You’ve got a tricky customer email to answer, and an AI chat assistant could help you get the wording right in thirty seconds. The fastest way is to paste in the whole thread: the customer’s name, their account details, the history of their order. Should you?
The short answer: most of what you write for your business is fine to use with AI. A small slice of it never belongs there. The trick is knowing which is which, and building a habit so you don’t have to think hard about it every time.
Treat it like a conversation in a public place
Here’s the simplest rule I know. Treat any AI chat tool like a conversation in a coffee shop where you can’t see who’s sitting at the next table.
You’d happily talk through a scheduling problem, a social media caption, or how to word a polite reminder about a late invoice. You’d never read a customer’s credit card number or medical history out loud. The same instinct works for AI.
That’s not because these tools are secretly up to something. It’s because once information leaves your hands, you no longer control where it’s stored, who at the provider might review it, whether it’s used to improve the product, or what happens if your account is ever compromised. For ordinary business writing, none of that matters much. For sensitive personal details, it matters a lot.
What never goes in
Keep these out of general-purpose AI chat tools, every time:
- Social Security numbers, tax ID numbers, and driver’s license numbers
- Full bank account, credit card, or payment details
- Passwords, PINs, and login codes, yours or anyone else’s
- Medical or health information about customers, patients, or employees
- Anything you’ve promised to keep confidential, like contract terms under an NDA, a client’s legal matter, or unreleased financials
What’s usually fine
Most everyday business work involves none of that. These are generally safe to use with AI help:
- Drafting a reply about scheduling, shipping, or a general question
- Social media captions, newsletters, and blog posts
- Job descriptions and interview questions
- Cleaning up a proposal or contract template, without the client’s sensitive details
- Summarizing your own meeting notes, as long as they don’t include anything on the “never” list
The placeholder trick
What about the in-between cases, like a sensitive customer email you genuinely want help wording? Swap out the identifying details before you paste, then put them back afterward.
Instead of pasting this:
Jane Morales, account #4471-2290, was double-charged $312.40 on her Visa ending 8812…
Paste this:
A customer was double-charged on their card. Help me write a short, warm apology that explains the refund will post in 3 to 5 business days.
You get the same help with the wording, and the AI never sees anything that identifies a real person. You add the real details back yourself before you hit send.
In my book for small business owners, a bookkeeper named Priya caught herself halfway through pasting a client’s full name, Social Security number, and account details into a chat assistant. She deleted the message, and now she uses placeholders like “the client” or “Client A” for anything sensitive. It costs her a few seconds per email.
The two-second pause
You don’t need a policy binder. You need a habit. Before you paste anything into an AI tool, glance at it and ask one question:
Does this contain a number or detail that could identify a real person in a sensitive way?
If not, go ahead. If it does, swap in placeholders first. After a week or so, the check becomes automatic.
Check your settings once
Most AI chat tools let you control whether your conversations are used to improve the product. It’s worth five minutes to look.
In ChatGPT, for example, you can go to Settings → Data controls and turn off “Improve the model for everyone.” ChatGPT also offers Temporary Chats, which don’t appear in your history and aren’t used to train OpenAI’s models. And OpenAI says that by default it doesn’t use content from its business plans, such as ChatGPT Business and Enterprise, to train its models (OpenAI’s data controls FAQ has the details). Other providers, including Google, Microsoft, and Anthropic, have their own settings and business plans, so check the data or privacy section of whichever tool you use.
Changing a setting is a good step, but it isn’t permission to paste sensitive information. The placeholder habit still applies.
Does your industry have extra rules?
Some businesses work under legal rules about where data can go. Healthcare is the clearest example, and finance, legal services, and insurance often have their own requirements. A general-purpose chat tool may not meet them, even with training turned off. If you’re in one of these fields:
- Find out what your industry requires. Often one question to your industry association, compliance adviser, or attorney answers it.
- For anything touching regulated data, use tools built for your industry that come with the agreements and protections it requires.
- Keep using general tools freely for work that touches no regulated data, like your newsletter or a job post.
I’m not a lawyer, and this isn’t legal advice. If you’re unsure, a short conversation with someone who knows your industry’s rules will save you trouble later.
Choosing tools you can trust
You don’t need to read every privacy policy word for word. A few shortcuts go a long way:
- Stick with established providers for anything involving real customer information.
- Make sure there’s a clear privacy policy you can actually find.
- Be wary of free tools with no obvious business model. If it isn’t clear how a tool makes money, your data may be part of the answer.
The short version
- Most business writing is fine to use with AI.
- Never paste Social Security numbers, account or card numbers, passwords, or health information.
- Use placeholders for anything sensitive, and add the real details back yourself.
- Take two seconds before every paste.
- Check your data settings once, and find out whether your industry has extra rules.
