This is a practical checklist, not legal advice. If your firm handles health, financial or children's data at scale, talk to your data protection adviser before you start.
The short answer
The Information Commissioner's Office has been clear that data protection law applies to AI the same way it applies to anything else. Using ChatGPT or Claude doesn't change your duties as the business that collected the data. What changes is how easy it becomes for someone to paste a client's details into a tool the business doesn't control.
So the question isn't "is AI allowed?" It's "have we set it up so client data only goes where we've agreed it can go?"
1. Get everyone off personal accounts
Personal plans are made for individuals. OpenAI says content on its personal plans can be used to train its models unless the user opts out. On its business plans (ChatGPT Business and Enterprise) it says it doesn't train on your data by default. Anthropic says the same of Claude Team and Enterprise.
Business plans also put the account in the company's hands: an admin adds and removes people, and the work doesn't leave with an employee. Microsoft 365 Copilot and Gemini in Google Workspace sit under the business terms you already have with Microsoft or Google.
2. Check you have a processor contract
When a supplier handles personal data on your behalf, it's normally your processor, and UK GDPR requires a written contract with specific terms: what the processing is for, that the supplier only acts on your instructions, security, confidentiality, sub-processors, help with people's rights requests, and deleting or returning the data at the end.
For AI tools this is usually a data processing addendum on the business plan. OpenAI offers one for its business products. Check it's in place, keep a copy, and add the tool to your record of processing if you keep one.
3. Decide what may go in
Sort your data into three piles and write them down.
- Fine: your own templates, price lists, public information, anonymised examples.
- Fine on approved business tools: ordinary client and customer details needed for the job, such as names, emails, order history and correspondence.
- Only with extra checks: health information, financial account details, criminal records, children's data, anything about someone's ethnicity, religion or beliefs. Most first projects don't need any of this.
4. Keep a person in the decision
Drafting a reply, summarising a file or filling in a report is low risk when a person checks the result. Letting AI make a decision about someone on its own, such as rejecting an applicant or refusing credit, is a different matter. UK GDPR has specific rules on solely automated decisions with significant effects on people, and the ICO's guidance covers the safeguards. Keep AI as the drafter and a person as the decider.
5. Tell people
If you use AI on customer data in a way they wouldn't expect, say so in your privacy notice. The ICO's guidance has a chapter on transparency for AI. A plain sentence is usually enough: what you use it for, which supplier, and that a person reviews the results.
6. Do an impact assessment where the risk is higher
A data protection impact assessment is required where processing is likely to be high risk to people. Routine drafting and admin rarely is. Anything that profiles people, makes decisions about them or uses sensitive data might be. The ICO publishes an AI and data protection risk toolkit you can work through.
The ICO notes its AI guidance is under review after the Data (Use and Access) Act, so check its site for the current version before relying on the detail.
A one-page AI policy
Our free policy generator writes one from six questions.
Most small businesses need one page, not a manual. It should say:
- Which tools are approved, on which plan, and who administers them.
- What data may go into them, using the three piles above.
- That a person checks every AI result before it goes to a client or customer.
- That nothing about a person is decided by AI alone.
- Who to ask when unsure, and how to report a mistake.
We go through this page with your team before any client data touches a new tool.
Sources
- Information Commissioner's Office, Guidance on AI and data protection
- Information Commissioner's Office, AI and data protection risk toolkit
- Information Commissioner's Office, What needs to be included in a controller and processor contract
- OpenAI, Business data privacy, security and compliance
- Anthropic, Claude plans and pricing
- OpenAI, ChatGPT pricing and plan comparison
We checked every figure and rule above against these sources by hand on 8 October 2026. Plans and prices change often. If something here is out of date, tell us at hello@leverloft.com.