Do ChatGPT and Claude Train on Your Chats? Check These Settings

ChatGPT and Claude can use eligible consumer conversations for model training, depending on your settings and account. Paying for a personal subscription does not, by itself, establish that your chats are excluded.

The safest starting point is the account you actually use: personal or organisational, the product or feature involved, and the current training preference. Do not rely on a screenshot of a policy from before a service changed its terms.

ChatGPT’s model-training control

OpenAI’s data-controls guide explains that disabling its model-improvement setting excludes new conversations from training while allowing them to remain in history. On the web, look under your account settings and Data controls.

The same guide distinguishes model training from memory and from deletion. It also says temporary chats are not used for model improvement while they remain temporary, but may be retained for up to 30 days for safety. Saving one turns it into a regular chat governed by the relevant account settings.

These are the provider’s stated controls, not an independent audit of its backend processing. Read the current documentation if the setting names or behaviour differ in your account.

Claude’s training choice

Anthropic’s consumer-terms update covers Free, Pro and Max accounts, including associated Claude Code use. It describes a model-training choice available in privacy settings and an extended five-year retention period for eligible data when training is allowed.

Anthropic distinguishes those consumer services from services under its commercial terms. The update also explains that turning training off affects future training runs; it does not undo training already started or completed.

A claim that “Claude never trains on consumer chats” is therefore too broad. So is a claim that every Claude product follows identical terms.

Separate the questions before comparing providers

A training opt-out should not be described as a promise of zero storage, zero access or instant erasure. Those are different assertions requiring their own evidence.

A practical account check

Open the actual workspace where you chat. Check the training preference, review any memory or personalisation settings that matter to you, and read the deletion or retention information separately. Repeat the check if you switch accounts, plans or products.

For confidential work, use the service and account your organisation has approved. A consumer privacy toggle is not a substitute for checking whether a document should be uploaded at all. You can often ask a useful question using a short fictional example instead of an identifiable customer record.

Privacy is one part of an ethics comparison

Our OpenAI–Anthropic comparison considers the wider record. You can also inspect OpenAI’s score and Anthropic’s score.

Those company-level assessments do not determine your account settings. Use them alongside the current product policy, and record when you checked it: a service’s reputation is a poor replacement for a specific, verifiable privacy choice.

Sources checked on 24 September 2026. Ownership and policies can change after this review.