The useful question about AI in a charity is not which tool to try. It is what happens when the tool is wrong, and who carries that.
Answer it, and most decisions become easy. A summary of your own meeting notes, which you will read before anyone else sees it, is low risk. A funding application, a safeguarding record, or an answer given to a distressed person is not.
If the underlying problem is that your systems don’t talk to each other and nobody has time to fix it, that’s what our digital and tech support is for – with or without AI in the picture.
Start with what these tools actually do
The tools most charities use generate text that reads as though a knowledgeable person wrote it. That is what they optimise for. Accuracy is a frequent side effect rather than the objective.
So when a tool doesn’t know something, it does not stop. It produces something plausible: a statistic that sounds right, a quotation nobody said, a reference to guidance that doesn’t exist, a beneficiary example with convincing detail. It does this in the same confident register as everything else, which is exactly why it slips through.
This isn’t a flaw in a particular product to be avoided by picking a better one. It’s how the technology works. Every decision below follows from it.
Where the risk is low enough
Three conditions make AI use straightforward. You already know the subject well enough to spot an error. Nothing goes out without a person reading it. And a mistake would be embarrassing rather than harmful.
Work that usually clears all three: tidying up something you have already written, turning your own notes into a first draft, shortening a document, suggesting subject lines, drafting a routine reply you will edit, translating your own text where someone will check it, and giving you a starting structure for a document you’ll then write properly.
Notice the pattern. In each case, the substance came from you, and the tool worked on the wording. That is the safe direction of travel.
Where it isn’t
The reverse – asking a tool to supply substance you don’t have – is where charities get into trouble.
Anything going to a funder. An invented figure or an example that didn’t happen is not a typo in an application; it’s a false statement made to obtain money. Funders increasingly ask whether AI was used, and the reasons are set out more fully in why grant applications fail.
Anything going to a regulator. Your annual report, your accounts narrative, anything filed with the Commission. These are documents trustees sign.
Facts, figures and law. If you ask for the current threshold, the relevant regulation or the sector statistic, you may get a confident answer that is out of date, from another country, or invented. Everything of this kind has to be checked against the actual source, at which point the tool has saved you nothing.
Anything a person in difficulty will rely on. A chatbot answering opening hours is fine. A chatbot in front of someone who might be in crisis is a safeguarding question, not a service design one, and the honest answer for most small charities is not yet.
Decisions. A tool can lay out options. It cannot weigh them against your charity’s purposes, and it is not accountable for the outcome.
What must not go in
This is the part with legal consequences, and it gets the least attention.
Text typed into a public AI tool leaves your control. Depending on the service and its settings, it may be stored, reviewed by staff at the provider, processed outside the UK, or used to improve the model.
So: no beneficiary names or circumstances. No safeguarding information, ever. No case notes, referrals or health details. No supporter lists or donor data. No staff or trustee personal information. No confidential board papers.
“I’ll anonymise it” is less protective than it sounds. In a small charity working in one area, the combination of an age, a condition and a location often identifies someone to anyone local.
Under UK GDPR, your charity is the data controller and remains responsible for what happens to personal data you put into a third-party tool. If you intend to use AI with anything containing personal data, that needs a proper look at your lawful basis, your privacy notice and the provider’s terms – and the ICO’s guidance on AI and data protection is where to start.
Business accounts of some services offer stronger data terms than the free versions. That changes the analysis, but it doesn’t remove the need to do it.
Not sure what your team is already putting into these tools?
In most small charities, the answer is that nobody has asked, and the honest starting point is finding out rather than writing a policy. A free Charity Consultation will help you work out where you stand.
Trustees remain responsible
AI changes nothing about accountability. If a decision is taken on the basis of an AI-produced analysis, it is still the trustees’ decision, and the same seven principles apply – whether the board was sufficiently informed, whether it took account of what was relevant, whether it can justify the outcome. That framework is in how trustees should make decisions that hold up.
“The software suggested it” is not a defence. Nor is it a reason the Commission would find interesting.
The practical consequence: where AI has contributed to something significant, the board should know it was used, and someone should have checked it.
A policy short enough that people read it
Most small charities need one page, not ten. Six things:
Which tools are approved, and whether free or paid accounts are used, since the data terms differ.
What must never be entered – the list above, written out plainly, so nobody has to interpret it.
What has to be checked before anything is sent, published or filed, and who does the checking.
Where AI must not be used at all: funding applications, regulatory filings, safeguarding, and anything a service user relies on.
Whether and how you tell people. Being open that a chatbot is a chatbot, with an easy route to a human, is the minimum. For written communications, disclose where it matters and where people would reasonably want to know.
And who owns this – one named person who keeps the list current and answers questions.
That belongs with your other charity policies, with a review date, and it should describe what your team actually does rather than what you would like them to do.
The question nobody asks first
Before adopting any of this: what is the problem?
If your charity is stretched, AI may help at the margins. It will not fix an income model that doesn’t work, a board that can’t decide, a database nobody maintains, or a donation page losing half the people who reach it. Those are the things that actually constrain small charities, and none of them has a software answer.
Time saved on drafting is real and worth having. It’s also the smallest of the problems most small charities have. If AI adoption is being discussed as a strategic priority, it’s worth checking what it’s displacing.
The charities getting value from these tools are using them narrowly, on work they already understand, with someone checking the output. The ones getting into difficulty are using them to produce things they cannot verify.
If what you actually need is systems that work together and a team that trusts them, that’s a defined piece of work – and a one-off digital project will get you there.
