- On a business plan, ChatGPT, Claude, Copilot and Gemini are reasonably safe for everyday business work. None of them trains on your content by default.
- The real risk is personal accounts. Staff using free or personal plans for company work put client data where you can't see it, control it or delete it.
- Some data should never go into any AI tool without a specific agreement: passwords, account numbers, government IDs and patient health information.
This is the question we hear most from owners, usually right after they find out someone on their team has been pasting client emails into ChatGPT for months. The short answer: it depends far more on which plan your team uses than on which AI company makes it.
Below is what actually happens to your data on each plan, where the real risks are for a small or mid-sized business, and the six steps that make AI use safe. This is practical guidance from an AI consulting firm, not legal advice. If you're in a regulated industry, have your attorney or compliance officer review your setup.
What happens to your data on each plan
These are the defaults as published by each vendor as of November 2026. Settings and terms change, so check the vendor's current privacy page before relying on any of them.
| Plan | Used to train models? | Who controls it |
|---|---|---|
| ChatGPT Free, Go, Plus, Pro (personal) | Yes by default, unless the user turns off "Improve the model for everyone" in Data Controls | The individual user |
| ChatGPT Business and Enterprise | No, not by default | Your company's admin |
| Claude Free, Pro, Max (personal) | Only if the user allows it in privacy settings. If allowed, Anthropic says data can be kept up to five years | The individual user |
| Claude Team and Enterprise | No, not by default | Your company's admin |
| Microsoft 365 Copilot (work account) | No, not by default | Your Microsoft 365 admin |
| Gemini in Google Workspace | No, not by default | Your Workspace admin |
Sources: OpenAI, Anthropic, Microsoft and Google privacy and plan documentation, checked November 2026.
The pattern is simple. On business plans, the company controls the account and the vendor doesn't train on your content by default. On personal plans, each employee makes their own choice, and you never see it.
What "not used for training" doesn't mean
A business plan solves the training question. It doesn't make the tool a vault. Owners should know four things:
- Your data is still stored. Conversations and files sit on the vendor's servers for a retention period, which varies by plan and admin settings.
- It can be reviewed in limited cases. Vendors may review content to investigate abuse or security issues, as their terms describe.
- It can be subject to legal requests. Like email or any other cloud service, stored data can be requested in litigation or by authorities.
- It isn't privileged. Typing something into AI doesn't make it confidential in the legal sense.
In practice that puts a business AI plan in the same category as your email provider or cloud file storage. Fine for normal business work, with sensible limits on the most sensitive data.
The real risks for a small business
1. Personal accounts
This is the big one. A University of Melbourne and KPMG study of more than 48,000 people in 47 countries, published in 2025, found that almost half of employees admit using AI in ways that break company policy, including uploading sensitive company information into free public tools. 57% said they hide their use of AI at work. When that happens on a personal account, you have no record of what was shared and no way to remove it, and it leaves with the employee when they do.
2. Wrong answers sent with confidence
AI tools sometimes state incorrect facts, figures or policy details as if they were certain. The data risk gets the headlines, but a wrong number in a client proposal or a made-up clause in a contract summary does more damage in most small businesses.
3. Tools that can see too much
Copilot and Gemini can draw on the files a user has access to. In many businesses, years of loose sharing mean staff technically have access to payroll, HR or client folders they never open. AI will find them. Fix permissions before you switch these features on.
4. Connected apps
Connecting AI to email, your CRM or cloud storage is where much of the value is. It's also where access widens. Connect only what a task needs, and know who approved each connection.
5. People who leave
On a business plan, removing someone's access takes a minute. On personal accounts, it isn't possible at all.
6. Look-alike apps and extensions
App stores and browser extension stores are full of tools with names like "ChatGPT Assistant" that aren't made by OpenAI. Some request broad permissions. Approve specific tools by name and source.
What never to paste into an AI tool
- Passwords, login codes and security keys
- Social Security, passport and driver's license numbers
- Bank account and card numbers
- Patient health information, unless a signed BAA covers the tool
- Anything under an NDA that forbids it
- Drafts of emails, proposals and marketing
- Internal procedures, SOPs and training material
- Meeting notes and call summaries
- Spreadsheets with identifying details removed
- Public information about clients and markets
A simple habit covers most of the gray area: replace names and identifiers before pasting ("Client A", "the property on Main Street"). The AI rarely needs them to do the job.
Six steps to make AI use safe
Find out what's in use today
Ask the team, without blame, which AI tools they use, for what and on which account. The answers are usually more than you expect.
Pick one business plan
Choose the assistant that fits how your team works and buy business seats. Our comparison of ChatGPT, Claude, Copilot and Gemini walks through the choice.
Write a one-page policy
Name the approved tool, list what must never be pasted in and say who reviews output before it reaches clients. Start from our free AI acceptable use policy template.
Check the admin settings
Turn on single sign-on if you have it, set data retention to what you need, and limit which apps can be connected.
Clean up file permissions
Before switching on AI that reads your files, review who can see what in SharePoint, OneDrive or Google Drive.
Train the team on real examples
Fifteen minutes with your own emails and documents ("this one's fine, this one needs the account number removed") does more than any written rule. More in AI training for employees.
If you're in a regulated industry
- Healthcare, dental and med spas: protected health information needs a signed Business Associate Agreement under HIPAA that covers the specific product. As of November 2026, ChatGPT Business and Claude Team don't offer one. Microsoft's BAA can cover Copilot on work accounts and Google's can cover Gemini in Workspace once accepted. Confirm in writing.
- Financial advisors and insurance: client communications can be subject to recordkeeping and supervision rules, including AI-drafted messages. Check approved tools with your compliance team or broker-dealer.
- Law firms: state bar guidance applies. In Florida, Ethics Opinion 24-1 (2024) allows generative AI but requires protecting confidentiality and supervising output.
- Accounting and tax: client tax information carries its own federal confidentiality rules. Remove identifying details or use a tool your firm has vetted for it.
How we help
Every AI Opportunity Audit we run includes a readiness and data privacy review: which tools are in use, what's being shared, and what to change. We also write AI policies with clients and train teams on them in our in-person workshops.
Common questions
Does ChatGPT use my business data to train its models?
On ChatGPT Business and Enterprise, OpenAI says it doesn't train on your content by default. On personal Free, Go, Plus and Pro accounts, the setting that allows training is on unless the user turns it off. That's why company work belongs on a business plan, not personal accounts.
Is ChatGPT confidential?
Not in the way a conversation with your attorney is. Even on business plans, your content is stored by the vendor, kept for a period of time and can be subject to legal requests. Treat it like any other cloud software: fine for normal business work, not a place for passwords, account numbers or regulated data unless your agreement covers it.
Is Claude safer than ChatGPT for business data?
On business plans the two are broadly comparable: neither trains on your content by default, both offer admin controls and single sign-on. The bigger difference is between business plans and personal accounts, not between vendors.
Can I put patient information into ChatGPT?
Only if your practice has a signed Business Associate Agreement with the vendor that covers the product you're using. As of November 2026, OpenAI doesn't offer one for ChatGPT Business, so standard business plans aren't suitable for protected health information.
What's the first step to make AI use safe at my company?
Move everyone off personal accounts onto one approved business plan, and give them a one-page policy listing what must never be pasted in. Those two steps remove most of the risk.
This article is general information, not legal, tax or compliance advice. Prices, plans and vendor terms change; confirm current details with each vendor before you rely on them. ChatGPT, Claude, Microsoft 365 Copilot, Gemini and other product names are trademarks of their owners. FlowBack AI is not affiliated with or endorsed by these companies except as stated.
Carter Simmons is the founder of FlowBack AI, an AI consulting firm in Tampa. He holds 13 AI certifications from Anthropic (the maker of Claude), Google and Microsoft, and FlowBack AI is an OpenAI Select Partner. The firm runs AI Opportunity Audits, advisory programs, implementation and team workshops for established businesses in Tampa Bay and nationwide.
Newsletter
The Owner’s AI Brief
Twice a month for owners of established businesses: one AI tool you can use in 15 minutes, with the dollar math behind it.
Subscribe to the newsletterNot sure what your team is putting into AI?
Every AI Opportunity Audit includes a readiness and data privacy review. Book a free 15-minute call and we'll tell you whether it makes sense for your business.
Speak with a consultant →