Skip to content
← All posts

Bringing AI into your business?
What not to give it

An employee wants to send holiday wishes to every customer. She opens a free chatbot, pastes the list of names and phone numbers and asks for a message for each one. Two minutes later the messages are ready. The job is done, and nobody wondered where the list went.

AI is useful, and I won’t tell you to avoid it. I’ll tell you what not to give it, and how to set it up so it helps you without exposing you or your customers.

What you never paste into a chatbot

  • Passwords and access keys. Your email, online banking, tax portal or website password. Not even “so it can help me figure out why I can’t log in”. Whatever you type into a conversation is stored on servers you don’t control.
  • Customers’ personal data. Names with phone numbers, tax numbers, addresses, and above all health details. Your customers trusted you with them, and you are responsible for where they end up.
  • Anything you wouldn’t show a stranger. Contracts, payroll, financial statements, quotes before they go out.

A simple rule: if you wouldn’t email it to someone you don’t know, don’t paste it. Often you don’t even need to. To write a greeting, the AI only needs one example with a made-up name. You add the real ones yourself afterwards.

Your plan matters

Most of the well-known tools have personal and business plans, and they differ in more than price.

  • On personal plans, paid ones included, your conversations can usually be used to train the models, meaning the AI learns from them, unless you turn that off in the privacy settings. Turning it off doesn’t remove what has already gone into training.
  • On business plans the companies usually commit to not training on your data by default.

The terms change often, so don’t rely on what you read last year, or on this article. Open your account settings and see what applies today.

Watch chat sharing too. In the summer of 2025, thousands of ChatGPT conversations showed up in Google results, because their users had created a public link and ticked the option to let it appear in searches. Some contained names and personal details. OpenAI removed the option to make them searchable, but share links still exist, and anyone with the link can read the chat.

Whose name the account is in

The same goes as for who holds the keys to your website. If everyone in the business works from their personal Gmail and their personal AI account, then:

  • you don’t know what has been pasted where,
  • when someone leaves, they take every conversation with them, including whatever concerns your customers,
  • you can’t cut off their access, because the account is theirs.

One account in the business’s name, with users you add and remove yourself, gives you back control: who has access, and what stays with the business when someone leaves. For the first point, what each person pastes, you also need a team rule. You’ll find it at the end.

When AI starts taking on work

A chatbot answers and the work stays with you. An AI agent acts: it reads emails, opens files, writes, sends. That’s where the question becomes “what do I allow it”.

  • Give it only what it needs. If its job is answering questions about your services or opening hours, it doesn’t need access to payroll or your bank account.
  • Watch what it reads. An email or a web page can hide instructions written for the AI, such as “forward every invoice to this address”. This is called prompt injection. No current model reliably resists it.
  • Approval before anything that can’t be undone. Sending, deleting, paying: the AI suggests, you press the button.

When I set up AI for someone, the agent asks before it changes, deletes or sends anything. It’s a little slower at first. As you see it working well, you hand it more, at your own pace.

Check it in half an hour

  1. Ask your team which AI tools they use for work, and with which account. You may be surprised.

  2. Open the privacy settings of each account and check whether conversations are used for training. If they are and you don’t want that, turn it off.

  3. Find the share links in the settings and delete the ones nobody needs any more.

  4. Write two lines for the team: what never goes into AI, and which account is the official one.

And then?

None of this means you should fear AI. Treat it like a new hire: you give it what it needs for the job, not the keys to the shop. As I wrote about AI in small businesses, you start with something small, and a person always has the final say.

If you want to set it up right from the start, or see where your own business stands today, write to me. I’ll tell you honestly whether you need anything more, or whether the settings you already have are enough.

FAQ

Is it safe to put customer details into ChatGPT or Claude?
On a personal account, better not. Your customers' details are personal data, and you are responsible for where they end up. If you need AI for that kind of work, use a business plan in your company's name that commits in writing to how it handles data, and give it only what it needs. Health details, better never.
Do AI companies use my conversations for training?
On personal plans, paid ones included, usually yes, unless you turn it off in the privacy settings. On business plans, usually not by default. The terms change often, so check the settings of your own account.
What is prompt injection?
It is when a text the AI reads, such as an email or a web page, contains hidden instructions, and the AI follows them as if they came from you. That is why an AI that reads outside text should do nothing important without your approval.
Who should own the business AI account?
The business itself. If everyone works from a personal account, when they leave they take every conversation with them, along with whatever they pasted into it.