Is ChatGPT safe to use for my business? What should I never put into it?

ChatGPT is safe enough for most small business work, with one hard line: never paste anything you would not be willing to email to a stranger. Customer records, passwords, employee files, and anything covered by a contract you signed stay out of the box. Everything else is usually fine.

That is the whole answer. The rest is why, what the two big companies say in writing, and the one thing nobody tells you about the delete button.

What actually happens to what you type into ChatGPT?

Three separate things, and people mash them together into one worry.

First, the chat is stored on the company’s servers so it can answer you. True of every one of these tools, and not optional.

Second, on some plans your chat may be used to train the model. That is the part people mean when they say “is it safe.”

Third, on some plans a person at the company may read a sample of chats to check quality. Google says so plainly in its own help pages. Most owners have no idea it is even possible.

Which of the three applies to you depends on the version you are signed into. Not on how careful you are, and not on how you word the prompt. The account decides it.

Does the free version use my business data to train the model?

On a personal account, usually yes, unless you go and turn it off. On a paid business account, no, by default. Here is what the two companies most small businesses actually use put in their own documentation. I checked both again on 7 October 2026.

Where you are typing	Used to train the model by default?	What the company says
ChatGPT free or Plus, personal account	Yes, unless you turn it off	OpenAI says it may use your content to train its models, and points you to Data Controls to opt out.
ChatGPT Business, Enterprise, or the API	No	OpenAI: “By default, we do not train on any inputs or outputs from our products for business users.”
ChatGPT Temporary Chat	No	Temporary chats do not appear in your history and are not used to train the models.
Google Gemini with Keep Activity on	Yes, and a person may read it	Google uses that activity to improve its services “including training generative AI models,” with help from human reviewers.
Google Gemini with Keep Activity off, or a temporary chat	No	Kept for 72 hours so it can answer you and so Google can catch abuse, then dropped.

Sources: OpenAI Help Center, How your data is used to improve model performance, and Google’s Gemini Apps Privacy Hub, last updated 24 September 2026. Both are the companies’ own published policies, not a third party summarizing them. Policies change, so check the page yourself rather than trusting my table a year from now.

What should I never put into ChatGPT?

Seven things. This list does not change based on which plan you are on, because the plan is not the only way information leaks. Your own screen in a coffee shop counts too.

1. Customer records. Names attached to addresses, phone numbers, card details, medical information, or anything financial about a real person.
2. Passwords, logins, and API keys. Anything that would let somebody in the front door of something you own.
3. Anything covered by an agreement you signed. Nondisclosure agreements, vendor contracts, franchise paperwork. Read what you already promised before you paste.
4. Employee records. Pay, write-ups, health information, immigration paperwork. This is the one small businesses get wrong most often, because it feels internal.
5. Unreleased pricing and live bids. Especially if you are bidding against somebody else this week.
6. Raw financials. Full bank exports, tax returns, payroll files. Summaries and made-up example numbers are fine. The actual file is not.
7. Anything you would not email to a stranger. The catch-all. If the thought of a stranger reading it makes you wince, that is your answer.

Notice what is not on that list. Your marketing. Your service descriptions. Your rough draft of an awkward email. Almost everything a small business owner actually wants help with is nowhere near this line.

Does deleting the chat delete the data?

Not entirely, and this is the part that gets left out of every “AI safety tips” post I have read.

Google says that if a chat has already been looked at by a reviewer, deleting your activity does not delete that copy. It has been disconnected from your Google account and it is kept for up to three years.

Read that again slowly. The delete button reaches the copy in your history. It does not reach back and pull the copy somebody already pulled out for review. None of that is hidden. It is written on the page in public. But it means the honest mental model is once it is in, assume it is in.

Which is why the seven-item list matters more than any setting. Settings protect you going forward. Nothing protects you backward.

So what is actually safe to put in?

Most of your week, honestly. The work that eats a small business owner’s evenings is almost never confidential. It is repetitive.

Safe with no special setup: writing your own marketing copy, drafting a reply to a review, turning rambling notes into a clean email, explaining a piece of jargon somebody sent you, planning a week of posts, working out how to say something difficult politely.

Safe with a small change: anything that involves a real customer, if you strip the name first. “A customer is upset that their order was three days late, help me write an apology” is a completely different thing from pasting the customer’s name, address, and order number. You get the same quality of help from the anonymous version. There is no reason to hand over the identity.

That one habit, take the name out first, does more than every other precaution combined.

Do I need to pay for the business version?

Only if the answer to one question is yes: does confidential information genuinely need to go into this tool for the work to happen?

For a contractor writing job descriptions and quotes, no. Turn training off on the free account, keep the seven items out, and you are done. For a practice handling patient information, or a firm under a client confidentiality clause, the business tier is the price of using the tool at all, and you should be talking to whoever handles your compliance before you start.

I do not sell these tools and I do not take a commission on any of them, which is why I am comfortable telling most owners the free version plus one setting change is enough.

How do I handle this in my own business?

Uncle Ron Media is one person, in Bowersville, working with small businesses across Hart, Franklin, Elbert, and Stephens counties. This company runs on AI systems I built for myself, and I have written before about exactly what they do and do not do.

The rule I published there is the same rule I would give you: a human reads everything before anything is sent. AI prepares, a person decides. That rule exists because these tools are equally confident whether they are right or wrong, and confidence is not accuracy.

The privacy version of the same rule is just as short. Client work gets described to the tool, never handed to it. Names come out. Logins never go in. It has cost me nothing in usefulness and it means I have never had to write an awkward email to somebody explaining where their information went.

When does none of this apply to you?

If you are not putting anything into these tools yet, none of this is a reason to stay out. The risk of a small business quietly using ChatGPT to write better emails is very close to zero. What gets businesses in trouble is pasting something in that they had already promised somebody they would protect.

Handle that one thing and the rest of it is just a faster way to get your Tuesday back. If you want the plain-English version of what these tools are actually for, that is what Plain AI means.

Frequently asked questions

Can my competitor see what I typed into ChatGPT?
No. Nobody is searching your chats. The realistic risk is not a competitor reading your conversation, it is your information becoming part of what the model learned, or a reviewer at the company reading a sample. Both are handled by the plan you are on and the settings you choose.

How do I turn off training on a free ChatGPT account?
In ChatGPT it is under Settings, in Data Controls. OpenAI also offers an opt out through its privacy portal. Once you opt out, new conversations are not used to train the models. It does not reach backward through old ones.

Is it against the law to put customer data into ChatGPT?
There is no single law that says do not paste this into a chatbot. What there usually is, is a promise you already made: your own privacy policy, a client contract, a card processing agreement, or health and financial rules for your industry. That paperwork is what decides it, not the tool. If you are unsure, ask whoever wrote that paperwork.

Are the other AI tools any different?
The pattern is the same across all of them. The free consumer version usually trains on what you type unless you say otherwise, and the paid business version usually does not. Check the plan you are actually signed into, because most people are signed into the free one.

What is the single easiest thing I can do today?
Take the name out. Before you paste anything about a real customer or employee, delete the identifying part. It takes four seconds and it removes most of the risk.

Should I write a rule for my staff?
Yes, and it can be one sentence on one page: do not put customer information, logins, or anything under contract into an AI tool, ever. That single sentence covers more ground than a ten page policy nobody reads.

Not sure whether your own setup is a problem?

If you are not sure whether what you are already doing is a problem, that is a 20 minute conversation, not a project. Book a free 20 minute assessment and I will tell you plainly whether you need to change anything, including if the answer is no. Get in touch here.