy guide.

ChatGPT: turn off Improve the model for everyone

OpenAI's Data Controls FAQ says that when Improve the model for everyone is off, new conversations are not used to train OpenAI models. They still appear in chat history.

1. 1

   Open Settings

   Click your profile icon and choose Settings.

2. 2

   Go to Data controls

   Select Data controls.

3. 3

   Switch off Improve the model for everyone

   Turn it off and select Done. This also covers Codex tasks on a personal plan.

Temporary chats are not saved to history and, according to OpenAI, may be kept for up to 30 days for safety. For business use, OpenAI's enterprise privacy page states that ChatGPT Business and Enterprise data is not used for training by default, unless the workspace explicitly opts in, for example through feedback.

Claude, Gemini and Microsoft Copilot settings

Claude (Free, Pro, Max)

Anthropic's model improvement settings article says to open your settings from your name in the menu and switch off Help improve our AI models. If a user allows training, Anthropic says it may keep de-identified chats for up to five years. Incognito chats are not used for training even when the setting is on. For Team, Enterprise and the API, Anthropic says inputs and outputs are not used for training by default.

Gemini app (personal Google account)

Google's Gemini Apps Privacy Hub explains that while Keep Activity is on, Google uses activity to improve its services, including training generative AI models, and human reviewers read a subset of chats. With Keep Activity off, future chats are not used for training unless you send feedback, but are still kept for 72 hours. Change it at Gemini Apps Activity. Work Workspace accounts follow different terms: Google says prompts are not used for training without permission.

Microsoft Copilot

Microsoft's Microsoft 365 Copilot privacy page states that prompts, responses and data accessed through Microsoft Graph are not used to train foundation models. The consumer Copilot app, signed in with a personal Microsoft account, is different: Microsoft's Copilot privacy FAQ says users can opt out of model training in Copilot's privacy settings.

Canva, LinkedIn and other tools staff forget about

AI features now sit inside design, writing and social tools, not just chatbots. Two common ones:

• Canva: its Trust Center privacy page says user content is used to improve AI features only if it matches your Privacy controls (Profile, Your account, Privacy controls). For Business and Enterprise accounts the control is disabled and content is not used.
• LinkedIn: the LinkedIn generative AI FAQ says posts, profile data and other content may be used to train content-generating models unless the member turns off Data for Generative AI Improvement (Settings, Data privacy). Private messages are excluded.

Check meeting note-takers, grammar tools and browser extensions too. Many are free products with their own training terms, and they often see more confidential material than a chatbot does. Open-weight model apps from other providers can carry different risks again, as covered in our guide to DeepSeek, Qwen and open models.

What an AI training opt-out does not cover

Switching training off is worthwhile, but it is narrower than many people assume.

• Feedback buttons: Anthropic keeps the whole conversation for up to five years when you press thumbs up or down, and OpenAI and Google can use submitted feedback too.
• Safety review: chats flagged for misuse can still be reviewed and retained by all the major vendors, whatever the training setting.
• Past training: Anthropic and LinkedIn both say opting out does not affect training that has already happened.
• Storage: the vendor still stores your chats to provide the service, often in another country. An opt-out does not change where data lives.
• Personal data rules: the PDPA still applies to what staff paste in, training or not.

Staff checklist: AI training settings

Send this to anyone who uses AI tools for work. It takes about ten minutes per person.

Vendors rename settings and change defaults several times a year. Re-check this list each quarter, and whenever a tool announces new terms.

Do it yourself or get help?

A small team can work through the checklist in a staff meeting. The harder part is moving everyone onto company-owned accounts and setting admin controls so the defaults are enforced, not left to each person.

Our Security Awareness Training (S$390) covers safe AI use, what never to paste, and these settings for your tools. If you run Microsoft 365 or Google Workspace, our Microsoft 365 / Google Security Lockdown (S$590) sets the admin controls for Copilot or Gemini alongside the rest of your tenant security.