Stop ChatGPT, Claude and Gemini training on your business data

To stop your chats training the AI, switch off one setting per tool: Improve the model for everyone in ChatGPT, Help improve our AI models (Model Improvement) in Claude, and Keep Activity in Gemini. On the free and personal paid plans all three are on by default. The business plans (ChatGPT Business and Enterprise, Claude for Work, Gemini in Google Workspace) don’t train on your data unless you opt in.

Below are the exact clicks as of October 2026, then what the switch doesn’t do. Read that second part before you tell your team they’re covered.

Which plans train on your chats by default?

Tool	Trains by default	Off by default
ChatGPT	Free, Plus, Pro	Business, Enterprise, Edu, API
Claude	Free, Pro, Max (and Claude Code on those plans)	Claude for Work (Team, Enterprise), API
Gemini	Personal Google accounts (Keep Activity is on by default if you’re 18 or over)	Work accounts in Google Workspace

The split follows who pays. If staff signed up with a personal email, even on a paid plan, assume their chats can be used for training until someone checks the setting.

“Training” here means the company may use your conversations to build future versions of its large language model. Once your text is in a trained model, you can’t get it back out.

How do I stop ChatGPT using my data for training?

1. Click your profile picture, then Settings.
2. Open Data Controls.
3. Switch Improve the model for everyone off.

On the phone app it’s in the same place: open the sidebar, tap your profile, then Data Controls. The setting follows your account, so doing it once covers every device you’re signed in on.

Your chat history stays as it was. OpenAI says new conversations simply won’t be used for training. If you only want one conversation kept out, start a Temporary Chat instead: it isn’t saved to your history or memory, isn’t used for training, and is kept for up to 30 days for safety checks before it’s deleted.

How do I stop Claude using my data for training?

1. Click your name, then Settings.
2. Open Privacy (or go straight to claude.ai/settings/data-privacy-controls).
3. Switch Help improve our AI models off.

The phone app has the same path. Anthropic’s help pages call this setting Model Improvement, so you may see both names.

Two details from Anthropic’s privacy pages are worth knowing. With the setting on, new and resumed chats can be kept for up to five years for training. With it off, your chats aren’t used for new training runs, but anything already in a run that has started, or a model already trained, stays in. Incognito chats are never used for training, even with the setting on.

How do I stop Gemini using my data for training?

1. Go to gemini.google.com.
2. Click Settings & help, then Activity.
3. Near the top, click On, then Turn off (or Turn off and delete activity to clear your history at the same time).

This one costs you something, unlike the other two. With Keep Activity off, Gemini stops saving your chats to your history, and some connected apps, including the Google Workspace ones, stop working. If you’d rather keep your history, use a Temporary chat for anything sensitive. Google says those aren’t used to train its models and are kept for 72 hours.

On a work account, your Google Workspace admin controls these settings and you may not be able to change them. Google says Workspace doesn’t use customer data to train its models without the customer’s permission.

Does switching it off make my chats private?

No. It stops one use of your data. Four things carry on:

• Short-term storage. ChatGPT temporary chats are kept for up to 30 days. Gemini keeps chats for 72 hours even with Keep Activity off. Deleted Claude chats leave Anthropic’s systems within 30 days.
• Safety review. All three can still check conversations for misuse. Claude chats flagged for a policy breach can be kept for up to two years, and Gemini chats already seen by human reviewers are kept for up to three years, separated from your account.
• The thumbs-up and thumbs-down buttons. Rating a reply can send the whole conversation back for training. Anthropic keeps feedback for five years. Google says chats with Keep Activity off can still be used for training if you send feedback. On sensitive chats, leave the buttons alone.
• What’s already been used. No setting pulls your chats out of a model that’s already trained, and OpenAI and Google describe their switches as covering new chats. Delete old chats you’d rather weren’t stored.

You still have to decide what goes into an AI tool in the first place.

Should a small business just pay for a business plan?

If your team pastes in client details, contracts, HR notes or anything under an NDA, yes. The business plans exclude your data from training by default and put the accounts under one admin, so you’re not relying on each person’s settings. OpenAI’s business terms also state that its staff only access conversations to resolve incidents, recover chats with your permission, or where the law requires it.

If your team only uses AI for public-facing work like social posts, blog drafts or rewriting your own website copy, personal plans with training switched off are usually fine. Write down which kind of work goes where, so nobody has to guess.

A 15-minute check for your team

1. List every AI tool your team uses and whether each account is personal or business. Ask; don’t assume.
2. Have everyone on a personal account switch training off using the steps above, and screenshot the setting.
3. Agree what never goes into a personal account: client names, financial details, anything covered by a contract.
4. Tell people to use Temporary or incognito chats for one-off sensitive questions, and to skip the rating buttons on those chats.
5. Put a reminder in the diary for three months’ time. These settings and their names change, so check them again then.

What to do next

Once the settings are sorted, put the tools to work on content that’s safe to share. The brand voice guide playbook shows you how to make ChatGPT, Claude or Gemini write like your business, and the ChatGPT content calendar playbook gets a month of social posts planned in one sitting.