Does AI train on your data? ChatGPT, Claude, Copilot and Gemini, plan by plan

Business plans from OpenAI, Anthropic, Microsoft and Google don't train AI models on your content by default. Consumer plans are different: ChatGPT Free and Plus train on your conversations unless you opt out, and Claude Free, Pro and Max train on them if you allow it. The biggest exception on business plans is feedback: a thumbs up or down can send the whole conversation to the vendor.

The table below shows the default for each plan as of October 2026, with the setting names each vendor documents.

AI training defaults by plan (October 2026)#

As of October 2026, from each vendor's documentation (see Sources). Vendors change these policies, so check before you rely on them.

Does ChatGPT train on your data?#

It depends on the plan.

• Free, Go, Plus and Pro: OpenAI's help centre says that for services for individuals "we may use your content to train our models". To opt out, turn off Improve the model for everyone under Settings > Data controls, or select Do not train on my content in OpenAI's Privacy Portal. Either works for ChatGPT conversations and Codex tasks. Opting out applies to new conversations.
• Temporary chats aren't used to improve OpenAI's models while they remain temporary.
• Feedback overrides your opt-out. If you give a thumbs up or down, OpenAI says "the entire conversation associated with that feedback may be used to train our models", even if you've opted out.
• Business, Enterprise, Edu and the API: OpenAI says it doesn't use inputs or outputs from these plans to improve its models by default. For the API, this has applied since 1 March 2023. API organisation owners can enable data sharing, which lets members share feedback and Playground chats with OpenAI.

The practical risk for most organisations isn't the business plan. It's staff pasting work content into a personal ChatGPT account with training switched on. The OAIC recommends, as best practice, that organisations don't enter personal information, particularly sensitive information, into publicly available generative AI tools.

Does Claude train on your data?#

Anthropic changed its consumer terms on 28 August 2025.

• Free, Pro and Max: Anthropic uses your chats and coding sessions to improve its models if you allow it. New users choose during sign-up, and Anthropic's pricing page lists these plans as "Opt-out". You can change it any time under Settings > Privacy > Help improve our AI models. If you allow training, Anthropic keeps your data for up to five years; if you don't, it keeps the existing 30-day retention period.
• Turning it off stops new chats being used in future training. Data already used in training runs or trained models stays there.
• Incognito chats aren't used for training, even with the setting on. Conversations flagged by safety classifiers may still be used for safety work.
• Team, Enterprise and the API: Anthropic says it won't use inputs or outputs from its commercial products to train models by default. The exception is feedback: a thumbs up or down stores the whole conversation for up to five years, and it may be used for training. Primary Owners and Owners of Team and Enterprise plans can turn off the Rate chats setting under Organization settings > Data and Privacy.
• Amazon Bedrock: AWS says inputs and outputs aren't used to train models and aren't shared with model providers. Anthropic's consumer changes explicitly didn't apply to API use through Bedrock.

Does Copilot use your data for training?#

Microsoft's position for work accounts is consistent:

• Microsoft 365 Copilot: "Prompts, responses, and data accessed through Microsoft Graph aren't used to train foundation LLMs, including those used by Microsoft Copilot."
• Copilot Chat: Microsoft says prompts and responses aren't used to train the underlying foundation models. When enterprise data protection applies, Copilot Chat shows a green shield next to the New Chat button.
• Feedback: Microsoft may use optional feedback to improve Copilot, but says it doesn't use that feedback to train the foundation models. Admins can manage feedback for their organisation.
• Anthropic models with Data Retention: Some advanced Claude models in Copilot require Anthropic to retain data under Anthropic's own terms rather than Microsoft's. They're off by default and need a separate admin opt-in. Microsoft says Anthropic doesn't use that retained data for training without your express permission.

Does Gemini train on your data?#

For Google Workspace, Google says Gemini doesn't use your prompts, Workspace content or responses to train generative AI models without your permission, and points to the "Training Restriction" in its Service Specific Terms. Feedback is voluntary, can be kept for up to 18 months, and isn't used to train the models behind Workspace's generative AI services. On Google Cloud, Google says it won't use your data to train or fine-tune any AI/ML models without your prior permission or instruction.

Training isn't the only question#

A "no training" commitment doesn't mean your data is never stored, read or moved. Also ask:

• How long is it kept? Retention ranges from zero data retention on approved API accounts to indefinite chat history set by an admin.
• Who can review it? Abuse monitoring can involve human review. OpenAI keeps API abuse monitoring logs for up to 30 days by default.
• Where is it stored and processed? See our guide to AI data residency in Australia, or compare options in the AI data residency checker.

How to check and lock down training settings#

• List the accounts staff actually use, including personal ChatGPT, Claude and Gemini accounts used for work.
• Move work onto business plans where training is off by default, and record that in your AI policy.
• Decide on feedback buttons. Turn off Claude's Rate chats setting or Microsoft feedback if feedback shouldn't leave your organisation, and tell staff what a thumbs up shares.
• Check new models and features. Vendors sometimes attach different terms to new models, as Microsoft does for Anthropic models with Data Retention.
• Re-check after policy updates. Anthropic's 2025 consumer change shows defaults can move.

AusGPT doesn't use your prompts, documents or responses to train AI models. See our security page for how we handle your data.