ChatGPT is safe for business use when two things are true. Staff use a business workspace, where OpenAI does not train on your data by default. And five settings are checked: model training, feedback buttons, temporary chats, shared links and memory. Free and personal accounts can be used for training unless someone switches it off, so the risk sits in the account and the settings.
What actually happens to the data you type into ChatGPT?
What happens to data typed into ChatGPT depends on the account it was typed into. On a personal Free, Go, Plus or Pro account, conversations can help train OpenAI’s models unless the user turns that off. On a business workspace, training is off by default.
OpenAI’s enterprise privacy page states it does not train on data from ChatGPT Business, Enterprise or the API by default.
Our guide to keeping company data out of AI model training covers the account difference. This article spends its time on the settings.
Two other facts matter for an owner. On ChatGPT Business, OpenAI says deleted or unsaved conversations are removed from its systems within 30 days, unless the law requires longer. And workspace admins can view, export and delete the conversations staff hold in that workspace.
That second point surprises people. A business workspace gives the firm control of the record, which is what you want when someone leaves.
Which ChatGPT settings make it safe for business use?
Five ChatGPT settings decide most of the risk, and an owner can check all five in about 20 minutes. Each one lives in the account menu under Settings, then Data controls, or in the chat window itself.
| Setting | Where to find it | What it controls | What a small firm should do |
|---|---|---|---|
| Improve the model for everyone | Settings, Data controls | Whether new chats on a personal account can train OpenAI’s models | Turn it off on any personal account used for work, or move the person to the business workspace |
| Feedback buttons (thumbs up or down) | Under each ChatGPT reply | Whether that whole conversation can be used for training, even after opting out | Tell staff not to rate replies in chats that hold client or staff details |
| Temporary chat | Chat window, before the chat starts | A chat kept out of history and memory, held for up to 30 days for safety | Use it for one-off questions that touch sensitive material |
| Shared links | Settings, Data controls, Shared links, Manage | Who can open a link to a conversation | Delete old links from personal accounts; keep sharing inside the workspace |
| Memory | Settings, Personalisation | What ChatGPT remembers about the user between chats | Keep client names and personal details out of saved memories |
The table follows OpenAI’s data controls help page, as checked in October 2026. The sections below cover the two that catch most firms out.
Does the thumbs-up button undo your privacy settings?
The ChatGPT feedback button can override a training opt-out for the conversation it is attached to. If you rate a reply after opting out, OpenAI’s data controls page says “the entire conversation associated with that feedback may be used to train OpenAI models”.
The top-ranking guides we read for this article skip it. A member of staff can switch training off correctly, then click thumbs up on a helpful reply about a client dispute. That one click puts the whole chat back in scope.
The fix costs nothing. Add one line to your staff guidance: no rating replies in any chat that holds client, staff or financial details. On a business workspace the training default is already off, but the habit is worth building anyway.
Who can see a ChatGPT shared link?
A ChatGPT shared link from a personal account can be opened by anyone who has the link. OpenAI’s sharing help page says so plainly, and it means a forwarded link can travel well beyond the person it was meant for.
Links made inside a ChatGPT Business or Enterprise workspace behave differently. Only members of that same workspace can open them, so a link pasted into the wrong email shows nothing to an outsider.
One gap remains.
The same page notes that ChatGPT Business does not give admins a workspace-wide switch to turn sharing off, which Enterprise does. For a small firm that means a short rule in the policy, plus a quarterly look at Settings, Data controls, Shared links.
Is ChatGPT safe for confidential information?
ChatGPT is safe for confidential information only on a business workspace. Even there, use it only for material your contracts and regulator let you process with a third-party tool. Personal accounts should never hold client files, staff records or anything covered by a confidentiality clause.
A useful test is to ask who you would have to tell if the chat leaked. If the answer is a client, a regulator or an employee, the material belongs in the business workspace or nowhere.
Our AI acceptable use policy guide turns that test into a one-page rule.
Some material stays out of every chat tool. OpenAI’s own sharing guidance tells users not to include health information, financial details, passwords or account numbers in shared content. Treat that list as the minimum for staff guidance.
How much does it cost to make ChatGPT safe for a 12-person firm?
Moving a 12-person firm onto ChatGPT Business costs £2,160 a year on annual billing at current UK prices. As of October 2026, OpenAI’s pricing page lists a standard Business seat at £15 per user per month billed annually. Monthly billing is £18, from two users.
Here is the worked example for 12 staff.
- Annual billing. 12 users at £15 is £180 a month, or £2,160 a year.
- Monthly billing. 12 users at £18 is £216 a month, or £2,592 a year.
- Partial rollout. If only the 5 people who handle client material need it, annual billing is £75 a month, or £900 a year.
Option 3 is often where a small firm should start. It puts the sensitive work inside the workspace and leaves everyone else on a clear rule: no client material in personal accounts.
Bedrock AI’s view is that the subscription is the cheap part. The expensive part is finding out which jobs are worth putting into ChatGPT at all. That takes a diagnosis of the work before any money is spent on seats.
How do you check whether your staff are using ChatGPT safely today?
The quickest check is to ask every member of staff which AI tools they used last week, and on which account. Most owners find at least one personal login being used for work.
Our guide to finding the AI tools your staff use sets out a no-blame way to ask.
Then run the 20-minute settings check with each person who uses ChatGPT for work.
- Open Settings, then Data controls, and confirm Improve the model for everyone is off on any personal account.
- Open Shared links, select Manage, and delete any link that holds client or staff details.
- Open Personalisation and read what Memory has saved, removing anything about named clients.
- Agree the two habits: no rating replies on sensitive chats, and temporary chat for one-off sensitive questions.
- Write down which account each person will use for work from now on.
Step 5 is the one that lasts. Settings drift as OpenAI adds features, but a written list of who uses which account gives you something to check against next quarter.
FAQ
Is ChatGPT safe for business use on the free plan? The free ChatGPT plan is not a safe home for business data. Conversations on personal accounts can be used to train OpenAI’s models unless the user turns that setting off. The firm also has no admin view of what was shared.
Does ChatGPT Business train on my company’s data? ChatGPT Business does not train on your company’s data by default, according to OpenAI’s enterprise privacy commitments. The same default applies to Enterprise and the API. Feedback ratings and settings should still be covered in staff guidance.
Can my employer see my ChatGPT chats? On a ChatGPT Business workspace, admins can view, access, export and delete conversations held in that workspace, according to OpenAI. A personal account is outside the employer’s view, which is why personal accounts should not hold work material.
Is ChatGPT safe to use at work for drafting emails? ChatGPT is generally safe for drafting routine emails at work if the draft holds no confidential client or staff details. For anything sensitive, use the business workspace, and have a person read every draft before it is sent.
Can ChatGPT give wrong answers at work? ChatGPT can give wrong answers at work, including confident ones with made-up figures or sources. A business account protects the data and does nothing for accuracy. Have a named person check every output that reaches a client or shapes a decision.
Get ahead with AI by getting The Business Brain | Would rather talk it through? Book a free 15-minute call