AI for clinics and care providers, for the admin

Each example works on information with no client details in it, or with them removed first.

• Rotas. 'Here's availability for 14 carers, the visit runs and who drives. Draft next week.' You get a starting grid to fix, using initials rather than names.
• Policies. A first draft of a lone-working policy in your format, checked against current regulations and your own practice.
• Training. Your moving-and-handling procedure turned into a ten-question quiz for new starters.
• Letters. Appointment changes and fee updates to clients or families, written plainly and kindly.
• Recruitment adverts. A clear advert from your job description, with the pay and hours you set.
• Summarising guidance. A long public document boiled down to what changes for a small provider, before you read the sections it points to.

Team meetings can be summarised too, not consultations; the meeting notes page covers telling people first.

Health information is special category data under UK GDPR. The ICO says you need an extra condition under Article 9 to process it, and a DPIA for any processing likely to be high risk. In practice: no client health details in free AI tools at all, and a DPIA before any AI tool touches health information on a business plan.

For the rest of the admin, a business plan is still the safer place to work: OpenAI and Anthropic don't train on ChatGPT Business or Claude Team content by default, and Copilot Chat used with a work account doesn't feed Microsoft's models.

We couldn't confirm the CQC's current position on AI, so we haven't summarised it here; check its website before changing how records are handled. A short AI policy that staff have actually read is the most useful single control.