Do OpenAI Dots Train on My Data? Personalization Isn’t the Same as Training

On a personal ChatGPT plan, dots’ conversations and work may be used to train OpenAI models unless you opt out. In ChatGPT Business, Enterprise, and Edu workspaces, OpenAI says it does not train on that content by default. The deciding factor is your workspace and its data settings—not simply whether your dot “learns” about you. These are OpenAI’s published rules as of October 25, 2026. (Dots privacy FAQ)

The useful distinction: you can let a dot build context about your preferences while declining to contribute eligible work to OpenAI’s model training. That does not stop the dot from processing or retaining information needed to work for you.

Two different meanings of “learning”

OpenAI’s dots launch page says dots learn from feedback and get to know your preferences and standards. That describes the product’s personalization promise; it does not, by itself, establish that your interactions update the underlying model.

OpenAI documents two separate mechanisms:

• Personalization and context: Your dot can retain information from conversations and connected apps, and exchange memories and recent conversation context with ChatGPT. (Dots privacy FAQ)
• Model improvement: Eligible conversations and work may be used to train OpenAI models, depending on your plan and settings.

OpenAI explicitly treats memory and model-training controls as separate settings. Turning off training is therefore not the same as turning off personalization—or deleting what your dot already knows.

What can enter training?

For personal plans, the dots-specific policy says “Improve the model for everyone” controls whether dots’ conversations and work may be used for model improvement. The scope is broader than messages you type: it can include actions dots take, delegated work, automations, and connected-app data used to inform conversations. OpenAI says it takes steps to remove personal identifiers where possible before using eligible data. That is not a promise that no sensitive content can enter training. (Dots privacy FAQ)

There is an important qualification around background activity. OpenAI says it does not train directly on proactive research or the dot’s private research notes. However, information from that research can become eligible if it is brought into a conversation or task, depending on your settings. (Dots launch page)

For example, a detail found in a connected project document during background research is not necessarily excluded forever: if the dot uses it in an eligible task, the indirect-use provision can apply. “We don’t train directly on research notes” is narrower than “nothing read in the background can ever be used for training.”

How to opt out on a personal plan

OpenAI says data sharing is enabled by default in personal Free, Plus, and Pro workspaces. To turn it off on the web:

1. Open your ChatGPT account menu.
2. Select Settings → Data controls.
3. Turn off Improve the model for everyone and select Done.

When signed in, this setting applies across web and mobile. OpenAI’s general policy says that after opting out, new conversations will not be used to train its models; it does not promise to reverse training that already happened. (Data controls; model-improvement policy)

One exception matters: OpenAI says that if you submit response feedback, such as thumbs up or thumbs down, the entire conversation associated with that feedback may be used for training even after you opt out. If that conversation contains sensitive material, avoid submitting it through those feedback controls. (Data controls)

If your work also involves Codex, check its separate Include environments setting. The ChatGPT opt-out does not change that control for additional context from Codex environments. (Data controls)

For work data, use your organization’s approved workspace rather than assuming a paid personal Pro account has business protections. OpenAI’s business-data policy says business inputs and outputs are not used for training by default. That protection concerns training; your organization’s retention and access policies still apply.

No training does not mean no stored data

The dots privacy documentation sets out several limits worth knowing before connecting accounts:

Control	What it changes—and what it leaves behind
Turn off model improvement	Controls training eligibility; it does not delete saved chats or your dot’s context.
Turn off ChatGPT Memory	Stops memory sharing with your dot; it does not delete information the dot already received.
Disconnect an app	Stops new access through that connection; existing information in the dot’s context remains.
Delete your dot	Deletes its own context; separately stored files, conversations, Codex threads, and ChatGPT memories are not automatically deleted.

OpenAI also says you currently cannot inspect, delete, or directly edit individual dot memories. Limited human review can occur, including for safety reasons, even with model improvement turned off. These are vendor-documented boundaries, not an independent verification of how the service operates. (Dots privacy FAQ)

If your concern is contributing data to model training, turn off model improvement before sharing sensitive material. If your concern is access or retention, also limit connected apps and consider deleting your dot’s context, while separately managing files, chats, and ChatGPT memories. The training toggle is useful, but it is not a general “keep everything private” switch.