# Green AI for charities

> A guide for charities and non-profits on using AI where it helps the mission, with low cost and energy, safe beneficiary data and honest communications.

Source: https://greenai.guide/guides/charities/  
Updated: 2026-11-19

## In short

For a charity, green AI means using AI only where it clearly helps your mission, choosing small and free tools where they are enough, and protecting the people you serve. A short responsible-use policy, a few supplier questions and honest communications cover most of what a small organisation needs.

## Key points

- Most charities already use AI: the Charity Digital Skills Report 2026 found 79% of UK charities routinely use it.
- Energy per text request is small, but image generation and long automated workflows use far more, and volume adds up.
- Use AI where it saves real staff time or improves services; skip it where a template, search or existing tool will do.
- Never put personal data about beneficiaries into a public AI tool without a lawful basis and a check of where the data goes.
- A one-page AI policy covering purpose, data, energy and honesty is enough for most small charities to start.

## Why should a charity think about green AI?

Because AI is already part of most charities' work, and how you use it affects your costs, your data risks and how
well your actions match your values. Thinking it through now is cheaper than fixing problems later.

AI use in the sector is now the norm. The Charity Digital Skills Report 2026, based on 807 responses from UK
charities, found that 79% routinely use AI, up from 76% a year earlier. The same report found that 35% lacked trust in
AI tools, up from 15%, and that only 36% of small charities had an AI policy. Interim findings published earlier in
2026 found that 39% of charities were worried about AI's impact on the environment.

The figures are from the UK, but the same questions apply to non-profits everywhere: limited budgets, sensitive
data, and supporters who expect you to live your values.

## How much energy does a charity's AI use actually take?

For ordinary text tasks, not much per request. For images, video and heavy automation, a lot more. What matters most
is the type of task and how often it runs.

- **Text prompts.** Google reports that a median text prompt to its Gemini apps used 0.24 Wh. OpenAI's chief
  executive has said an average ChatGPT query uses about 0.34 Wh. Both are self-reported, using the companies' own
  methods, so treat them as rough guides.
- **Images.** In tests by Luccioni, Jernite and Strubell, generating images used on average about 2.9 kWh per 1,000
  images, more than a thousand times the 0.002 kWh they measured for classifying 1,000 pieces of text.
- **The bigger picture.** The International Energy Agency estimates data centres used around 1.5% of the world's
  electricity in 2024, with AI the main driver of growth. Each organisation's share is small; the total is not.

So a few staff drafting letters with AI is a small footprint. Regenerating dozens of campaign images, or an
automated system that runs AI on every incoming email, is where the energy and cost go.

## When is AI worth it for a charity?

AI is worth it when it frees staff time for the mission, improves a service, or does something you could not
otherwise afford, and when the risk to the people you serve is low. It is not worth it when a simpler tool would do,
or when errors could harm someone.

Ask four questions before adopting a new use:

1. **What is the benefit?** Hours saved, more people reached, better decisions. Be specific.
2. **Is there a simpler way?** A template, a mail-merge, a search, an existing feature in software you already pay
   for.
3. **What could go wrong?** Wrong information given to a vulnerable person, bias, leaked data. Who checks the output?
4. **What does it cost?** Subscription, staff time to check results, energy, and the cost of switching later.

Uses that often pass the test:

- drafting and editing routine text, such as funding bids, reports and social posts, with a human editing the result;
- summarising long public documents, such as consultations or research;
- translating information for communities, checked by a fluent speaker where it matters; and
- accessibility support, such as captions and plain-English versions.

Uses that need much more care:

- anything that makes or shapes decisions about individual beneficiaries;
- chatbots giving advice on health, money, legal or safeguarding matters; and
- generating images of people or places that supporters might take as real.

## Which free tools and courses can help?

Many charities can do most of what they need with free or low-cost tools and the AI built into software they already
use. Learning to use them well saves more energy and money than buying more.

- **Use the AI in your existing software** (office suites, email, design tools) before adding new subscriptions.
  Check its data settings.
- **Choose smaller or "lite" models** for simple tasks where you have the option. UNESCO and University College
  London found that smaller, task-specific models could cut energy use by up to 90% on suitable tasks, and that
  shorter prompts and responses could cut it by more than half.
- **Write short, clear prompts** and ask for short answers. You will usually get better results too.
- **Avoid repeated regeneration.** Refine the request instead of pressing "try again" many times.
- **Take a short course.** Free introductions to AI and to responsible and green AI are listed in our
  [courses directory](/courses/). Skills, more than tools, are often what holds a small team back, so this is time
  well spent.

Our [use AI efficiently](/practice/use-ai-efficiently/) guide has more everyday tips.

## What should a charity's AI policy cover?

A good charity AI policy is short, practical and owned by someone senior. It should tell staff and volunteers what
they may use, what they must never do, and who to ask.

Include:

- **Purpose:** AI is used to support the mission, not for its own sake.
- **Approved tools:** a short list, with who approves new ones.
- **Data rules:** no personal or sensitive data about beneficiaries, donors, staff or volunteers in public AI tools;
  what is allowed in tools covered by a data agreement.
- **Human checking:** who reviews AI output before it is published, sent or acted on.
- **Energy and proportion:** prefer the simplest tool; avoid heavy image and video generation unless it adds clear
  value; switch off AI features you don't use.
- **Honesty:** when you tell people that AI was used, and how you describe it publicly.
- **Review:** when the policy will be looked at again, for example every year.

Our [green AI policy](/practice/green-ai-policy/) guide includes a template you can adapt.

## What should we ask AI suppliers?

Ask about data first, then energy and cost. Even small charities can ask these questions, and suppliers who cannot
answer them clearly may not be the right choice.

- **Data:** Where is our data stored and processed? Is it used to train your models? Can we switch that off? How
  long is it kept? Can we delete it?
- **Security and compliance:** Which data protection laws do you follow, for example UK GDPR or the EU GDPR? Will
  you sign a data processing agreement?
- **Energy:** Can you tell us the energy use per request, or which data centres you use and how they are powered?
- **Model choice:** Can we use a smaller model for simple tasks?
- **Cost and lock-in:** Is there a charity discount? Can we export our data if we leave?

The [sustainable AI procurement](/practice/sustainable-ai-procurement/) guide has a fuller list.

## How should we handle beneficiary data?

Treat beneficiary data as the most sensitive thing you hold, and keep it out of general-purpose AI tools unless you
have checked the law, the contract and the risk. The people you serve may be vulnerable and may not be able to object.

- **Lawful basis.** Make sure you have a lawful basis to use personal data in this way, under your local data
  protection law.
- **Impact assessment.** Do a data protection impact assessment before using AI on personal data, especially for
  anything new or high-risk.
- **Minimise.** Remove names and identifying details where you can; use the least data needed.
- **Keep humans responsible.** AI should not make decisions about individuals' support on its own.
- **Ask the people affected.** Find out how the people you serve use and feel about AI. Talking to them can shape
  better and fairer uses.

## How do we talk about our AI use honestly?

Be open about what you use AI for, and modest about its environmental footprint. Supporters trust charities that
explain their choices; they are wary of claims that sound too good.

- Say what AI helps with ("we use AI to draft first versions of reports, which staff then edit").
- Explain how data is protected.
- Don't call your AI use "green" or "carbon neutral" without figures and a method. Efficiency gains can also lead
  to more use overall, the [rebound effect](/glossary/#rebound-effect).
- Label AI-generated images where supporters might take them for real photographs.
- If you campaign on climate or nature, be ready to explain why a particular use is worth it.

See [honest green claims](/practice/honest-green-claims/).

## Should a charity build its own AI?

Rarely. Building or training your own model is expensive and uses far more energy than using an existing one, and
most charity needs are met by tools that already exist. If you have a specialist need, such as sorting thousands of
case notes or answering common questions from a trusted knowledge base, start by asking whether an existing small
model, or a simple search of your own documents, will do.

If you do commission something, ask the developer to use the smallest model that meets your quality standard, to
reuse answers rather than generate them each time, and to report how much the system is used and what it costs to
run. Plan for who will maintain it when the project funding ends.

## What to do next

1. **List the AI tools in use** across your charity, including those built into existing software and those staff
   use on their own.
2. **Write a one-page AI policy** using our [policy guide](/practice/green-ai-policy/), and agree who owns it.
3. **Set a clear data rule** today: no beneficiary, donor or staff personal data in public AI tools.
4. **Pick one task** where AI could save real time, try it for a month, and check the results honestly.
5. **Cut waste:** shorter prompts, fewer regenerations, and AI features switched off where they aren't used.
6. **Book staff and volunteers onto a free course** from the [courses directory](/courses/).
7. **Ask your main AI supplier** the data and energy questions above at your next renewal.

## Frequently asked questions

### Is it wrong for an environmental charity to use AI?

Not in itself. The question is whether a use helps the mission enough to justify its cost. Light, well-chosen use of text AI has a small footprint; heavy image or video generation and always-on automation deserve more thought.

### Can we use free AI tools safely?

Often, for tasks without personal or sensitive data, such as drafting a generic newsletter. Check the terms: some free tools may use what you type to train models. Don't enter beneficiary, donor or staff personal data.

### Do we need an AI policy?

Yes, even a short one. It tells staff and volunteers which tools are allowed, what data must never go in, and when a human must check the output.

### How do we explain our AI use to supporters?

Say plainly what you use AI for, how people's data is protected, and that staff check what it produces. Avoid claiming your AI use is green or carbon neutral unless you can show figures.

## Sources

- [Trust issues rise as charities' AI adoption continues (Charity Digital Skills Report 2026)](https://www.civilsociety.co.uk/news/trust-issues-rise-as-charities-ai-adoption-continues.html), Civil Society (2026)
- [AI adoption among charities shows no sign of slowing down (interim findings, Charity Digital Skills Report 2026)](https://www.civilsociety.co.uk/voices/zoe-amar-ai-adoption-among-charities-shows-no-sign-of-slowing-down.html), Civil Society (2026)
- [Smarter, smaller, stronger: resource-efficient generative AI and the future of digital transformation](https://www.unesco.org/en/articles/ai-large-language-models-new-report-shows-small-changes-can-reduce-energy-use-90), UNESCO and University College London (2025)
- [Power Hungry Processing: Watts Driving the Cost of AI Deployment?](https://arxiv.org/abs/2311.16863), Luccioni, Jernite and Strubell, ACM FAccT (2024)
- [Measuring the environmental impact of delivering AI at Google scale](https://arxiv.org/abs/2508.15734), Elsworth et al., Google (2025)
- [The gentle singularity](https://blog.samaltman.com/the-gentle-singularity), Sam Altman (OpenAI) (2025)
- [Energy and AI](https://www.iea.org/reports/energy-and-ai), International Energy Agency (2025)
