Does OpenAI train on the files you upload to ChatGPT?

Training: it depends which service

OpenAI draws a clear line between consumer and business offerings.

Consumer services. Content may be used to improve model performance, and OpenAI says explicitly that content may include files that are uploaded. There is a setting controlling this, in ChatGPT's data controls.

Business offerings. OpenAI states it does not use content submitted to business offerings such as the API and ChatGPT Enterprise to improve model performance.

Retention: files no longer follow their chat

This is the part that changed, and most advice about it is now wrong. Files used to live as long as the conversation holding them. Since Library storage arrived in May 2026 they do not. OpenAI's File storage and Library article states plainly that deleting a chat containing files does not delete those files saved to Library.

The exceptions OpenAI names are content already de-identified and disassociated from your account, and content that must be retained for security or legal reasons.

Note that OpenAI's older File Uploads FAQ still carries the previous chat-linked wording, which is why the outdated version of this advice is everywhere including OpenAI's own help centre.

Files uploaded as knowledge to a custom GPT are retained until you delete that GPT, which is worth remembering: a GPT you built and forgot is still holding whatever you gave it.

There is one documented way to upload without the file being kept at all: OpenAI states that files uploaded in a Temporary Chat are not saved to Library or to your account. It does not exempt you from the upload rate limits, so it is a retention measure rather than a way round the caps.

Deleting things, and why it is worth doing anyway

Beyond privacy, there is a practical reason. Your storage cap is 25 GB per user, shared across chats, Projects and custom GPT knowledge, and it is cumulative. It does not clear with time.

So deleting old files does two useful things at once: it removes the document and it frees the quota. Deleting the file is also the only thing that frees the quota, which is why "upload limit reached" so often appears to people who have not uploaded anything recently, and why clearing chat history does not shift it.

Reasonable precautions

Not paranoia, just proportionate care.

Redact before uploading. Names, account numbers, addresses, salaries. A contract analyses just as well with the parties called Company A and Company B, and you have removed the part that mattered.

Grep code for secrets. .env files, API keys in config, connection strings in test fixtures. Bundling a directory sweeps all of it in. Check before you send, not after.

Know what your employer allows. Many organisations have a policy on this, and a fair number of people find out what it was after uploading something.

Prefer excerpts. Sending the clause you have a question about rather than the whole agreement is better privacy and usually a better answer, because there is less irrelevant material competing for attention.

What this extension does not change

Worth being explicit. Using any tool to get content into ChatGPT does not alter what OpenAI does with it once it arrives. Your data controls setting and your plan govern that, and a browser extension has no influence over either.

What a locally-processing extension does mean is that there is no additional party. The content goes from your machine to OpenAI, and nowhere else along the way. That is worth checking on any tool in this category, since not all of them work that way.