Field Notes
What to check before AI writes your care home's notes
The adult social care sector now has its own guidance on AI. A few plain checks to run before you let a tool draft your records.
The sector wrote down what good looks like
For a while, anyone weighing up an AI tool for a care home had to work out the ethics on their own. That changed in March 2025, when the first AI in Social Care Summit at the University of Oxford put out a Call to Action, guidance, a white paper on generative AI, and a pledge for technology suppliers. The work was led by Oxford's Institute for Ethics in AI, the Digital Care Hub and Casson Consulting. A year on, the March 2026 summit agreed the shape of an AI in Social Care Alliance to carry it forward. So before you shortlist anything, you have sector guidance to hold a supplier against, and you did not have to write it yourself.
Ask whether the supplier signed the pledge
The Pledge for Technology Suppliers sets out principles a responsible AI supplier should meet: co-production with the people who use and give care, transparency, safety, fairness, data protection and accountability. Several suppliers have already signed. So ask yours plainly. Have you signed the pledge, and if not, why not. Show me how a care worker would know when the words in front of them were drafted by AI. A refusal is not always a red flag, but the reasons a supplier gives will tell you a good deal about how they think.
Treat AI-written notes as a draft, not the record
Generative AI can be fluent and confidently wrong in the same sentence. The Information Commissioner's Office is clear that the accuracy principle applies to what these tools produce, and that a higher bar applies where the output feeds decisions about a person. Care records are exactly that kind of output. Keep a person reading, correcting and signing off every note before it stands as the record, the same way you would with an agency worker on their first shift. The tool saves the typing. It does not carry the responsibility.
Do the data protection work before go-live, not after
Care notes are health data, which UK law treats as special category and expects you to handle with extra care. That means a data protection impact assessment before the tool touches a real record, not once it is already in use. Write down who the data controller is, where any audio or text goes, who can see it, how long it is kept and how it is deleted. If the supplier cannot answer those questions in writing, you are not ready to trial the tool, whatever the demo looked like.
Trial one note type beside your current way
Do not switch the home over in one go. Pick one kind of entry, run the AI draft next to the note a person would have written, and compare them for a fortnight. Count how often the draft needed correcting and what it got wrong. If it earns its place, widen it slowly. If it does not, you have spent two weeks and no more. This is the sort of scoping our AI Operations Sprint is built for, and our newsletter carries the next set of notes as the guidance develops.