If you don’t want OpenAI to train its models on what you type into ChatGPT or Codex, open Settings > Data controls in ChatGPT and turn off “Improve the model for everyone”. That one switch covers ChatGPT chats and Codex tasks on Free, Plus and Pro personal accounts. But it is not the whole story: OpenAI’s help pages list several things that switch does not cover. Clicking thumbs up or thumbs down can still send the whole conversation for training. Codex has a separate “Include environments” switch. Voice recordings have their own toggles. And OpenAI’s Data controls help page, updated on October 30, 2026 (US time), now says content from Daybreak cybersecurity sessions on a personal account may be used for training even when the switch is off. This guide walks through every setting to check, in the order that matters, with nothing to install and no code.

Everything here was checked on October 31, 2026 against OpenAI’s Help Center articles Data controls in ChatGPT, How your data is used to improve model performance, ChatGPT Voice, Troubleshooting OpenAI Daybreak and Using Daybreak in Codex CLI.

Key facts:

• On by default: for personal ChatGPT Free, Plus and Pro accounts, data sharing for model training is enabled by default. (Data controls)
• One switch covers chats and Codex tasks: turning off “Improve the model for everyone”, or choosing “Do not train on my content” in the Privacy Portal, is enough for both. You don’t need to do both. (Model improvement)
• It only affects new conversations. (Model improvement)
• Feedback overrides it: a thumbs up or down can put the entire conversation into training on Free, Plus and Pro, even after you opt out.
• Daybreak exception: during Daybreak (Trusted Access for Cyber) sessions on a personal account, OpenAI “may use content from your interactions to improve our models even when ‘Improve the model for everyone’ is off.” (Data controls)
• Business, Enterprise, Edu and the API are not trained on by default.

Who needs to do this

Your account	Trained on by default?	What to do
ChatGPT Free, Plus or Pro (personal)	Yes	Follow every step below
Signed out of ChatGPT	Yes, unless you change it in that browser	Step 1, signed-out version
ChatGPT Business, Enterprise, Edu, ChatGPT for Healthcare	No, unless your organization signed a separate agreement	Nothing for training; your admin’s retention rules still apply
OpenAI API	No, by default	Nothing, unless your API org owner turned on data sharing
Teen account linked to a parent	A parent or guardian may control this setting	Ask the parent or check Parental controls

Sources for the table: Data controls in ChatGPT and How your data is used to improve model performance.

OpenAI’s pages name Free, Plus and Pro for the default. They don’t spell out ChatGPT Go in that sentence, so if you’re on Go, open the same settings screen and check what the switch shows rather than assuming.

Step 1: Turn off “Improve the model for everyone”

This is the main switch. It applies to your whole account once you’re signed in, on web and on the phone apps.

On the web:

1. Click your account menu (your name or picture).
2. Select Settings.
3. Select Data controls.
4. Select Improve the model for everyone, turn it off, and select Done.

On iPhone or Android:

1. Open the sidebar.
2. Tap your profile icon to open Settings.
3. Tap Data controls.
4. Turn off Improve the model for everyone.

If you use ChatGPT without signing in: open Settings and turn the same switch off. That choice is saved only in that browser, so you’d have to do it again in another browser or after clearing its data.

Alternative route: OpenAI’s Privacy Portal has a Do not train on my content option. OpenAI says either route is enough for ChatGPT conversations and Codex tasks.

What turning it off does not do: it does not delete or hide your chats. Regular chats stay in history, archived chats stay archived, and project chats stay in their projects. Archiving a chat also doesn’t change whether it can be trained on; the switch is what decides. If you want something gone, delete the chat or the account separately.

Step 2: Stop rating answers, or accept what it means

This is the exception most people miss. On Free, Plus and Pro, if you click thumbs up or thumbs down on a response, “the entire conversation associated with that feedback may be used to train our models, even if you have opted out of model training” (OpenAI).

So the rule is simple: in any chat you’d rather keep out of training, don’t press the thumbs. One rating counts for the whole conversation, not just that reply.

On Business, Enterprise and Edu this is different: feedback does not permit training on the conversation, and workspace admins can turn feedback off under Workspace settings > Allow response feedback in the Admin Console.

Step 3: Check Codex’s separate “Include environments” switch

If you use Codex, the agent that works on code for you inside ChatGPT, there’s a second switch. Your Codex tasks follow the main ChatGPT setting from Step 1. But Codex has its own Include environments setting in Codex’s settings. It controls whether extra context from your Codex environments, the cloud workspaces where Codex runs your code, can be used for training.

OpenAI is explicit that changing your ChatGPT setting or using the Privacy Portal does not change this one (Data controls). Open Codex’s settings, find Include environments, and turn it off if you don’t want environment context used.

If you can’t find the switch, for example because you’ve never used Codex in the cloud, don’t assume that means it’s off. OpenAI doesn’t document a default for it. The safe move is to ask OpenAI support through the Help Center what the setting is on your account.

Step 4: Check the voice toggles

Voice has its own controls in the same Settings > Data controls screen. According to the ChatGPT Voice help page:

• Audio and video clips from voice chats are not used for training unless you chose to share them with Include your audio recordings or Include your video recordings. Those toggles only appear once “Improve the model for everyone” is on.
• But the transcripts of your voice chats count as normal chat content. If “Improve the model for everyone” is on, they may be used.
• If you stop sharing clips, new clips won’t be used. Clips that were already disassociated from your account may still be used.

So if you finished Step 1, transcripts are covered. If you once turned on audio or video sharing, check those toggles too.

Step 5: Use Temporary chat for one-off sensitive questions

If you normally keep training on but want a single conversation kept out, start a Temporary chat. While a chat stays temporary, OpenAI says it:

• doesn’t appear in your chat history,
• doesn’t create or update memories,
• isn’t used to improve OpenAI models,
• may be kept for up to 30 days for safety purposes.

The catch: if you save a temporary chat, it becomes a regular chat and follows your normal training setting. And remember Step 2: a thumbs rating is feedback, so avoid it there too.

The new exception: Daybreak sessions

This one only affects a small group, but OpenAI’s Data controls page spells it out in its October 30, 2026 update, and it drew attention this week, so it’s worth understanding.

Daybreak is OpenAI’s Trusted Access for Cyber program. Approved security professionals get models that refuse fewer cybersecurity requests, in tiers called Daybreak Blue and Daybreak Red. To get it as an individual you need an eligible paid plan, Advanced Account Security and at least one physical FIDO2 hardware security key, plus identity verification at chatgpt.com/cyber (Troubleshooting Daybreak). We covered the program’s tiers when GPT-5.6-Cyber launched behind Daybreak Red.

The Data controls page says:

“During Daybreak sessions on your personal account, we may use content from your interactions to improve our models even when ‘Improve the model for everyone’ is off. This helps us to strengthen safeguards against cyber misuse. Your ‘Improve the model for everyone’ setting still applies to your other activity when not in a Trusted Access session.”

What that means in practice:

• If you’ve never applied for Daybreak, nothing changes for you. Your switch from Step 1 still applies.
• If you’re approved on a personal account, treat anything you do with Daybreak turned on as potentially used for training, whatever your setting says. That includes details about systems you’re testing.
• Daybreak is something you turn on. In Codex CLI you type /daybreak to toggle it, and Codex replies “Daybreak on. Applies to new turns.” It’s also saved as the default for new sessions. In Codex Desktop it’s a Daybreak toggle in the Advanced model picker (Using Daybreak in Codex CLI). If you’re an approved user, check that it isn’t left on as your default when you’re doing unrelated work.
• Organization access is different. Business and Enterprise workspaces aren’t trained on by default. OpenAI’s Daybreak help also says Daybreak access doesn’t change an organization’s retention or Zero Data Retention settings; those stay as configured.

A practical rule for security people: if a client engagement forbids sharing data with a vendor for training, don’t run it through Daybreak on a personal account. Use an organization account where your company’s data terms apply.

Quick checklist

Run through this once on each account you use:

1. Settings > Data controls > Improve the model for everyone is off (or Privacy Portal “Do not train on my content” is done).
2. You don’t use thumbs up or down in chats you want kept out.
3. Include environments is off in Codex’s settings, if you use Codex.
4. Voice Include your audio recordings and Include your video recordings are off, if they appear.
5. You use Temporary chat for one-off sensitive questions, and don’t save it.
6. If you’re a Daybreak user on a personal account, Daybreak is off unless you need it.

Troubleshooting and FAQ

I turned the switch off. Are my old chats removed from training? OpenAI says that after you opt out, it won’t use your new conversations. The pages don’t promise anything about content shared before. If you want old chats gone, delete them; deleting a chat is separate from the training setting.

I can’t change the switch, or it’s greyed out. OpenAI says parent or guardian settings and workspace controls can stop you from changing it. On a work account, your admin controls it, and by default business workspaces aren’t trained on anyway.

Does turning off Memory stop training? No. Memory is about personalizing answers; the training switch is a separate setting. Turning off Memory also doesn’t disable the safety features that may use limited context in rare high-risk situations.

Do I need to change this on my phone and computer separately? Not when you’re signed in. The setting is tied to your account and applies on web and mobile. Only signed-out use is per browser.

Are my conversations with OpenAI Support used? They may be used to improve OpenAI services, including models, if “Improve the model for everyone” is on. Turning it off covers them too.

I connected Gmail or iCloud to ChatGPT. Is that data covered? The same switch applies. We walk through it in how to connect iCloud Mail and Calendar to ChatGPT: turn training off before you connect if you don’t want that content in the pool.

I use Codex with the new models. Anything different? No. Model choice doesn’t change training settings. If you’re setting up Codex, our guide to using GPT-6.1 Terra in ChatGPT and Codex covers the model picker; the switches in this guide apply whichever model you pick.

For more plain-language AI setup guides, see the 7minai homepage.