Yes. On a personal ChatGPT Free, Plus or Pro account, your new chats can be used to train OpenAI's models unless you turn off a setting called Improve the model for everyone. It lives in Settings, under Data controls.
That switch is only one of four things I check before client work goes anywhere near an AI tool. The other three are Temporary Chat, memory and shared links, and they trip people up more often than training does.
I checked every setting name and path below against OpenAI's and Anthropic's help pages on 23/10/2026. These menus move around, so if yours looks different, the linked page is the source of truth.
How do I stop ChatGPT training on what I type?
Open your account menu, select Settings, then Data controls. Select Improve the model for everyone, turn it off and select Done. On iOS and Android it's the sidebar, then your profile icon, then Data controls.
Per OpenAI's Data controls page, sharing is on by default for personal Free, Plus and Pro accounts. A few details from the same page that people miss:
- It covers new conversations from that point. It doesn't delete or hide your saved chats.
- Voice mode has its own separate control for training.
- If you give a thumbs up or thumbs down on a reply, OpenAI says the entire conversation attached to that feedback may be used for training, even with the setting off.
- Business, Enterprise, Edu and Healthcare workspaces aren't used for training by default.
Turn it off whatever plan you're on. It takes thirty seconds, and it's the first of the ChatGPT settings I'd change on any new account.
Is Temporary Chat safe for client work?
Safer than a normal chat, but it isn't a vault. OpenAI's Temporary Chat page says a temporary chat stays out of your history, doesn't create or update memories and isn't used to improve models while it stays temporary. It also says OpenAI may keep a copy for up to 30 days for safety.
Two gotchas. If you save a temporary chat, it turns into a regular chat and follows your normal settings. And you choose Personalized or Unpersonalized before you send the first message. You can't change it afterwards. If it touches a client, choose Unpersonalized.
What does ChatGPT's memory hold on to?
Whatever it decides is useful, and that can include something you typed once and forgot. Memory and training are two separate settings. Turning off one doesn't touch the other.
Find memory under Settings, then Personalization, then the Memory section. OpenAI's memory page warns that deleting a chat doesn't necessarily delete a separate saved memory created from it. So clearing a chat isn't enough. Open the memory list and check.
Are shared ChatGPT links really public?
For personal accounts, yes. OpenAI's sharing page says anyone who has the link can view the conversation, there's no expiry date, and you can't limit it to named people. It also tells you not to put health information, financial details, passwords or account numbers in shared content.
My rule is the billboard rule. If you wouldn't put it on a billboard, don't put it in a chat you plan to share. Links get forwarded to people you've never met.
This isn't only a ChatGPT problem. Researchers found around 6,800 shared Meta Muse chats in public DNS records in October 2026. They opened a few hundred and found a small-business loan application, a bank fraud alert with partial account details and a staff holiday calendar. Meta paused Muse sharing on 20/10/2026.
To check yours: Settings, Data controls, then Manage next to Shared links. Delete one, or open More actions and select Delete all shared links. Deleting a link stops new access. It doesn't remove a copy someone already saved.
And in Claude?
Same idea with different labels. Anthropic's page on model improvement says to go to Settings, then Privacy, and look for the toggle called Help Improve our AI models. With it off, new chats and coding sessions aren't used for future model training. Conversations flagged by safety systems may still be used for safety purposes, and anything already used in training isn't removed.
For sharing, Settings, then Privacy, then Shared chats lets you unshare chats on the Free, Pro and Max plans. Anthropic's public links page says turning a link off doesn't delete copies others already made. Team and Enterprise chats can only be shared inside your organisation.
What would I never paste into AI?
Settings only help with what's left after you've decided what goes in. These stay out of my chats, whatever the toggle says:
- Money details. Bank statements, BSB and account numbers, your tax file number.
- Health details. Medicare numbers, diagnoses, a staff member's medical certificate.
- Other people's details. Client names with contact details, signature blocks, screenshots of DMs.
- Passwords, logins and API keys. Use a placeholder like [API KEY] instead.
- Confidential business material. Anything under NDA, unannounced pricing, staff performance issues.
For client work, swap names for roles (Client A, my VA) and describe the document instead of pasting it: "a contract with vague scope and 30-day payment terms" gets you the same help.
This matters even more once AI can reach your accounts. When you connect Claude to Gmail and your calendar, everything in your inbox is within reach of the chat, so the same rules apply to what's already there.
Your 15-minute check:
- Turn off Improve the model for everyone (ChatGPT) and Help Improve our AI models (Claude).
- Open Shared links in ChatGPT and Shared chats in Claude. Delete what you don't need.
- Open ChatGPT's memory list and delete anything with a client's name in it.
- Start your next client task in an Unpersonalized Temporary Chat.
Quick answers
Does ChatGPT Business train on my chats?
Not by default. OpenAI says it doesn't use content from Business, Enterprise, Edu or Healthcare workspaces to train its models. Your workspace admin can still set its own retention rules.
Does deleting a chat delete its share link?
Yes. OpenAI says deleting a conversation removes its shared conversation link.
Does a shared link include my memories?
No. OpenAI's memory page says Memory Sources aren't included when you share a conversation through a link. The words you typed in that chat still are.
Want help setting this up properly?
I run the Wright Mode membership, so I have an obvious interest here. Settings are a ten-minute job. Working out which client tasks can use AI, and which need a scrubbed version first, is the harder part, and it's easier with someone to ask. Inside Wright Mode you get the monthly live Build with Brooke, fortnightly Co-Work and Q&A, and a private community where you can bring a real client scenario and get an answer.
