Opting out of AI training changes whether a provider may use eligible future conversations to improve its models. It does not usually delete your chat history, erase old training runs or disable safety review. The control also sits in a different place for ChatGPT, Claude and Gemini.
This guide shows how to opt out of AI training on the current consumer versions of those three services and what each switch leaves behind.
Documentation checked October 31, 2026. Settings can vary by account type, age, region, managed workspace and app version.
ChatGPT: turn off model improvement
On a personal ChatGPT Free, Plus or Pro workspace:
- Open your profile menu and choose Settings.
- Select Data controls.
- Turn off Improve the model for everyone.
OpenAI’s Data controls documentation says you can keep chat history while this setting is off. The choice applies to eligible new activity rather than deleting conversations already stored in the account.
If you are signed in, the setting follows your account across the web and mobile apps. Managed Business, Enterprise, Edu and Healthcare workspaces use organization-level policies, so the personal toggle is not the whole answer there.
There are two exceptions worth remembering. First, providing feedback with a thumbs-up or thumbs-down may allow the associated conversation to be used for training. Second, OpenAI documents a separate training control for voice mode. Review both if you use them.
For a wider account-level view, see the ChatGPT privacy reference.
Claude: disable Help Improve our AI models
For Claude Free, Pro and Max accounts:
- Open your name or profile menu.
- Choose Settings.
- Select Privacy.
- Turn off Help Improve our AI models.
Anthropic’s model-improvement privacy instructions say the change stops new chats and coding sessions from being used in future model-training runs. It also stops previously stored chats and coding sessions from being used in future runs.
That is not retroactive deletion. Anthropic says data already included in training that has begun, or in models already trained, remains part of that work. Conversations flagged by safety classifiers may also still be used for internal trust-and-safety models, policy enforcement or safety research.
These instructions cover consumer products, including Claude Free, Pro and Max and those plans’ use of Claude Code. Anthropic directs commercial customers using Claude for Work or the API to separate commercial-product documentation.
The Claude privacy reference collects the related retention and account controls.
Gemini: turn off Keep Activity
In Gemini Apps, the relevant consumer control is Keep Activity:
- Open Gemini.
- Open the account or settings menu.
- Go to Gemini Apps Activity.
- Turn Keep Activity off.
Google’s Gemini Apps Privacy Hub says future chats are not used to improve its AI models when Keep Activity is off, provided you do not submit feedback. Temporary chats are also excluded from model training.
Google still retains temporary chats and chats made while Keep Activity is off with your account for 72 hours. It says this period supports responses, short-term context, safety and protection against failures. Turning the switch off therefore does not mean an interaction vanishes immediately.
Feedback is another separate path. If you submit it, Google may collect the feedback, associated context and content needed to understand it. The Privacy Hub also says trained human reviewers examine some collected data for service, safety and improvement purposes. Avoid entering confidential material that you would not want reviewed.
Read the Gemini privacy reference for the connected-app and retention details that the activity switch does not replace.
The switches are similar, but not identical
All three controls reduce the use of eligible future consumer activity for model improvement. Their names and boundaries differ:
- ChatGPT uses Improve the model for everyone. History can stay on, while feedback and voice have separate considerations.
- Claude uses Help Improve our AI models. Anthropic distinguishes future training runs from training already under way or completed.
- Gemini uses Keep Activity. Chats made with it off still have a documented 72-hour account retention window, and feedback changes the rule.
Do not assume that one product’s wording carries over to another. Recheck the provider’s own page after a major app update.
Opting out is not deletion
A training preference controls a purpose for using data. Deletion controls whether a saved conversation or account record remains. They are separate actions.
Turning training off generally does not:
- delete existing chat history;
- erase a model that was already trained;
- remove a conversation from safety systems immediately;
- delete saved memories, files or connected-app records;
- override a managed-workspace retention policy.
If you also want to remove stored content, use the product’s chat or account deletion tools. For ChatGPT, our guide to deleting ChatGPT history explains why archives, memories and separately saved files need their own checks.
Use temporary modes for sensitive one-off work
Temporary modes can reduce what becomes part of normal history, but they still have documented retention and safety boundaries.
OpenAI says Temporary Chat does not appear in regular history, create or update memories, or train its models; it may be retained for up to 30 days for safety. Google says Gemini temporary chats are not used to train its AI models and are retained with the account for 72 hours.
Those modes are useful for limiting future account history. They are not a substitute for reading the provider’s privacy terms, removing uploaded files or following an employer’s rules for confidential information.
A quick privacy check
After changing a setting:
- Reload the app and confirm the toggle still shows the intended state.
- Check whether you are in a personal or managed workspace.
- Review separate feedback, voice, audio or live-sharing controls.
- Avoid submitting feedback from a sensitive conversation.
- Use deletion tools separately when removal is the goal.
- Revisit the setting after a major product or account change.
The practical rule is simple: treat training, history, memory, feedback and deletion as separate controls. Turning off one does not silently complete the others.
