Whenever I talk to other charity leaders about AI, the conversation turns to ethics within a few minutes. Someone has read that every search uses a bottle of water, someone else has a friend who’s an illustrator and is worried about their livelihood, and someone asks whether an organisation built on values should be using this technology at all.
Those are fair questions, and I think charities should be asking them more seriously than most businesses do, because our reputation rests on doing the right thing. I use AI every day at First Days, and I’ve had to work out where I stand on each of these concerns. Here I set out what I’ve found, including where the worries are overstated and where I think they’re right.
“Every search uses a bottle of water”
This is the claim I hear most often, and it’s a good example of a real concern that has become distorted as it’s been passed around.
It seems to trace back to a 2023 study by researchers at the University of California, Riverside and the University of Texas at Arlington, which estimated that ChatGPT could use around a 500ml bottle of water over a conversation of 20 to 50 questions. That figure included water used to generate the electricity as well as water used to cool the data centre, and the paper hadn’t been peer-reviewed at the time. Somewhere along the way “a bottle per 20 to 50 questions” turned into “a bottle per search”, which overstates even the original estimate by a long way.
More recent measurements put the figure much lower. Google tracked its Gemini assistant over a year and found that a typical text prompt used about 0.26ml of water, or roughly five drops, although that figure counts only the water used for cooling and not the water used to generate the electricity. The true number for any given tool probably sits somewhere between the two, and it varies a great deal depending on the model, the data centre and the local climate.
So the bottle-per-search idea is a myth, but the underlying point isn’t. Data centres do use significant amounts of water, and where they’re built in areas that are already short of water, that matters to the communities around them. It’s a reason to use AI thoughtfully and to ask providers hard questions, rather than a reason to feel guilty every time you ask for help with a funding bid.
The bigger picture on energy
The per-prompt figures are small, and I think it’s right to say so, because guilt about individual use doesn’t help anyone make better decisions. The industry-wide picture is a different matter, and this is where I think the concern is justified.
The International Energy Agency estimates that data centres used 415 terawatt-hours of electricity in 2024 and could use around 945 terawatt-hours a year by 2030, with AI as the main driver of that growth. How much of that comes with extra carbon depends on whether new demand is met by renewable energy or by fossil fuels, and that decision sits with governments and the large technology companies rather than with a small charity in Berkshire.
I find it helpful to separate the two levels. At the level of an individual charity, using AI for writing and admin adds very little to your footprint compared with heating your building, your staff travel or replacing laptops. At the level of society, the rapid growth of AI is a genuine climate question, and charities with an environmental mission in particular have every right to raise it loudly. You can hold both positions at once, using AI sensibly in your own work while pressing for the industry to be powered cleanly and to report its impact openly.
“AI is destroying creativity”
There are really two worries bundled together here, and I think one of them is much stronger than the other.
The first is about fairness to creative people. AI models learned from huge amounts of writing, art and music, much of it created by people who were never asked and never paid. Writers, illustrators and musicians are right to be angry about that, and the argument is far from settled. In the UK, the government had proposed letting AI companies train on copyrighted work unless creators opted out, but in March 2026 it stepped back from that plan after a strong backlash from the creative industries, and it has said it will gather more evidence before deciding what to do. I don’t think anyone should pretend this question has been resolved.
The second worry is that using AI makes us less creative ourselves, and here I think the answer depends on how you use it. If you let AI write everything, you’ll end up with bland, interchangeable copy that sounds like every other charity, and funders and supporters will notice. If you use it to take care of the admin, the reformatting and the first rough draft, you get more time for the parts of the work that need a human imagination: the story only you can tell about a family you’ve supported, or the campaign idea that comes from knowing your community. In my experience it has given me more room for creative work, because I’m spending less of my week on paperwork.
Where charities can make a real difference is in the choices we make about creative work. I wouldn’t use AI-generated images to stand in for real people or real stories, and I’d always choose to pay a local photographer or illustrator for work that represents our charity publicly. Being clear about when and how you’ve used AI is part of that too.
Bias and the people we serve
This one gets less attention than water or creativity, but I think it matters most for charities. AI learns from text written by people, so it picks up the assumptions and blind spots in that text. It can describe families in poverty in stereotyped ways, default to a narrow idea of what a “typical” person looks like, or miss the realities of communities that are under-represented online.
For drafting a newsletter, that mostly means reading what it produces with a critical eye and correcting anything that doesn’t reflect the people you know. The risk becomes much more serious if AI is used to make or shape decisions about people, such as who is eligible for support, how referrals are prioritised or how someone’s needs are assessed. I don’t think AI should be making those calls in a charity, and if it’s used anywhere near them, a person who knows the family should always make the final decision.
Where I’ve landed
I don’t think there’s a version of using AI that’s free of ethical questions, any more than there is for flying to a conference, banking with a high-street bank or buying laptops. What I can do is use it in a way I’m comfortable explaining to our trustees, our families and our funders. These are the principles I work to at First Days:
1. Use it where it genuinely saves time, mostly for drafting, managing daily tasks and admin, and skip it where it doesn’t help.
2. Keep people’s personal information out of any tool that doesn’t have proper data protection terms.
3. Keep humans in charge of decisions about people, and never let AI replace a relationship with someone we support.
4. Pay creative people for creative work that represents the charity, and don’t use AI images to stand in for real people or stories.
5. Be open about how we use it, with our team, our trustees and the people we work with.
6. Keep asking questions of the companies that make these tools, about their energy, their water use and how they treat the people whose work they learned from.
The charities I worry about most are those that either avoid AI entirely and fall further behind on admin, or use it without thinking about any of these questions. Most of us will end up somewhere in the middle, and I think that’s a reasonable place to be, as long as we stay thoughtful about it.
If you’d like to go further, I’ve put together a set of free guides on using AI safely in a charity, along with a course for leaders who want to make it part of how they work.
*Sources
• Euronews, study on ChatGPT’s water use per conversation (April 2023)
• DeepLearning.AI, summary of Google’s study on Gemini’s energy, water and emissions (September 2025)
• Data Center Dynamics, IEA projections for data centre electricity use
• Lewis Silkin, UK government rethinks its position on copyright and AI (March 2026)*