How to Turn Off AI Training on Your Data (ChatGPT, Claude, Gemini, Microsoft Copilot, Grok and More)

This guide shows where the opt-out setting lives in the major AI tools, and which business plans say they do not train on your data by default. We checked each vendor's official help or privacy page on October 24, 2026. Menus change often. If a step here does not match your screen, open the official link for that tool.

Read this first

• Free and personal accounts are the main risk. Business and enterprise plans usually have stronger default protections. Personal accounts usually do not.
• Staff using personal accounts is the biggest risk. You can set up a business account and still have people using their own free account on the side. Settings on one account do not change another.
• Turning off training is not the same as the vendor not storing your data. Most vendors still keep chats for a period of time. Many still allow safety review, legal holds, or review of chats you flag or rate. Turning training off usually applies to new chats only. It does not undo anything already used.
• Do not paste regulated or client data into an AI tool your firm has not approved. A setting is not a safe substitute for a policy.

Summary table

ChatGPT (OpenAI)

Personal plans (Free, Plus, Pro): sharing your chats to improve the model is on by default.

Turn it off on the web:

1. Open your account menu and choose Settings.
2. Choose Data controls.
3. Turn off Improve the model for everyone.
4. Select Done.

On iOS or Android: open the sidebar, tap your profile icon (this opens Settings), choose Data controls, then turn off Improve the model for everyone.

Good to know:

• The setting follows the signed-in account on web and mobile. If you use ChatGPT signed out, change it in Settings; that choice is saved in that browser only.
• OpenAI also offers a Privacy Portal option called "Do not train on my content."
• Turning it off does not delete your existing chats.
• Temporary Chat chats are not saved to history, do not create memories, and are not used to improve models. OpenAI says they may be kept up to 30 days for safety.
• Thumbs up or down feedback can let OpenAI use that conversation.
• Voice mode and some other features have their own separate controls. Check the linked page.

Business plans: OpenAI says that by default it does not use content from ChatGPT Business, Enterprise, Edu, or Healthcare workspaces to train its models. It says the same for inputs and outputs sent through the API Platform. Workspace admins control retention. If someone explicitly opts in to share data (for example through feedback), that data can be used.

Sources: Data controls FAQ · Enterprise privacy at OpenAI

Claude (Anthropic)

Free, Pro and Max plans (including Claude Code on those plans):

1. Select your name from the settings menu.
2. Choose Settings, then Privacy.
3. Under Help Improve our AI models, turn the toggle off.

The mobile app uses the same path: your name → Settings → Privacy.

Good to know:

• This applies to new chats and coding sessions. Data already used, or already part of a trained model, stays.
• Anthropic says chats flagged by its safety systems may still be used.
• Incognito chats are not used for training.
• If you give thumbs up or down feedback, Anthropic stores that whole conversation for up to 5 years and may use it for training.
• Programs you explicitly join, such as testing programs, can involve training.

Business plans: Anthropic says that by default it does not use inputs or outputs from its commercial products (Claude for Work, the Anthropic API, Claude Gov and similar) to train its models. The exceptions it lists are feedback or bug reports you submit and anything you explicitly choose to allow. Team and Enterprise owners can turn off the thumbs up/down feedback button under Organization settings → Data and Privacy.

Sources: Change your model improvement settings · Consumer plans: is my data used for training? · Commercial plans: is my data used for training?

Gemini (Google)

Personal Google accounts (adults 18+): the Keep Activity setting is on by default.

Turn it off on a computer:

1. Go to gemini.google.com and open Settings & help.
2. Choose Activity.
3. Near the top, select On, then Turn off (or Turn off and delete activity).

On iPhone or iPad: open the menu, tap your profile picture or initial, choose Gemini Apps Activity, then On, then Turn off.

Good to know:

• With activity off, Google still saves chats for up to 72 hours to provide the service and for safety.
• Chats that human reviewers have already looked at can be kept for up to 3 years, disconnected from your account. Deleting your activity does not delete them.
• Feedback you submit can be used.
• Some features that use connected Google apps are not available when activity is off.
• Google warns not to enter confidential information you would not want a reviewer to see.
• Temporary chats are not used to train models and are kept 72 hours.
• Live audio and video improvement is a separate setting.

Work or school accounts: Gemini activity settings are controlled by your Workspace administrator, not by each user.

Google Workspace: Google says Workspace does not use customer data to train generative AI models outside the customer's domain without prior permission or instruction. Ask your admin to confirm which Gemini features are turned on for your firm.

Sources: Gemini Apps Activity · Gemini Apps Privacy Hub · Generative AI in Google Workspace privacy hub

Microsoft Copilot

Microsoft has two very different situations. Make sure staff know which one they are in.

Copilot with a work or school account (Microsoft 365 Copilot and Copilot Chat)

Microsoft says prompts, responses, and data accessed through Microsoft Graph are not used to train foundation models. There is no per-user training switch to flip.

What your firm should do instead:

• Confirm staff sign in with their work account (look for the "Protected" green shield in Copilot Chat).
• Remember that Copilot only shows people data they can already open. If permissions are too loose, Copilot makes that easier to see. See our article Will Microsoft Copilot Show Employees Files They Shouldn't See?.
• Microsoft says Copilot interactions are kept and are available for audit and eDiscovery through Microsoft Purview.

Sources: Enterprise data protection in Copilot · Microsoft 365 Copilot privacy · Copilot privacy and protections

Copilot with a personal Microsoft account (new app, from August 18, 2026)

Microsoft's current page says prompts, responses and your file contents in the Microsoft Copilot app aren't used to train foundation models. Optional feedback may be used to improve Copilot but not to train the foundation models. Because of that, the new app has no training switch. You can still manage chat history and personalization:

1. Open Copilot and go to Settings.
2. Use Personalization to control saved memories and related options.
3. To export or delete all your Copilot activity history, use the Microsoft privacy dashboard (linked below).

Sources: Copilot for individuals: your activity history · Copilot privacy controls (new app)

If you still use the older personal Copilot app

Microsoft says that if you have not updated to the new version, its older privacy articles apply. In the older app:

1. On copilot.com, select your profile icon and your profile name, then Privacy. In the Windows or macOS app, use profile icon → Settings → Privacy. On mobile, use menu → profile icon → Account → Privacy.
2. Turn off Training on conversation activity (and Training on voice conversations if shown).

This covers future conversations only. Microsoft says it does not stop other uses such as safety, security, or compliance, and human review cannot be opted out of. Microsoft lists some groups that are excluded from training already, including work and school accounts.

Source: Older app: Copilot privacy controls · Older app: Privacy FAQ

Grok (xAI)

Grok app and grok.com: content may be used for training by default.

1. In the mobile app, go to Settings → Data Controls and turn off Improve the model.
2. On grok.com, go to Settings → Data and turn off Improve the Model.

Good to know:

• The opt-out covers new conversations.
• Private Chat (ghost icon, top right) is not used for training. xAI says those conversations are deleted within 30 days.
• Feedback you give voluntarily may be used for training.
• Authorized staff may review conversations.
• xAI says that if you use Grok signed out in some regions (not the EU or UK), there is no opt-out. Sign in and use the setting.

Grok inside X (the social network): xAI's FAQ points to the X Help Center for those settings. We could not open the X Help Center page when we checked, so we have not verified the exact menu path. Start at the xAI Consumer FAQ and X's own Help Center.

Business and enterprise: xAI's Enterprise FAQ says it does not use business data, including prompts or outputs, to train its models. It also says it may offer free credits in exchange for agreeing to training, so read any offer before you accept.

Sources: xAI Consumer FAQ · xAI Enterprise FAQ

Perplexity

For Free, Pro and Max accounts, "AI data retention" is on by default.

1. Open your Account settings.
2. Go to Preferences.
3. Find the AI data retention toggle. If it is on, your data can be used for AI training.
4. Turn it off.

Good to know:

• It applies to data collected after you opt out. Previously collected data cannot be removed.
• The setting is per user.
• Perplexity says data may still be processed to run the service, meet legal requirements, and improve the product.

Business plans: Perplexity says Enterprise data is never used for AI training. If an Enterprise plan is downgraded to a non-Enterprise plan, consumer defaults apply and each person must opt out.

Source: Data Collection at Perplexity

Meta AI

This is the one we could not give you a clean opt-out for. Meta's transparency page says the information it uses to develop and improve its AI includes publicly shared Facebook and Instagram posts and people's interactions with its AI features. We did not find an official Meta help page that documents a simple "turn off training" switch for Meta AI chats for US users.

What Meta does document:

• Delete your Meta AI chats and media: in the Meta AI app, tap Menu → Settings → Data & privacy → Manage your information. From there you can export your information, delete all chats and media, or remove public posts you shared. Deleting is not the same as opting out of training.
• EU and UK: people there have a right to object to some uses of their information and can submit an objection through Meta's Privacy Rights Request channel. Meta may deny an objection if it has compelling grounds.

For firms: do not use Meta AI for client or confidential information.

Sources: AI at Meta: training data transparency · Manage your information on Meta AI and Vibes · How to object on Meta products

GitHub Copilot

For individual Copilot Free, Pro, Pro+ and Max plans, GitHub says that starting April 24, 2026 it may use your interactions (inputs, outputs, code snippets and context) to train and improve AI models.

Turn it off:

1. On GitHub, click your profile picture (upper right), then Copilot settings.
2. Find Allow GitHub to use my data for AI model training and set it to Disabled.

The setting belongs to each personal GitHub account. If you do not see it, you may be signed in with a Copilot Business or Enterprise license.

Business plans: GitHub says it does not use Copilot Business or Copilot Enterprise customer data to train AI models, under its Data Protection Agreement, and the setting is not shown for these plans.

Source: Managing GitHub Copilot policies as an individual subscriber

Mistral (Le Chat, called "Vibe" in Mistral's current help pages)

Mistral says that in certain cases your inputs and outputs may be included in its training programs, and that you can opt out at any time. Personal Vibe users are not opted out by default.

On the web (Admin panel):

1. Select Vibe under Manage.
2. Under Privacy, turn off Allow your interactions to be used to train our models.

On iOS or Android: Settings → Data & Account Controls (under Account) → untick Enable data sharing.

Good to know: the Vibe toggle and the API (Mistral Studio) toggle are separate. For the API, open Privacy in the Admin panel and turn off the toggle under Anonymous improvement data.

Business plans: Mistral says Enterprise customers are opted out by default, and Teams administrators can disable training for the whole organization.

Source: Mistral: Can I opt out of training?

DeepSeek

DeepSeek's Terms of Use (last updated March 27, 2026) say it may use inputs and outputs, with de-identification, to provide and improve its services. They say that if you do not want that, you can turn off Improve the model for everyone. The Terms do not give a menu path, and we did not find an official page that does, so we cannot give exact steps here. DeepSeek is operated by a company in mainland China and its Terms are governed by PRC law. Many firms will want to keep client data out of it entirely.

Source: DeepSeek Terms of Use

What to do at your firm

1. Use business accounts for anything that touches client work. Business and enterprise plans from the vendors above say they do not train on your data by default. Free accounts often do.
2. Set an approved-tools policy. List the AI tools staff may use, and which kinds of information may go into each. Say plainly that personal accounts are not approved for client data. See our AI acceptable-use policy guide.
3. Never paste regulated or client data into an unapproved tool. That includes tax IDs, health information, account numbers, privileged legal material and anything covered by an NDA.
4. Have staff turn off training on any personal account they already have, using the steps above. This lowers risk. It does not make a personal account approved.
5. Review settings every quarter. Vendors rename menus and change defaults. Re-check the linked pages and your admin console.
6. Ask your admin or IT provider to confirm which AI features are on for your Microsoft 365 or Google Workspace tenant, and who can use them.

For a broader look at what is realistic for law, accounting and financial firms, see Practical AI for Law, Accounting, and Financial Firms.

FAQ

Does turning off training delete my chats?

No. In most tools, turning off training only changes how future chats are used. Your history stays until you delete it, and the vendor may keep copies for a set period for safety or legal reasons. Data that was already used for training usually cannot be pulled back.

If we pay for a business plan, are we safe to paste client data into it?

A business plan that says it does not train on your data by default is a much better choice than a free account. But "not used for training" is not the same as "approved for your regulated data." Check your own contracts, client agreements and regulator rules, and make sure admin settings such as retention and sharing are set the way you want. Your firm's policy should say what is allowed.

What if our staff already used their personal accounts?

Do not panic. Ask them to turn off training in those accounts, stop using them for client information, and move that work to an approved business tool. If real client data was pasted in, talk to your compliance lead about whether any notice is required.

Related Blue Tree Technology® pages

• Will Microsoft Copilot Show Employees Files They Shouldn't See?
• AI acceptable-use policy
• Practical AI for Law, Accounting, and Financial Firms
• Cybersecurity services

Verification sources

All pages were checked on October 24, 2026. These are the vendors' own pages. Menus and defaults change, so always check the page for the latest.

• OpenAI: Data controls FAQ · Enterprise privacy
• Anthropic: Model improvement settings · Consumer training FAQ · Commercial training FAQ
• Google: Gemini Apps Activity · Gemini Apps Privacy Hub · Workspace generative AI privacy hub
• Microsoft: Copilot activity history (new app) · Copilot privacy controls (new app) · Older app privacy controls · Enterprise data protection · Microsoft 365 Copilot privacy
• xAI: Consumer FAQ · Enterprise FAQ
• Perplexity: Data Collection at Perplexity
• Meta: AI at Meta transparency · Manage your information on Meta AI · How to object
• GitHub: Managing Copilot policies
• Mistral: Privacy and data controls
• DeepSeek: Terms of Use

Last verified: October 24, 2026. This guide is general information, not legal advice.

Still have a question?

Ask Little Sprout for general information. To talk about which AI tools fit your firm and how to set them up safely, schedule a free 15-minute discovery call.