Will it leak my customers' data? The honest version.
It's the first question every business owner asks, and the one most AI sellers answer with a shrug and the word "secure". Here's where your data actually goes, who can see it, and the five things to insist on before you let any AI near your customer list.
Where the data goes
When you paste a customer's details into ChatGPT, or an AI tool reads your bookings, the text is sent to a computer owned by the company running the model (OpenAI, Anthropic, Google), processed, and an answer comes back. That part is unavoidable; it's how the thing works. The questions that matter are what happens to the text afterwards, and who else could see it on the way.
There are three different situations, and they get lumped together by people who want to frighten you or sell to you.
1. Free consumer accounts
The free ChatGPT or Claude app on your phone, with default settings, may use what you type to improve the model. That's in the terms. It means a customer's name and problem could in theory end up as a grain of sand in the next version. The risk of it coming back out in a recognisable form is tiny, but it isn't zero, and more to the point it isn't yours to take: it's your customer's data.
2. Business accounts and the proper way in
Paid business tiers, and the technical "doors" that software uses to talk to these models, come with a different promise: your data is not used for training, is kept only as long as needed to answer, and is covered by a contract you can read. This is how every tool we build connects. It's the same promise your email provider makes, and it's enforceable.
3. The "AI employee" from the advert
This is the one to watch. When you sign up to a third-party "does it all" tool, your data goes to them first, then to whichever model they use. Now there are two companies in the chain, and the small one is the weak link. Ask them where your data is stored, who at their company can see it, and what happens to it when you cancel. If the answer is vague, that's your answer.
The five things to insist on
- No training on your data. In writing. Every reputable provider offers this on business terms.
- Read-only wherever possible. An AI that briefs you about bookings doesn't need permission to change them. Give it the least it needs.
- Your systems stay the source. The customer list lives in Wix, Xero, Shopify, wherever it already is. The AI reads it; it doesn't keep its own copy that nobody remembers to delete.
- Every answer shows its source. "Mrs Hartley is due a follow-up" should come with "because her last visit was 2 September and you set six weeks". If it can't show its working, don't act on it.
- A person sends anything to a customer. The AI drafts the reminder; you press send. Nothing goes out of your business without someone reading it.
The UK side
Using AI on customer data doesn't change what the law expects of you under UK GDPR; it's the same duty you have with a spreadsheet. You need a lawful reason to process the data, you need to keep it secure, and you need to tell customers in your privacy notice that you use software providers to run the business. For a small firm that's a paragraph, not a project. If you handle health information, as a clinic does, the bar is higher and worth getting right, which is why the first thing we build for a clinic is always read-only.
Learn this yourself, in an afternoon
Most of the risk isn't in the tools; it's in habits. Someone pastes a client email into a free app to tidy the wording, someone uploads the price list to "summarise it", someone's phone keeps the chat history. Our half-day covers exactly this, for your team, with your own systems: where your data lives, which AI accounts train on it and which don't, how to set the switches, what should never go into a public AI at all (customer lists, contracts, anything you'd call your IP), and how to get the benefit without the exposure. You leave with a one-page rule sheet in your own words that staff can actually follow.
What we do
All five of the above, every time, or the build doesn't happen. We're not a lawyer and won't pretend to be; for anything beyond a plain privacy notice, a local data protection consultant is cheap and worth it. What we can promise is that nothing we build sends your customers anywhere they didn't already live, and that you'll always know what it's reading.