Is it safe to put customer data into AI? A small business guide
What GDPR and other privacy rules expect when you use AI, what the big AI companies promise, and 8 questions to ask any AI vendor first.
Ali Hassan
· 6 min read
Quick answer
It can be, if you use business plans or APIs that don't train on your data, have a data processing agreement with the vendor, and send only the data the task needs. In the UK and EU, the regulator expects most AI use with personal data to have a short data protection impact assessment. Avoid putting customer data into free or personal AI accounts.
"Can I just paste this client's email into ChatGPT?"
Someone in your business has probably already done it. Maybe it was a contract, a complaint, or a spreadsheet of customer names. It saved them twenty minutes, and nobody thought twice.
Most of the time, nothing bad happens. But "probably fine" isn't a policy. Here's the plain-English version of what the rules say, what the big AI companies promise, and the questions I'd ask before any customer data goes near an AI tool.
This isn't legal advice. If you handle health, financial or children's data, talk to a data protection specialist. For everyone else, this gets you most of the way.
The question isn't "AI or no AI". It's "where does the data go?"
Using AI with personal data isn't banned anywhere I know of. What matters is the same thing that matters with any software you use: who gets the data, what they're allowed to do with it, how long they keep it, and where it's stored.
The biggest difference in practice is between free or personal accounts and business accounts or the API.
- OpenAI says that by default it doesn't use data from ChatGPT Business, Enterprise or its API platform to train its models.
- Anthropic says that under its commercial terms, content from Team and Enterprise plans and the API isn't used for training by default.
- Google's own pricing page marks the free tier of its Gemini API as "used to improve our products", while the paid tier isn't.
In other words: the free version is often paid for with your data. The business version usually isn't.
Your team
- Is this data actually needed for the task?
- Can names and details be removed first?
The AI tool or app
- Business plan or personal account?
- Is there a data processing agreement?
The model provider
- Is our data used for training?
- Which country is it processed in?
Storage and logs
- How long is it kept?
- Can we delete it?
What the law expects, in plain English
UK and EU: GDPR
If you serve customers in the UK or EU, the UK GDPR or EU GDPR applies. For AI, three parts matter most.
1. You're responsible, the vendor helps. You're usually the "controller" (you decide why the data is used) and the AI provider is your "processor". The law says you must have a written contract with any processor that covers things like only acting on your instructions, keeping data secure and deleting it at the end. In practice, that's the vendor's data processing agreement (DPA). Business plans from the big providers offer one. Free consumer accounts generally don't.
2. You'll probably need a risk assessment. The UK regulator, the ICO, says that in the vast majority of cases, using AI with personal data is likely to be high risk and so triggers the legal requirement to do a data protection impact assessment (DPIA) before you start. For a small business, this can be a few pages: what data, why, what could go wrong, and what you're doing about it. The ICO has a free template.
3. Watch where it travels. Many AI providers process data in the US. For UK businesses, the government's UK-US data bridge allows transfers to US organisations certified under the Data Privacy Framework. Otherwise you need other safeguards, which a good vendor's DPA will cover.
One note: the ICO says its AI guidance is under review after the Data (Use and Access) Act 2025, so details may change. The basics above won't.
EU: tell people when they're talking to AI
The EU AI Act's transparency rules apply from 2 August 2026. AI systems that talk to people must be built so people know they're dealing with AI, and deepfakes must be labelled. If you have EU customers and a chatbot, say clearly that it's AI. It's good manners anyway.
US, Canada and Australia
There's no single federal AI privacy law in the US, but the FTC has warned that AI companies can be in trouble if they break their privacy promises to customers, and that quietly changing terms to allow AI training may be unfair or deceptive. The lesson for you: whatever your own privacy policy promises customers, your AI use has to match it.
In Australia, the privacy regulator is more direct. Its guidance recommends that organisations do not enter personal information, especially sensitive information, into publicly available AI tools.
8 questions to ask any AI vendor
Whether it's a big provider or a small app built on top of one, these are the questions I'd want answered in writing:
| Question | A good answer sounds like |
|---|---|
| Do you train on our data? | "No, not by default. It's in our terms." |
| Do you offer a data processing agreement? | "Yes, here it is." |
| Where is our data processed and stored? | A clear list of countries or regions |
| How long do you keep our inputs and outputs? | A specific number of days, or zero retention |
| Which AI provider do you use underneath? | A named provider, with their terms linked |
| Who on your side can see our data? | Limited staff, for support or abuse checks only |
| Can we delete our data? | "Yes, and here's how." |
| What happens if you have a breach? | A clear process and a notification time |
The fifth question catches a lot of people out. Many AI apps are a layer on top of OpenAI, Anthropic or Google. You need the app's terms and the provider's terms to be good.
A simple policy you can adopt this week
You don't need a 30-page document. Five rules cover most small businesses:
- Approved tools only. Name the AI tools staff can use for work, on business accounts.
- No personal accounts for customer data. Ever.
- Send the minimum. Remove names, account numbers and anything sensitive if the task doesn't need them.
- A person checks anything that goes to a customer. AI drafts, people decide. I explain why in Can AI actually answer my customers?.
- Write it down. Keep the DPA, your DPIA and the list of tools in one folder. If a customer or regulator asks, you can answer in minutes.
If you're weighing up which AI to use at all, it also helps to know what it actually costs. The business plans that protect your data are rarely the expensive part.
Not sure whether your current AI use would pass these checks? Book a free 20-minute call. Tell me what tools your team uses and what data goes into them, and I'll tell you honestly what's fine and what I'd change.