What You Can Paste Into ChatGPT At Work: Two Checks
Two checks decide what you can safely paste into ChatGPT at work: which account you are signed in to, and what your company policy says. Credentials never go in either way.
What you can paste into ChatGPT depends almost entirely on which account you are signed in to. Credentials never go in, on any account. On a personal or free account, keep client information out of it. On a company business account, client work may be exactly what it is for, provided your policy says so.
August 2025, Claude. Anthropic reversed the default on its Free, Pro and Max plans: your chats now feed training unless you opt out, and those chats are kept for five years. If you picked Claude because it did not train on your chats, that reason has expired.
The direction of travel. Every major consumer AI tier now defaults to training. There is no longer a brand you can choose to avoid this. Your protection is the setting, or a business account.
Checked September 2026. We re-check this every quarter.
One Rule That Never Changes When You Paste Into ChatGPT
Credentials never go in. Not a password, not an MFA code, not an API key, not a recovery phrase. If it unlocks something, it does not belong in a chat window. That holds every time you paste into ChatGPT, on any tier, at any company.
That is the only universal rule. Everything else depends on two checks, and both take about ten seconds.
Check One: Which Account Are You Signed In To?
What you can paste into ChatGPT depends almost entirely on which account you are using. Consumer tiers and business tiers have opposite defaults. Same interface, same model, very different contract behind it.
| Tool | Personal or free tier | Business tier |
|---|---|---|
| ChatGPT | Free and Plus train on your chats by default. You can opt out in settings. | Business, Enterprise and API do not train on your content by default. |
| Claude | Free, Pro and Max: training on by default since August 2025, switchable off in settings. | Team and Enterprise: your content stays out of training by default. |
| Gemini | Free trains by default and conversations may be reviewed by a person. | Gemini for Workspace on a paid business account does not train on your content. |
The pattern holds across all three. Personal tiers train unless you stop them. Paid business tiers do not, because somebody signed an agreement saying they would not.
Tell which account you are on in five seconds. Open the account menu and look at the sign-in email address. If it is your work address, you are probably on the company account. If it is a personal address, you are on a personal account no matter which computer you are sitting at.
Check Two: What Does Your Policy Say?
A business license is necessary. It is not sufficient. Three things still sit on top of it.
- Your client agreement may say more than the license does. Some contracts restrict where client information may be processed at all.
- Professional obligations sit on top. Privilege, confidentiality and regulatory duties do not care what tier you bought.
- Your company policy decides, not you. If there is an approved tool list, that list is the answer.
So the honest version is this. Check your subscription first, then check your policy. When the two disagree, the policy wins. Your company AI policy is where the real boundary lives.
What Counts As Sensitive When You Are Moving Fast
Most of what people paste into ChatGPT is completely fine. The accidental disclosures are rarely dramatic. They happen because something looked harmless in the moment.
- A name plus context is identifying even when the name alone is not.
- Screenshots carry file names, folder paths, open tabs and whatever is in the sidebar.
- Financial and health details are sensitive whether or not a name is attached.
- Unreleased work is confidential until it is not.
Error messages are the most common accidental disclosure there is. People paste just the error, which is sensible, and the error contains a full file path, and the file path contains a client name and a matter number. Read the whole string before you paste it.
The Safe Rewrite Habit
You rarely need the real details to paste into ChatGPT and get a useful answer. Strip the identifiers and keep the structure of the problem. It takes a few seconds and it removes the question entirely.
Before: Draft an email to Mrs. Alvarez at Brightline Freight explaining that her claim under matter 2024-1187 was denied because the filing deadline passed.
After: Draft a short, professional email telling a client their claim was denied because a filing deadline was missed. Empathetic, factual, offers a next step.
The second version gets you the same draft. You fill the names back in yourself.
How To Stop ChatGPT Training On What You Paste
If you are on a free or personal account, this is worth two minutes right now.
On the web
- Open the account menu and select Settings
- Select Data controls
- Switch off Improve the model for everyone, then confirm with Done
On mobile
- Open the sidebar and tap your profile icon to open Settings
- Tap Data controls
- Turn off Improve the model for everyone
The switch is tied to the account, so the phone app, the desktop app and the browser all follow it, and none of your past chats are deleted. OpenAI describes this in its page on how your data is used to improve model performance.
Claude and Gemini each have their own switch, and Gemini's has a real cost attached. AI training on your data walks through each one.
Turning it off is not retroactive. Every one of these settings stops future conversations being used. Whatever went in before you flipped the switch stays in. A setting change cannot undo a paste that already happened.
If You Have Already Pasted Something You Should Not Have
If you paste into ChatGPT something you should not have, tell us the same day. Do not delete the chat and hope.
We are not going to be annoyed with you. Knowing what went where is what lets us judge whether anything actually needs to be reported, and that judgment gets much harder a week later. Our guide on how to report a security incident lists what to send us.
Common Questions
Can I paste into ChatGPT if my company pays for it?
Usually yes for company information, and the business tier does not train on it. Client information still depends on your client agreements and your policy. Ask before the first time you paste into ChatGPT for a client matter, not after.
Is it safer to paste a screenshot than text?
No. It is usually worse. A screenshot carries everything visible on screen, including file paths, folder names and other open documents.
Does deleting the conversation undo it?
Not reliably, and not for anything already used in training. Deleting is housekeeping, not a remedy.
What about pasting our own internal documents?
On an approved business account that is often exactly what the tool is for. On a personal account, treat internal documents as off limits.
Not Sure What You Can Paste Into ChatGPT? Ask First.
Describe the text to us, without sending the text itself, and we will tell you whether it is safe to paste into ChatGPT on your account. Asking takes two minutes, and nothing can be pulled back once it is pasted.