Let's start with an honest question
When you hear "AI", what comes to mind? A clever robot? A cheat code? Something that might take your job?
Let's put all that to one side for a moment. In this first level, we'll work out what an AI assistant like ChatGPT, Gemini, Claude or Microsoft 365 Copilot is really doing when it writes for you. Then we'll look at where it helps in charity work, where it lets you down, and what the sector is actually doing with it. No hype, just the plain picture.
So how does it actually write?
Here's the simplest way to think about it. You know how your phone suggests the next word as you type a text? An AI assistant does the same thing, on a much bigger scale.
It learned from a huge amount of written text. When you give it a prompt, it predicts the most likely next word, then the next, then the next, until it has a full answer. It's very, very good at sounding like a fluent human writer.
But here's the catch. Predicting what sounds right isn't the same as knowing what is true. Unless a tool is set up to search or read your documents, it isn't checking facts. It's producing likely-sounding text.
Why does it sound so sure when it's wrong?
This is the part that catches people out. Say you ask for the deadline of a local grant scheme. The AI may give you a neat, confident date that's completely made up. People call this a "hallucination".
Why does it happen? In September 2025, OpenAI researchers published a paper arguing that the way these models are trained and tested rewards guessing over saying "I don't know". A guess sometimes scores a point. "I don't know" never does. So the model learns to sound sure of itself.
The Charity Commission said much the same in an April 2024 blog post. It warned that generative AI can "confidently produce inaccurate, plagiarised, copyright infringing or biased results" without any awareness that something's wrong. It also reminded trustees that they remain responsible for decisions, and those decisions shouldn't rest on AI-generated content alone.
So the rule of thumb is simple: confident tone tells you nothing about accuracy.
Where it genuinely helps
Now for the good news. For lots of everyday charity tasks, AI is a brilliant first-draft partner. Think about:
- Turning your scribbled meeting notes into a tidy summary
- Drafting a friendly volunteer thank-you email
- Suggesting five different ways to open a social media post
- Rewording a dense paragraph into plain English
- Brainstorming questions for a volunteer feedback survey
What do these have in common? You know the subject, you can spot mistakes quickly, and nothing bad happens if the first draft is a bit off.
Where it falls short
AI struggles when the task needs facts it doesn't have, like your charity's real outcomes, current funder rules or local service details. It knows nothing about the family you supported last Tuesday. And it can't take responsibility. It can't decide whether someone is at risk, or whether a grant application is honest.
Sorting your own tasks
Here's a handy way to sort any task into three buckets.
Good fit: low-risk drafting and tidying where you can easily check the result. Rewording a newsletter intro, say.
Use with care: tasks where AI can help, but facts, tone or fairness really matter, so you check every line. A first draft of a funder report is a good example. The structure's fine, but every number must come from your own records.
Keep human: judgements about people, safety and responsibility. Deciding whether to raise a safeguarding concern, choosing who gets an emergency grant, or comforting a grieving supporter on the phone. These stay with you.
Not sure where something fits? Ask yourself: "If this were wrong, who would be hurt, and would I notice?"
We'll cover which tools you're allowed to use, what must never go into a prompt and how to check output properly in the next few levels. For now, just get used to sorting.
What is the sector actually doing?
So, are other charities really using this stuff? Yes, and fast.
The Charity Digital Skills Report 2026, led by Zoe Amar and Nissa Ramsay, came out in July 2026. It found that 79% of organisations are now routinely using AI. That's up from 76% in the 2025 report. The 2026 survey had 807 responses from UK charities, and 71% of them were small charities. It also found that 45% use AI for grant fundraising, and 43% named data protection as a barrier.
Notice how we quoted that? Every time you use a statistic, say three things: who ran it, when it was published, and who it covers. A UK survey from mid-2026 tells you a lot about UK charities in 2026. It tells you much less about a US food bank, or about next year. Statistics on AI go stale quickly, so always mention the date.
Your turn
Grab a notebook. List ten tasks from your actual week, both the dull ones and the important ones. Next to each, write "good fit", "use with care" or "keep human", plus one short reason.
Then look over your list. Which "keep human" tasks surprised you? Hold on to that list, because we'll come back to it.
Checklist
- I can explain next-word prediction in one sentence
- I know a confident tone doesn't mean a correct answer
- I've sorted ten of my tasks into three buckets
- I can quote one AI statistic with its source, date and coverage
Key takeaways
- AI assistants predict likely text. They don't automatically check facts.
- They can sound completely sure while being wrong, so tone is no guide to truth.
- AI is great for low-risk drafting, needs care where facts matter, and shouldn't make judgements about people or safety.
- The Charity Commission says trustees stay responsible for decisions, and those decisions shouldn't rest on AI output alone.
- Charity Digital Skills Report 2026 (July 2026, 807 UK respondents): 79% of organisations routinely use AI, up from 76% in 2025.