Turning off model training in ChatGPT is useful, but the setting does one specific job. It tells OpenAI not to use new conversations from your account to improve its models. It does not erase your chat history, disable Memory, delete your account, or make every interaction anonymous. Those controls are separate, and using the wrong one can leave data in places you meant to clear.
This guide follows OpenAI’s current documentation rather than treating “privacy mode” as a single switch. It explains the training preference, chat history, Memory, Temporary Chat, deletion, Voice, Codex, and the different defaults for business products and the API. Product labels and placement can change, so check the linked OpenAI pages if your interface differs.
Turn off training for a personal ChatGPT account
When you are signed in on the web, open your profile menu, select Settings, choose Data Controls, and turn off Improve the model for everyone. In the mobile app, open the sidebar, tap your profile icon, select Data Controls, and turn off the same setting. OpenAI’s Data Controls FAQ also documents a signed-out control under the question-mark menu on the web.
For a signed-in account, the preference syncs across web and mobile. You do not need to repeat the change on every device. OpenAI says that after you opt out, new conversations will not be used to train its models. Your ordinary chats can still appear in history, because history and training are different functions.
You can also submit a “do not train on my content” request through OpenAI’s Privacy Portal. OpenAI says it continues to honor earlier opt-outs submitted through support or its privacy form. If you manage more than one account, confirm the setting while signed into each one. An account-wide preference for one login does not establish the state of another login or a workplace account.
What the training opt-out changes
OpenAI says content from services for individuals, including ChatGPT and Codex, may be used to train models. The opt-out changes how new content is used for model improvement. It is best understood as a forward-looking training preference, not a deletion request for material already submitted.
There is an important feedback exception. Even after opting out, you can choose to rate a response with thumbs up or thumbs down. OpenAI says the entire conversation associated with feedback may then be used for training. Avoid submitting feedback from a thread that contains information you do not want included for that purpose. OpenAI also says support conversations may be used to improve its services, including models, when training is enabled in Settings.
The control does not stop OpenAI from processing a prompt to answer it, operating the service, enforcing its rules, or handling data for safety, security, and legal reasons described in its policies. OpenAI’s Privacy Policy says no internet or email transmission is fully secure or error free. Opting out reduces one use of new content; it is not a promise of secrecy, zero retention, or immunity from account compromise.
Training, history, Memory, and deletion are not interchangeable
Chat history is the list of conversations saved to your account. Turning off Improve the model for everyone does not remove those chats. They remain available unless you delete them or use a mode that does not save them to history.
Memory is personalization. When enabled, ChatGPT can use information from chats, files, and connected apps to shape later responses. Its controls are under Settings, Personalization, Memory, according to OpenAI’s Memory FAQ. Turning off training does not turn off Memory, and turning off Memory does not turn off training. For a deeper account of the current Memory system, see our ChatGPT Memory and controls guide.
Deleting a chat and deleting a memory are also separate. In the legacy saved-memories system, OpenAI says a saved memory is stored separately from chat history. Deleting the source chat therefore may not delete the saved memory. Removing a saved memory does not erase mentions already present in old conversations. Under the newer memory-summary system, fully removing a detail may require deleting every source where it appears, including current and archived chats, files, the memory summary, and connected-app data. The “Delete and turn off memory” command does not delete past chats.
Deletion addresses retained account content rather than the training preference. Delete individual conversations when you no longer want them in your history. Account deletion is a broader, permanent action documented separately by OpenAI. Exporting data gives you a copy; it does not delete the source. Before making an irreversible change, use the export option in Data Controls if you need records.
Use Temporary Chat for a conversation without history or Memory
Temporary Chat combines several useful limits for one conversation. OpenAI says these chats do not appear in history, do not use or create personalization memories, and are not used to improve its models. Start a new chat and select the pill-shaped Temporary button near the top-right of the page. The exact presentation can vary as the product changes.
Temporary does not mean instantly destroyed. OpenAI’s Temporary Chat FAQ says a copy may be kept for up to 30 days for safety purposes. The Data Controls FAQ says Temporary Chats may be reviewed only to monitor for abuse and are deleted from OpenAI’s systems after 30 days.
Temporary Chat still follows enabled Custom Instructions. OpenAI also says it may use limited information from prior conversations for safety and security in rare, high-risk situations, even though Temporary Chat does not use personalization Memory. If you use a GPT with actions, information sent through an action goes to a third party and is governed by that recipient’s privacy policy. That party may keep it longer than 30 days or use it for other purposes.
Voice data has more than one control
A Voice conversation can produce a transcript, other files, and audio or video clips. OpenAI’s Voice documentation says that when Improve the model for everyone is on, transcripts and other files from Voice conversations may be used for training, depending on the plan and settings. Turning that main setting off therefore matters for Voice transcripts too.
Raw clips have separate choices. OpenAI says it does not train on associated audio or video clips unless you choose to share them, including through the “Include your audio recordings” or “Include your video recordings” settings. In personal Free, Plus, and Pro workspaces, those clip-sharing choices require Improve the model for everyone to be on. Business, Enterprise, and Edu users cannot share Voice clips for training through those controls.
Retention also depends on the Voice experience. OpenAI says Live and Advanced Voice clips are stored with the transcript and retained for 30 days. Deleting the chat deletes associated clips within 30 days, subject to stated security, safety, or legal exceptions and a caveat for clips already disassociated after a user chose to share them. Standard Voice audio is deleted after transcription unless you chose to share audio for training. Archiving a Voice chat is not deletion.
Codex, business workspaces, and the API
For individual services, OpenAI groups ChatGPT and Codex under the same general training explanation. Its documentation says the ChatGPT training controls apply to ChatGPT conversations and Codex tasks. Codex also has separate controls for allowing training on full environments in Codex Settings. OpenAI specifically warns that changing the ChatGPT setting or using the Privacy Portal does not change those full-environment Codex settings. Codex users should review both places rather than assume one toggle covers every surface.
Business products have a different default. OpenAI says it does not train on inputs or outputs from ChatGPT Business, ChatGPT Enterprise, ChatGPT Edu, and the API Platform by default. An organization may explicitly opt in to share selected data, such as feedback or evaluation data. That business default does not mean data is never retained or reviewed. Workspace administrators may control access and retention, while API retention varies by endpoint and eligibility. OpenAI’s Enterprise Privacy page and API documentation should govern decisions for organizational data.
Do not move confidential work from a managed workspace into a personal account merely because both display the ChatGPT name. Different products, workspace settings, connected services, and administrator policies can apply. Follow your organization’s rules and confirm where a conversation is being created.
A practical privacy review
- Open Data Controls and set Improve the model for everyone to your intended choice.
- Review Memory separately, including the memory summary, legacy saved memories if shown, old chats, files, and connected apps.
- Check Voice clip-sharing controls if you use Voice, and inspect Codex full-environment settings if you use Codex.
- Use Temporary Chat when you do not want a conversation in history or personalization Memory, while remembering the up-to-30-day safety retention and third-party action limits.
- Delete chats or your account only when deletion is the actual goal. Export first if you need a copy.
Controls are most effective when paired with data minimization. Remove passwords, access tokens, account numbers, personal identifiers, confidential client material, and unnecessary health or financial details before sending a prompt. A training opt-out cannot protect information that another participant, connected app, compromised device, GPT action, or copied output exposes elsewhere. Our ChatGPT beginner guide also explains why important outputs still need human review.
FAQ
Does turning off Improve the model for everyone delete my chats?
No. OpenAI says regular conversations still appear in history after training is turned off. Delete a chat separately if you want it removed from history, or use Temporary Chat for a new conversation that should not appear there.
Is turning off Memory the same as opting out of training?
No. Memory controls personalization across conversations. Data Controls govern whether eligible conversations help improve models. Review both settings because changing one does not automatically change the other.
Are Temporary Chats completely private and immediately deleted?
No. They are not used for training, do not appear in history, and do not use or create personalization memories. OpenAI may keep a copy for up to 30 days for safety. Custom Instructions can still apply, and GPT actions can send data to third parties under separate policies.
Are ChatGPT Business and API data used for training by default?
OpenAI says no. Business offerings and the API are opted out by default unless an organization explicitly chooses to share data. Retention, administrator access, safety review, and legal obligations are separate questions, so consult the relevant workspace agreement and official documentation.