How to stop staff pasting client data into ChatGPT without banning AI
Banning ChatGPT rarely works. To stop staff pasting data into ChatGPT, stack five measures in order: a short written policy, a quick training session, a paid business plan, locked-down account settings, and an automatic check that takes sensitive items out before anything is sent. The first four set the rules. Only the fifth catches the paste that breaks them. A firm of 5 to 50 people can start all five this week.
Why a ban is not the answer
AI saves your team real time. Summarising a long email. Drafting a letter. Tidying a spreadsheet formula. Take it away and the work does not stop. It moves to personal phones and personal accounts, where you can see nothing.
Regulators notice this too. The ICO has flagged that ad hoc experimentation by staff with agentic AI can complicate Data Protection Officer oversight, as unstructured use may lead to unanticipated processing of personal information, according to Covington Inside Privacy. The aim is structure, not prohibition. AI is welcome. Client data stays out of it.
Five steps to stop staff pasting data into ChatGPT
Each step is cheap. Each one covers a gap the last one leaves.
1. Write a one-page AI policy
Keep it short enough to read in two minutes. It should say:
- Which AI tools are approved, and on which accounts.
- What must never go in: client names, bank details, National Insurance numbers, health details, passwords.
- What is fine: general questions, your own drafts with names taken out.
- Who to ask when unsure.
Then get each person to agree to it, with a date. A dated agreement is what clients and insurers ask to see.
2. Run a 15-minute training session
Show, do not lecture. Take a real task and do it both ways.
Worked example: Priya at Ashcombe wants a polite chaser for an overdue invoice. Unsafe: she pastes the whole email thread, with the client's name, address and sort code. Safe: she writes "Draft a polite chaser for Client A, invoice 4471, 30 days overdue." The answer is just as good. Nothing about the client went in.
Most careless pastes come from habit, not intent. Fifteen minutes of examples changes more habits than a long document.
3. Move everyone onto a paid business plan
Personal accounts are the biggest gap. On consumer ChatGPT accounts (Free, Plus, Pro), data is processed for training by default unless the user turns it off, while business accounts sit under contracts that prohibit OpenAI from training on your data, according to Secure Privacy.
The 2026 options, in US dollars:
| Plan | Price per user per month | Fits |
|---|---|---|
| Business (formerly Team) | $20 annual, $25 monthly. 2 to 200 seats. | Most small firms |
| Go | Roughly $35 to $40 | About 10 to 149 seats wanting more privacy protections |
| Enterprise | No list price. Reported quotes around $45 to $75. Roughly 150-seat minimum. | Larger firms |
Business and Enterprise figures are from Coworker.ai. Go is described by Inference.net as including most Enterprise-level privacy protections but fewer admin controls. For a fuller side-by-side, see which ChatGPT plan protects client data.
4. Lock down the account settings
Training opt-out lives in Settings, then Data Controls, under "Improve the model for everyone". On Business, Enterprise and Edu plans it is off by default. On personal accounts it must be switched off by hand, says the OpenAI Help Center.
Know what that switch does not do. Even Enterprise keeps data for 30 days for abuse monitoring, unless you separately set up Zero Data Retention. And no setting stops a person pasting a client list in the first place. Settings decide what happens after the paste. They cannot prevent it.
While you are there, check which other AI sites your team uses. Copilot, Gemini, Claude and DeepSeek all need the same thinking. Our guide on finding out which AI tools staff use covers how.
5. Add an automatic check before anything is sent
This is the step that catches mistakes. A browser check reads each paste, message or upload on AI sites. It takes out card numbers, bank details or names, or warns, or stops the send. Good tools swap the item for a label, so the question still goes.
Policy says what should happen. An automatic check makes it happen at 4.55 on a Friday.
Which tool category fits your size
- Under 5 people, low risk: policy, training and Business plan may be enough. A free redaction extension can help, though staff must remember to use it.
- 5 to 150 people handling client data: add a managed browser check across every PC. Accountants, solicitors, advisers and recruiters sit here.
- Larger firms on Microsoft E5: Microsoft Purview may already be in your licence, but it takes setup time.
- Firms with an IT provider: ask them. Many deploy browser checks across all their clients from one view.
Rule of thumb: if a single wrong paste would mean telling a client, you need step five, not just steps one to four.
A plan for this week
- Monday: write the one-page policy.
- Tuesday: send it round for dated agreement.
- Wednesday: run the 15-minute session.
- Thursday: move staff to a Business seat and check settings.
- Friday: trial an automatic check on a few PCs.
Where Helio Guard fits
Helio Guard is our tool for step five. An agent on each Windows PC adds an extension to Chrome and Edge. It checks pastes, messages and file uploads on listed AI sites. Sensitive items are taken out and replaced with labels. Client names can become a stand-in such as "Client A", then turn back when the answer is copied. Checks run on the PC with fixed rules, not an AI model. We never see what your team types. It also records each person's dated policy agreement. Windows only. No Mac. It costs £1 a person a month, and you can see the details on our pricing page.
Common questions about stopping staff pasting data into ChatGPT
Is turning off ChatGPT training enough to protect client data?
Turning off ChatGPT training is not enough on its own. The setting stops OpenAI using conversations to improve its models, but the client data has still left your firm. Business and Enterprise plans also keep data for a period for abuse monitoring. Training opt-out is worth doing, but it works after the paste, not before it.
Can staff still use ChatGPT on their phones?
Staff can still reach ChatGPT on a personal phone, and no PC-based check covers that. This is why the written policy and training matter. A clear rule that client data stays on work devices, with approved tools that are easy to use there, removes most of the reason to reach for a phone in the first place.
Do small firms really need an automatic check?
Small firms that handle client data usually benefit from an automatic check, because policy and training rely on people never slipping. One rushed paste of a client list can mean a difficult conversation. Firms with no client personal data, or very few staff, may decide a policy, training and a paid ChatGPT plan cover them well enough.
Related: How to stop employees sharing sensitive data with AI tools, not just ChatGPT, ChatGPT data protection for small businesses: a proportionate guide for 5 to 100 people, Is it safe to put client data into ChatGPT? What every firm should know