There's a fair chance someone on your team pasted a customer email into an AI chat tool this week and asked it to make the reply sound friendlier. Maybe on a work laptop, maybe on their own phone at lunch, maybe without mentioning it to you.
That isn't a scandal. It's where work is now. In Microsoft's 2024 Work Trend Index, 78% of AI users said they were bringing their own AI tools to work, and 80% at small and medium-sized companies. Those are shares of people who already use AI at work, from a survey of 31,000 full-time or self-employed knowledge workers (people whose jobs are mostly desk and computer work) in 31 markets, taken between February 15 and March 28, 2024. It isn't a measurement of your team. It is a fair hint that "we don't really use AI here" may be more hope than fact.
So the useful question is which of three stances you take: ban it, look the other way, or give people work accounts under a short written policy. Below you'll find the three side by side, what Google and Microsoft say in their own terms about personal versus work accounts, when each option honestly wins, and four decisions to write down whichever way you go.
One boundary up front: this isn't legal advice. Patient records, privileged legal files and anything a contract restricts belong with your own lawyer and your vendors' written agreements.
Ban, Ignore or Equip: The Three Stances Side by Side
"Work accounts" here means the business versions of these tools that your company signs up for and manages, rather than each person's personal login. The provider details come from Google's and Microsoft's own pages as of October 2026, quoted in the next section.
| Ban it | Look the other way | Work accounts plus a short written policy | |
|---|---|---|---|
| What data leaves, and under whose terms | On paper, none. Any use that carries on runs on personal accounts, under consumer terms each employee accepted. | Whatever each person pastes, under their own consumer terms, which differ from tool to tool. | Whatever staff paste, under the business agreement your company accepts. Microsoft says it acts as a "data processor" for organizations. |
| Who can see or review it | For anything that slips through, the provider. Google and Microsoft both say some personal-account chats are reviewed by people. | The same, except as the everyday arrangement rather than the exception. | Google says Workspace content isn't human reviewed outside your domain without permission. Microsoft says it won't use your data except as you instruct. |
| What you, the owner, can see | Nothing. A ban gives people a reason not to tell you. | Nothing, unless people volunteer it. | Depends on the plan. Google offers admins audit logs; Microsoft logs prompts and responses for auditing. |
| What happens when someone leaves | Any chat history stays in their personal account, out of your reach. | The same: it leaves with them. | The business owns the account. In Google Workspace, you can move what they own to a colleague before removing it. |
| How likely the rule is to be followed | Holds where you control the devices. Weak where everyone has the app on their own phone. | There's no rule, so each person draws their own line. | Better when the approved tool is as easy to reach as the personal one and the rules fit on a page. Nothing stops a personal phone. |
Only the third column gives you any say over the answers. That doesn't make it the automatic pick. It just shows the trade.
What Google and Microsoft Say About Personal vs. Work Accounts (as of October 2026)
Everything here is quoted from the providers' own pages as they read in October 2026. Terms like these change, so check the date on each page before you rely on it.
Google Gemini: personal account vs. Google Workspace
Google's Gemini Apps Privacy Hub for personal accounts, last updated September 24, 2026, says: "A subset of chats are reviewed by human reviewers (including Google's trained service providers) to help improve Google services." It asks: "Please don't enter confidential information that you wouldn't want a reviewer to see or Google to use to improve our services, including machine-learning technologies." Deleting your history doesn't erase everything, either: reviewed chats "are not deleted when you delete your activity. Instead, they are retained for up to three years." The same page notes: "If you have a work or school Google Account, your use of Gemini Apps may be subject to different data handling terms."
Those terms sit in Google's Generative AI in Google Workspace Privacy Hub, which applies to "a qualifying edition of Google Workspace." There, Google says: "Your content is not human reviewed or otherwise used for Generative AI model training outside your domain without permission." Admins can switch the Gemini app off, control conversation history for the organization, and use "a robust set of audit logs to track user activity and interactions with Gemini."
Microsoft Copilot: personal account vs. work account
Microsoft's Privacy FAQ for Microsoft Copilot, for the consumer version, says: "Some Copilot conversations are subject to both automated and human review for product improvement and digital safety purposes." And: "By default, we store conversation activity for 18 months." Whether those chats are used for model training is a setting the account holder controls, not their employer. The FAQ also says "We do not train Copilot on data from the following types of users" and lists "Users signed into Copilot with an organizational Entra ID account." Entra ID is Microsoft's name for the work sign-in a business sets up and manages.
For those work sign-ins, Microsoft's enterprise data protection page says it handles the data "as a data processor," which in everyday words means on the business's behalf: "We won't use your data except as you instruct." Prompts and responses "aren't used to train foundation models," and Copilot "applies your retention policies, supports audit of interactions, and follows your administrative settings." The catch: "The specific controls and policies will vary depending on the underlying subscription plan."
Microsoft's Copilot Chat privacy page says "both the prompt and response are logged and stored" for auditing in the business's Microsoft 365 account. It also gives staff an easy check: enterprise data protection shows as "a green shield along the top of the user interface next to the New Chat button."
ChatGPT: read OpenAI's current terms yourself
OpenAI publishes its own data terms for ChatGPT's personal and business plans. Those pages wouldn't load while I was writing this, and a summary from memory is exactly the kind of thing that goes stale, so read OpenAI's current terms directly. Look for the same answers as above: whether your data is used for training, whether people at the provider can review it, what an administrator can see, and what happens to someone's history when you remove them.
What no account type changes
Work-account terms cover where your data goes, not whether the answer is right. A confident, wrong paragraph in a customer email is just as wrong on a business plan. Nor do they settle regulated information on their own. Microsoft's enterprise data protection page says Copilot's support for the federal health privacy law (the Health Insurance Portability and Accountability Act) is "for properly configured implementations," and that it "doesn't apply to web search queries." Setup and contracts are questions for your lawyer.
When Each Option Honestly Wins
A ban wins when…
- The data can't go to any outside service yet. Patient records, privileged client files or anything that needs a signed vendor agreement your lawyer hasn't reviewed. "No AI tools with this data" is the honest default.
- A customer contract says so. Some client agreements limit where their information may go. Your AI rule has to match what you already signed.
- You control the devices. On shared front-desk computers or locked-down laptops, a ban is something you can actually enforce, not just a line on paper.
- It's a pause with an end date. "No AI tools with customer information until we decide, by the end of the month" buys time without pretending the question is closed.
The weakness is the part you can't see. In the same Work Trend Index, 52% of people who use AI at work said they were reluctant to admit using it for their most important tasks. That's a survey of AI users, not your staff, but a ban on tools people already like gives them one more reason to keep quiet.
Looking the other way wins when…
- It's a tiny team, and you've actually asked. If there are three of you and the honest answer to "what do you use it for?" is public material (a social post, ideas for product names, a spreadsheet formula with made-up numbers), a written policy may be more ceremony than protection.
- Nobody uses it. Ask first. There's no sense writing rules for a habit nobody has.
The trap is doing nothing by default rather than on purpose. "We never talked about it" isn't one stance. It's a separate, unwritten stance for every person on the payroll.
Work accounts with a short policy win when…
- Customer details are already going in. Names, quotes, complaint emails. Moving that work under a business agreement changes whose terms apply.
- People come and go. You want history and files to stay with the business rather than walk out the door.
- You already run on Google Workspace or Microsoft 365. Check what your current plan includes before you buy anything new.
The weakness: accounts don't change habits by themselves. If the work tool is slower to reach than the app on someone's phone, the phone wins. Someone also has to manage the accounts, and that job often lands on you.
So Which One Should You Pick?
Start with a question, not a policy. Ask your team, with no penalty for honest answers, which AI tools they use and what they put into them. Then:
- If nothing sensitive goes in and the team is small, a short written rule may be all you need. No purchase required.
- If customer or financial information goes in regularly, either move that work onto work accounts under a short policy, or keep that information out of every AI tool. Both are defensible. Customer details on personal accounts with no rule at all is the combination that's hard to defend.
- If regulated or contract-restricted information is involved, keep it out of AI tools until your lawyer has reviewed a vendor agreement.
None of those answers requires hiring anyone, us included. The written decisions below matter in every case.
Four Decisions to Put on One Page, Whichever Way You Go
Write them as the questions a new hire would ask on day one.
- "Which AI tools can I use for work, and signed in as whom?" Name the tools. With work accounts, say that work goes through the work login only (in Microsoft Copilot, look for the green shield). With a ban, say what it covers: every AI tool, or only certain information.
- "What never goes into a prompt?" Use examples from your own business rather than categories: a customer's card number, a patient's name next to the reason for their visit, the alarm code, anything you've promised in writing to keep confidential. People follow examples better than definitions. We looked at the same question of where customers' words end up, for phone agents, in whether voice AI is safe for a business phone.
- "Who reads it before a customer does?" Any AI-drafted email, quote or review reply gets read first by a named person, who owns it as if they'd written it. Drafting and sending are different jobs, the same argument we make in why a drafting agent shouldn't hold the publish key.
- "Whose account is it, and what happens when I leave?" With work accounts, the business owns them. Name who manages them, and on someone's last day, move what they own to a colleague before removing the account. Google's admin help explains why the order matters: "Data owned solely by the user—that isn't transferred before deleting their account—is permanently deleted." With personal accounts, whatever was pasted leaves with that person, which makes the second question matter even more.
Then add a review date. The terms keep moving: Google's Gemini privacy page for personal accounts was last updated September 24, 2026, shortly before this was written.
Frequently Asked Questions
Can I just tell staff to keep customer information out of ChatGPT?
Yes, as a start, but it only works if people know what counts. Write a short list of examples from your own business, such as account numbers, home addresses and anything a customer told you in confidence, so the line is clear.
Does a work account mean nobody at the provider ever sees our data?
Not exactly. As of October 2026, Google says Workspace content is not human reviewed or used for AI model training outside your domain without permission, and Microsoft says it won't use your data except as you instruct. The exact controls vary by plan, so read the current terms for the plan you actually have.
Is a work account enough for patient records or confidential legal files?
Not on its own. It depends on contracts and setup, not just the account type. Microsoft, for example, says Copilot's support for the federal health privacy law applies to properly configured implementations and not to web search queries. Ask your own lawyer and get the vendor's written agreement first.
What about ChatGPT's business plans?
Read OpenAI's current terms directly before deciding. Look for whether your data is used for training, whether people at the provider can review it, what an administrator can see, and what happens to a user's history when you remove them.
The Bottom Line
Your team's AI habits already exist. The choice is whether they run on rules you wrote or rules nobody wrote. A ban can be right. Doing nothing can be right for a tiny team that has actually checked. Work accounts can be right when customer details are already in play. What isn't optional is the page: which tools, what stays out, who reads it first and whose account it is. Keep it short enough that people will read it, and date it so you know when to look again.
If you'd like to talk through where AI fits in the rest of your business, start a conversation with us. You don't need to have picked a stance first.