Design the boundary. Then use the tool. For a charity, the useful question is not whether AI is allowed in the abstract. It is what data is being used, where it goes, what the AI is allowed to do, and who checks the result.
- Approved sources only.
- A known place where prompts and answers are processed.
- A person who reviews anything that matters.
- A clear rule for how long material is kept.
Charities handle information that deserves that treatment: records about the people they support, donors, volunteers and staff. AI can help with search, summaries and first drafts. It should not become an informal habit of pasting those records into a tool nobody has agreed.
Start with what the AI can see
Name the sources before anyone starts typing. A controlled project may use approved policies, reporting exports, curated folders or specific database views. It should not depend on staff pasting sensitive material into unmanaged tools.
If a source has not been approved for this job, it stays out. That includes notes that happen to be on someone’s desktop, a full case file “just to give it context”, and anything that would be special category data unless you have already decided, with a lawful basis and a separate condition, that this use is allowed. The ICO guidance on AI and data protection expects that care when an AI tool is the one doing the reading.
Name where the data goes
Your own system, or a public tab, are not the same thing. A public chat tool is someone else’s system reading the notes you paste into it. A private setup, on a system the charity controls, with agreed processing, is a different design. It is also easier to explain to trustees, staff and the people whose information you hold.
Write down the system that will hold the prompts and the answers. If the honest answer is “a public website”, stop and choose a different job, or a different tool. Hosting location alone does not settle the question, but you should still be able to say where processing happens. For public ChatGPT specifically, read can charities use ChatGPT with sensitive data.
People stay in charge
AI can draft. People still sign it off. AI can search, summarise and prepare. People still own decisions, exceptions, safeguarding judgement and anything that affects a person. A good system makes that review visible, rather than leaving it as a hope that staff will remember to check.
That is the same rule we apply when we build Ask Your Data: answers come from the charity’s own records, with a trail of what was asked, and a person approving what goes out. It is also the test for any staff assistant that prepares a first draft from approved sources.
Agree how long material is kept
Prompts, uploaded extracts and generated drafts are records of their own. Decide how long they are kept, who can see them, and when they are deleted. Do not let a helpful tool become an unofficial archive of case notes or donor details sitting in a chat history that nobody is reviewing.
Questions trustees will ask
A board should be able to explain the project in ordinary language. What job does this improve? Which approved information does it use? Where are prompts and outputs processed? Who reviews the result before it matters? Those questions are set out more fully in the trustee section of 12 questions to ask before your first AI implementation. The Charity Commission essential trustee guidance (CC3) keeps that duty with the board.
Questions
Can a UK charity use AI on case notes and service records?
It can, when the project names approved sources, where data is processed, who reviews the output, and how long material is kept. That is a design choice, not a blanket yes. Special category data needs a lawful basis and a separate condition under UK GDPR.
Is pasting into ChatGPT the same as a private workspace?
No. A public tab is someone else’s system. A private setup on a system the charity controls, with agreed processing and a person signing off what goes out, is a different arrangement.
Where should we start?
Pick one job that does not need the most sensitive records, or a job that can run on an approved extract with a named reviewer. If staff are already pasting sensitive material into a public chat, stop that first and agree a safer route.
If you want help designing a private, reviewable first job around your own records, book a free 40-minute call by phone or Teams. We’ll send a written summary of next steps.
Book a free call