Not by default. ChatGPT stores your conversations against your account, and on consumer plans they can be used to improve the model unless you turn that off in Data Controls. Business, Enterprise, and API traffic are excluded from training by default. Staff can access conversations for abuse review.
Short answer
Not by default. ChatGPT stores your conversations against your account, and on consumer plans they can be used to improve the model unless you turn that off in Data Controls. Business, Enterprise, and API traffic are excluded from training by default. Staff can access conversations for abuse review.
"Private" gets used to mean three different things in this conversation, and OpenAI's answer is different for each: is it stored, is it read by humans, and is it used for training.
Is it stored?
Yes. Every conversation is saved against your account so it can appear in your sidebar. That is the product working as intended. The exceptions are Temporary Chat, which is excluded from history and memory, and deleted conversations, which OpenAI says are purged within 30 days.
Can humans read it?
In limited circumstances, yes. OpenAI's policy allows authorised staff to review conversations for abuse investigations, safety enforcement, and legal obligations. Automated classifiers scan traffic continuously; human review is triggered by those flags rather than applied at random.
This is normal for a service of this kind, and it is also the reason "nobody will ever see this" is not a safe assumption for anything genuinely confidential.
Is it used for training?
This is the one that actually varies:
| Plan | Trained on by default |
|---|---|
| Free | Yes, unless you opt out |
| Plus / Pro | Yes, unless you opt out |
| Team | No |
| Enterprise / Edu | No |
| API / Platform | No |
On consumer plans, open Settings, then Data Controls, and turn off the model improvement toggle. It applies going forward only, so anything already collected is already collected.
What about legal holds?
Retention policies bend under court orders. During litigation an AI provider can be required to preserve data it would otherwise have deleted, including conversations users have removed. This is not specific to OpenAI; it applies to every cloud service you use. It is worth knowing because it means "deleted in 30 days" is a default, not a guarantee.
What a privacy conscious setup looks like
Turn off model training in Data Controls.
Use Temporary Chat for anything sensitive, so it never enters history or memory.
Do not paste credentials, client data, medical records, or anything covered by an NDA. No setting makes that safe.
Review Settings, Personalization, Manage memories periodically and delete what you do not want persisted.
Keep your own archive of the conversations you value, then delete aggressively on the platform side.
That last point is the part people skip. Privacy is easier when your working copy lives with you rather than in an account you do not control. Export your data from Data Controls, convert conversations.json into readable PDF or Markdown files, and store them locally. ChatExports does that conversion in your browser, so the transcripts never touch another server on the way.
Compared with the other assistants
Claude does not train on consumer conversations by default, which makes Anthropic the strongest of the mainstream defaults. Gemini's activity is tied to your Google account and human reviewers can see selected conversations, so Google recommends against entering confidential information. Copilot's behaviour depends heavily on whether you are signed in with a personal or a work account. In every case the shape of the answer is the same: stored yes, reviewable sometimes, trained on depends on the plan.
Frequently asked questions
Does ChatGPT read my conversations?
Automated systems scan all traffic for policy violations. Human reviewers can access conversations flagged by those systems or required for legal reasons.
Is Temporary Chat actually private?
It is excluded from your history, from memory, and from training. OpenAI still retains a copy for a short window for safety review before deleting it.
Can my employer see my ChatGPT conversations?
On a personal account, no. On a Team or Enterprise workspace, administrators have access to workspace data and compliance exports.
Is it safe to paste work documents into ChatGPT?
Only if your organisation permits it and you are on a plan excluded from training. For anything under an NDA, assume the answer is no.